Yuvo Health’s founding team is composed of 100% people who identify as BIPOC. I mention that because each of us has experienced, first-hand, the lack of equity in our healthcare system. And it’s driven all of us to push for better. 

My family moved from the Philippines when I was three. We moved around quite a bit—wherever my parents could find work—and I saw how so many doors were closed to them. Most formative for me was that we lacked health insurance for most of my childhood and Community Health Centers (CHCs) were my primary care providers. That is why I care so deeply about them. Without CHCs, I don’t know what options I would have had for care, except to go to the emergency room. This experience led me to work in healthcare, and more specifically, to support the amazing, yet historically disenfranchised communities that I’ve been a part of. 

After working for several companies, honing my experience in value-based care (VBC), I realized two things: value-based care can and will revolutionize how care is delivered in America, and it should support everyone, not just large health systems and health plans. That’s when I started talking with my co-founders about Yuvo Health. 

We recognized community health centers as the drivers of better health in under-resourced communities—communities many of us came from. CHCs deserve credit under value-based care, but they were not getting it. So instead of continuing to be frustrated by the lack of support for CHCs, we decided to do something about it. Thus, Yuvo Health was born. And now, we have supported almost a dozen Federally Qualified Health Centers (FQHCs), across two states who serve around 200,000 lives. Through this work, we’re able to give these CHCs access to VBC revenue and resources that they will use to increase capacity and provide care to more people in their communities.

2. For those not familiar with Community Health Centers (CHCs), describe how those fit in the health care landscape and their importance.

Community health centers are such a critical piece of the healthcare landscape. They are non-profit organizations that provide healthcare services to underserved communities and they have a huge footprint. There are 1,400 CHCs in the United States that together serve more than 31 million people, who are mostly on Medicaid or are underinsured or uninsured. They serve 11% of children in the U.S. and 33% of all people living in poverty. 

Many people aren’t aware of the breadth of services CHCs provide. In addition to primary care, CHCs are often the only point of access to dental care, vision services, chronic disease care and behavioral health services. 

Plus, it’s estimated that by providing more cost-effective care, CHCs save $24 billion in healthcare spending. So as the focus of healthcare transformation is on reducing spending while improving access to care and overall quality, CHCs present a huge opportunity to accomplish that.

3. Value-based care is a new frontier in healthcare. What are the challenges you see standing in the way of broader adoption?

The biggest challenge is that it isn’t easy. Data analytics, risk adjustment, coding, care coordination and patient planning are all crucial to success in value-based contracting and they all require dedicated resources and likely new technology. 

Another challenge is that making the switch to value-based care involves financial risk. Smaller practices, in particular, are hesitant to take it on because they often don’t have the human or capital resources required. But even if CHCs had the resources to do this, they are the only provider group that is regulatorily prohibited from taking on financial risk. 

And we can’t overlook the fact that change of any kind is hard. Primary care practices are stretched thin. They’re overworked. So, even with the potential for a more balanced workload and sustainable revenue in the future, shifting operations now can feel like a heavy lift and one they don’t have the bandwidth for.

4. How does Yuvo Health best help its partners in unlocking the potential of value-based care?

Some systemic and regulatory barriers prevent community health centers from entering into the most beneficial contracts—specifically, contracts that involve risk. Yuvo Health unlocks those contracting opportunities by assuming that risk for CHCs. This allows CHCs to participate in contracts where the potential for return is so much greater than it is in contracts that they could pursue on their own. 

We also increase the bargaining power of any individual health center, resulting in more favorable terms for them. Because we focus exclusively on CHCs, we understand the unique challenges they face and can advocate for fair and equitable contracts. 

Then, we work with our partners to help them be successful in these contracts. Our team of population health and quality improvement experts become advisors and go-to resources on everything from helping to increase the number of people who come in for their annual wellness visit to providing training on accurate coding and gap closure documentation.

5. In what ways do you see AI potentially impacting care delivery the most?

There are a few ways that stand out: 

The first way is reducing the administrative burden. AI copilots integrated into Electronic Health Records (EHRs) can streamline data management, improve efficiency, and reduce errors. This allows the practice to keep its focus on providing excellent care and reduces time spent on things like referrals, claims management, and even risk stratification and patient outreach. 

AI also has the potential to lead to more personalized care with more accurate diagnostic capabilities by analyzing large data sets instantaneously. This puts so much more information into the hands of the provider when considering treatments and prevention for each patient.

6. If you could change one thing about the US healthcare system, what would it be?

Ensure equitable access to healthcare regardless of where you live, your race, your insurance status, etc. 

To get there, we have to start by ensuring equitable funding for all healthcare providers. That’s why we’re so passionate about empowering community health centers. And, we have to acknowledge the institutional racism and classism that exists in our society, not just in healthcare. We can’t ignore how something like food insecurity or a lack of reliable transportation impacts a person’s healthcare. 

There also has to be a realignment of incentives, to prioritize keeping people healthy rather than treating them when they are sick.

7. Professionally, what are you the most proud of?

We achieved shared savings for multiple community health centers in the first year of our value-based contracting program. It’s a huge success, not only in delivering revenue to the health centers, but also in showing other contractors that partnering with CHCs is a smart investment. 

What makes me particularly proud of this achievement is that it reflects the amazing work of so many: the talented team of diverse and caring individuals we have built at Yuvo Health and the health centers who placed their trust in us from the start. I’m proud of my role in cultivating and leading this group to create real, lasting change.

8. As a first-time founder, what’s something you know now that you wished you had known when you got started?

This may be a function of the unique dynamics of the venture capital market right now, but I wish I had known that I would never stop raising capital—that it is not a transactional process. I’m building long-term relationships with potential and current investors, instilling and maintaining their belief in our organization, and keeping them engaged throughout our growth journey. It is time-consuming, but it is so important and so worth it.

9. How have your investors like AV8 been most helpful to you?

Our investor partners at AV8 have been fantastic thought partners to us. Ruchita serves as a board observer but continues to punch well above her weight. She’s always available for honest counsel and has made numerous connections for us with subject matter experts, talent, funders, and customers/partners.

10. What have you recently read or watched (books, TV, movies, etc) that others should know about?

The Founder’s Mentality: How to Overcome the Predictable Crises of Growth by Chris Zook and James Allen is an oldie, but a goodie that I’ve now re-read as a founder. I highly recommend it. 

The main message is what they call the paradox of growth—when companies get bigger, they add complexity and bureaucracy. Yet this complexity and bureaucracy kills further growth. Though I’ve always appreciated this perspective, it didn’t hit me as hard until now, since I have founded Yuvo Health and led its growth.

DataGalaxy is a project that has been developing for quite some time.

The idea sprouted after I first met Lazhar in 2009 on an analytics project concerning the deployment of smart meters. We quickly noticed the difficulty of collaborating on data between business teams and IT teams, and thought that it was a problem that absolutely needed more attention.

We continued to closely monitor the market’s progression, witnessing the ascension of data agility and Big Data. With each development, our understanding of the data infrastructure landscape deepened, revealing both its challenges and potential. Despite the existence of data governance solutions, none seemed truly adequate to address the complexities we encountered daily.

So, we decided to take matters into our own hands. In 2015, DataGalaxy was designed to tackle these challenges in a completely new way.

2. Given the explosion of data in businesses both large and small, what are some of the biggest data challenges or roadblocks you see within organizations today?

Without a doubt, the lack of data literacy.

We’re at a stage where all companies have become aware of the importance of data in continuing to innovate and progress in terms of competitiveness. Unfortunately, very few of them know how to support their teams towards a global understanding and relevant use of data.

While technical teams can get by with extra work, business teams have no solution and are overwhelmed by the amount of information to assimilate and skills to acquire. It’s extremely important to adopt a distinct approach when engaging with business teams, who can no longer afford to opt out of their organization’s data culture.

That’s why the Chief Data Officer’s primary focus is to instill a data-centric culture across the company. The CDO plays a crucial role, especially in supporting the business teams through this substantial transformation. It’s no easy task.

3. You founded the business in France, and now you are expanding in North America. What are some of the things you’ve had to overcome to make this happen? 

Right before we decided to expand, we had begun our linguistic shift to an English-first company, both externally and internally. Although our teams were predominantly French-speaking, we gradually began hiring more international profiles to ease the transition.

This change was crucial in overcoming our initial hurdle: the lack of brand awareness. We’ve started increasing our digital presence in the US market, complemented by our physical participation in events and conferences several times a year.

Recruitment has also been a significant focus. Culture plays a vital role here, especially in our sales process. Thankfully, we have a unique product and value proposition that caters to both markets’ needs and expectations.

4. How do you see AI impacting the future of data analytics and intelligence, especially given all the recent focus on language and images? 

In the realm of data, AI, particularly generative AI, stands to significantly enhance productivity and quality across the board. Specifically within our domain, AI is already expediting data documentation and suggesting corrections to data definitions. It’s also fostering the development of new applications and interfaces that bridge users with the knowledge repository we’re constructing.

However, AI isn’t a panacea. That’s why we remain very focused on prioritizing the user experience at the core of our platform, and also why we focus so heavily on the human aspect of change management and data culture.

5. As a founder, what’s something you’ve learned over the years that you wish you had known when starting your entrepreneurial journey?

One key lesson I’ve learned as a founder is the sheer importance of fostering a positive and inclusive company culture from the get go. While I initially focused heavily on product development and market strategy, I underestimated the impact of cultivating a supportive and collaborative environment for our team.

I’ve come to realize that a strong company culture not only contributes to employee morale and productivity, but also attracts top talent and sets us up for success in the long run. Moving forward, I prioritize traits such as empathy, communication, and adaptability when hiring new team members, as these qualities contribute significantly to our cohesive culture and drive our collective mission forward.

6. What are some technology trends in the enterprise data and compute landscape, other than AI, that you are following which you are excited about?

On the computing side, I’m trying to keep up with the technological advances in quantum processors, while at the same time trying to understand how they work (in layman’s terms). This technology will shatter all previous records for computing speed. As a result, it will pose a number of challenges, not least for cryptography, but also to further expand the capabilities of AI.

From a business perspective, I’m deeply interested in knowledge graphs and exploring how to leverage their potential.

7. DataGalaxy is now over 8 years old. How has your role as CEO changed over that time, and how would you describe it today?

Regardless of our age, it’s above all the size of DataGalaxy and our international presence that have had the greatest impact on my role as a CEO. From being a Swiss Army knife with numerous contacts in all fields (employees, customers, business partners, suppliers, banks, accountants, etc.), my role has evolved to become more of a management and representation role.

Without being original, I would say that my main challenge today is to succeed in surrounding myself with and empowering people who are much better and more knowledgeable than I am in their field, and providing them with a work environment where they can perform at their best.

8. How have your investors like AV8 been most helpful to you?

The support of our investors, particularly AV8, is indispensable when we need to reassess or view things from a different perspective. For instance, in terms of positioning the company, Baris provides valuable insights into market dynamics, drawn from his extensive interactions with various startups. In our GoToMarket strategy, he assists us in taking a step back to benchmark ourselves against other companies in the market and evaluate our progress against industry standards.  

9. What’s your go-to interview question for new hires, and what attributes and experiences do you gravitate toward?

One of our key interview questions for new hires and existing team members is about their vision for the future of DataGalaxy. We understand that we are all pieces of one puzzle, and we really invest in gathering as many perspectives as possible from our teams. Diversity of thought is important to us.

In terms of qualities, our core values revolve around being intentional, bold, clear, and humble. At this stage of DataGalaxy’s evolution, I would say that boldness is key. We’re committed to taking risks and forging new paths, and we value employees who share that hunger for innovation.

10. What have you recently read or watched (books, TV, movies, etc) that others should know about?

I recently finished reading ‘Decisive’, a book that focuses on the intricacies of decision-making and offers insights on avoiding the pitfalls of binary choices. I highly recommend it to anyone interested in refining their decision-making skills!

Source: Canva — Magic Media Editor — Insurtech future in emerging markets

Understanding the Challenge
Emerging markets, characterized by rapid economic growth and increasing urbanization, often face challenges in providing adequate insurance coverage to their populations. Factors such as regulatory barriers, lack of awareness, financial stability, and outdated distribution channels contribute to a significant gap in coverage.

The Power of Mobile Technology
One of the key drivers behind enabling insurance in emerging markets is the widespread adoption of mobile technology. In many emerging countries a large portion of the population owns smartphones, and Insurtech startups can capitalize on this connectivity to reach previously underserved communities.

Customized Solutions for Diverse Needs
Insurance in emerging markets is not a one-size-fits-all approach. Successful startups recognize the importance of offering customized solutions that cater to the unique needs of each market. Whether it’s micro-insurance for low-income households, weather-based insurance for farmers, or health insurance with flexible terms, Insurtech platforms have to tailor their offerings to address specific challenges faced by local communities.

Financial Inclusion Through Collaboration
Collaboration has been a key strategy for Insurtech startups looking to penetrate emerging markets. By partnering with financial institutions, NGO’s, mobile network providers etc. Insurtech’s can integrate insurance solutions with existing services, making them more accessible to a wider audience. These collaborations not only drive financial inclusion but also contribute to the overall economic resilience of the communities they serve.

Addressing Cultural Nuances
Cultural nuances play a crucial role in the success of Insurtech ventures in emerging markets. Successful startups invest time and resources in understanding local cultures and adapting their products accordingly. This approach builds trust among users and ensures that insurance solutions align with the values and expectations of the communities they serve.

Source: Amir Kabir — Market, Opportunities and Challenges in Emerging Markets

Insurance penetration serves as a metric for assessing the progress of the insurance sector in a country or region — It is computed by determining the ratio of total insurance premiums to the GDP (gross domestic product) in a specific year. Insurance density, coupled with insurance penetration, serves as a crucial metric for evaluating a country’s or region’s insurance sector development. Calculated as the ratio of premiums collected by insurance companies to the population, it is often expressed as the per capita premium in dollar terms.

The insurance density figures on all three continents are notably lower than the global average of $874.

Source: Best’s Review — January 2024 Edition

In terms of Insurtech maturity, Latin America is leading, while Africa and the Middle East are still in their very early stages of innovation, adoption, and disruption. Moreover, insurance and Insurtech penetration vary significantly among countries within each continent. As I highlighted in Part 1 of this series, the evolution of Insurtech in the United States unfolded gradually, spanning a period of more than two decades. Nevertheless, both Africa and the Middle East may have an advantage, as Insurtech’s in these regions can commence operations without the constraints of complex legacy systems. However, these endeavors encounter various other challenges that hinder their rapid scalability.

Latin America — A thriving and rapidly expanding insurance market, presenting significant opportunities for both incumbents and new entrants

Latin America (LATAM) emerges as a dynamic hub for innovation and entrepreneurship, a compelling reason for everyone to take a keen interest in the region’s startup landscape. With a staggering population of ~660 million individuals, almost twice that of the United States, it stands as a colossal force on the global stage. What’s truly striking is the remarkable ascent of the middle class in the region, experiencing a surge of over 50% in the past two decades, according to the OECD. The Latin American populace not only boasts significant numbers but also embodies a tech-savvy and youthful demographic, with a median age of 31 — seven years younger than both China and the United States. These insights underscore the region’s potential as a vibrant and influential player in the global landscape.

The weight of insurance on the LATAM economy accounts for ~3% of GDP, less than half the global average. In general, LATAM is an underdeveloped and underinsured market. In terms of life insurance, only 15% of Latin American adults who can afford such policies hold them, compared to 52% of U.S. adults. Similarly, less than 25% of vehicle fleets are insured in Latin America, whereas in the United States, this figure stands at 70%. What Latin America does have, however, is an over 80 percent digital adoption rate, which enables startups to expand the market.

Latin America has emerged as the region with the highest financial appeal for insurance companies, surpassing the global average that includes North America and Europe.

As per the 2023 Global Insurance Report from McKinsey & Company, the insurance market in LATAM is valued at $174 billion. Despite disparities, insurance penetration in LATAM is on an upward trajectory at a faster pace compared to other global regions.

Source: McKinsey — The Latin American Insurance Market

Over the past decade, the region has consistently outpaced other major insurance markets, boasting an impressive ~11% CAGR from 2018 to 2022 — outstripping North America, Asia, and Europe. Latin America’s insurance sector not only exhibits rapid growth but also proves highly lucrative, with a robust average Return on Equity (ROE) of 16.6% in 2022, surpassing global averages. Despite these promising figures, a substantial protection gap persists compared to more mature markets, with per capita insurance density at a modest $283 in Latin America versus $9,250 in the US. Local insurers dominate the market, commanding 64% of total gross written premiums, while global players carve a stronger presence in non-life insurance.

Efficiency emerges as a critical challenge for Latin American insurers, with higher expense ratios than their European counterparts, presenting significant opportunities for operational improvements.

This points to massive opportunities for Insurtech’s to support and enable the current market players, while expanding product portfolios, digitalizing customer interactions, creating new business models, and contributing to closing inequality gaps.

Key Trend — Progression of Open Insurance

A prominent trend in the region revolves around the emergence of open insurance, a novel business approach that advocates for the sharing of open data among organizations spanning various industries. This entails providing open access to information, including terms and prices, with the overarching goal of enabling providers to deliver products and services within digital ecosystems boasting lower operating costs. Simultaneously, this approach aims to enhance efficiency by reaching a broader audience while maintaining effective customer segmentation.

The convergence of Open Insurance and Open Banking gives rise to Open Finance, forming an ecosystem where an individual’s insurance and banking details are unified. This integration facilitates the development of more tailored products and a more assertive and user-friendly transactional experience.

Brazil, in particular, is at the forefront of adopting this model in Latin America. The insurance regulator, Susep (Brazilian Superintendence of Private Insurance), is already in Phase II of implementing a regulatory framework for open insurance. Moreover, since 2017, Susep has been running a ‘regulatory sandbox’ initiative. This program grants licenses to entities, allowing them to experiment with innovative projects for a specified period. During this time, they must adhere to a specific set of regulatory provisions that facilitate the controlled and delimited execution of their activities.

The Latin American Insurtech ecosystem presents a compelling mix of elements that attract global interest and foster growth. While Brazil leads in terms of investment volume, the number of ventures, user potential, and government interest, other countries like Mexico, Colombia and Chile boast similarly mature ecosystems. Brazil accounts for almost 50% of Insurtech investments in the region, followed by Mexico and Colombia. Currently they are around 500 startups focused on building in and around the insurance and Insurtech sector, with a cumulative funding of more than $1 billion (estimated based on various sources).

Life Insurance — A fundamental component for the rising middle class, playing a pivotal role in bridging portions of the existing protection gap

Life insurance serves as a crucial financial tool that provides individuals and their loved ones with a sense of security and stability. Beyond its primary function of offering a financial safety net in the event of the policyholder’s demise, life insurance plays a pivotal role in long-term financial planning. Additionally, life insurance can serve as an investment vehicle, offering opportunities for wealth accumulation and legacy planning. In this way, life insurance goes beyond a mere financial product, becoming a cornerstone for building and preserving financial well-being.

According to MAPFRE’s estimate, in Latin America, the insurance protection gap reached $267.2 billion, with 60% coming from the Life segment.

During the mid-1990s, numerous countries in the region embarked on reforms to their Social Security Laws, resulting in the transfer of life, disability, health, and retirement risks to the open private market. Within the LATAM region, the life insurance market presents considerable opportunities for new business development across three primary segments:

1. International Life Products — OffShore: This segment, currently valued at over $2 billion for life business premiums, is experiencing continuous growth propelled by the expanding High Net Worth and upper-middle-class demographics in LATAM. Health business premiums are estimated to surpass $600 million.
2. Local Existing Life Market — OnShore: Total premiums for life in the region amount to approximately $80 billion, with about $40 billion attributed to the six target markets (Mexico, Brazil, Colombia, Chile, Argentina, Peru). The historical profitability ratio averages at 19%.
3. Insurance Protection Gap: In 2022, the total premiums for the Insurance Protection Gap in LATAM were estimated at around $270 billion across all lines of business. The Life Insurance Gap alone is approximated at $162 billion, representing potential customers who currently lack life insurance but possess the financial means to afford it. These figures underscore the significant opportunities for market penetration and expansion in the LATAM insurance landscape.

Market Tailwinds

Throughout history, insurance penetration in LATAM has maintained a low trajectory, typically hovering at approximately 0.5% to 1.5% GDP for Life insurance. This trend has predominantly been linked to lower incomes, contributing to a limited awareness of the protective advantages that insurance provides. However, a positive transformation is currently unfolding, driven by the growth of the middle class, enhanced accessibility to competitive insurance products, and a generational shift. These factors collectively have spurred an upswing in insurance penetration within the region.

The primary catalysts for the growth of the insurance sector encompass increasing disposable income, enhanced availability of insurance education, a growing demand for financial stability and inclusion, and technological advancements that streamline the insurance purchasing process.

Prominent global reinsurers foresee a 2x increase in insurance penetration across specific countries in LATAM within the next five years. The advantageous impact of the demographic, with 66% of the population falling within the 15–64 age bracket and less than 10% aged over 65, continues to offer significant prospects in the retail consumption business. This encompasses diverse personal lines within the insurance sector, such as individual life and individual health. The emerging middle class in LATAM has witnessed an impressive growth, now representing approximately 30% of total households, underscoring substantial opportunities for insurers to cater to evolving consumer needs and preferences in the LATAM market.

Brokers and agents dominate distribution and remain highly relevant by adapting to client needs.

As indicated by a 2022 survey conducted by McKinsey and LIMRA, although Latin American consumers show a preference for digital interaction, with 67 percent initiating their life insurance search online, the majority of policies are not ultimately purchased through digital channels. When making purchasing decisions, consumers prioritize the security, trust, and convenience associated with in-person interactions.

At AV8, one of the companies in my portfolio is Olé Life, a full-stack digital insurance carrier operating in Latin America. The entrepreneurial team at Ole brings decades of insurance experience, having taken risks in Latin America and acquiring a comprehensive understanding of diverse risk environments. Leveraging their knowledge and expertise, the team has pioneered a novel AI-backed digital underwriting approach for delivering life insurance through online platforms and mobile applications. Olé has excelled in the recruitment, training, and retention of top life insurance agents in LATAM, equipping them with advanced technological tools to enhance their selling capabilities. Headquartered in Miami, Florida, Olé Life stands out as the first fully digital term life insurance product in Latin America that requires no medical exams for policies up to $1 million.

The core of Olé Life’s technology revolves around incorporating data science into its mobile and web applications, providing users with relevant information for a personalized and streamlined purchasing experience. This technology is accessible through embedded quoting engines and APIs, allowing B2B partners and agents to fully utilize and integrate the innovative platform.

As a fact Olé Life focuses on empowering brokers and agents by providing a streamlined, technology-driven approach, enabling them to sell life insurance more efficiently, effortlessly, and at an accelerated pace.

The cutting-edge AI-powered system utilizes data accumulated from decades of policy underwriting in Latin America to approve customers within minutes, offering same-day coverage. Olé Life currently collaborates with A+ rated international reinsurers, including Partner RE, RGA, Munich RE, and has a presence in 30 countries.

Source: Olé Life Management Presentation

Africa — The most underserved insurance market in the world offering vast growth potential coupled with inherent complexity

Dr. Osagyefo Kwame Nkrumah, Ghana’s inaugural prime minister and president, encapsulated a poignant paradox in his impactful declaration:

“It is Africans who are poor but not Africa.”

This statement resonates profoundly, drawing attention to the stark contrast between the continent’s abundant natural wealth and the widespread poverty experienced by its people. Africa’s vast natural resources are undeniable and could contribute significantly to the well-being of its population.

The African insurance market reflects this potential, reaching a substantial US$ 70.2 billion in 2022 according to Swiss Re Sigma Report, with projections indicating growth to US$ 123.8 billion by 2028, at a CAGR of ~8% during 2023–2028. However, despite its size, insurance penetration across Africa remains at a modest 2.8%. The industry’s value of ~$80 billion in Gross Written Premium (GWP) ranks very low globally, with 83% of premiums concentrated in seven key countries (South Africa, Kenya, Egypt, Nigeria, Algeria, Namibia, and Tunisia).

Source: Swiss RE Institute — African Insurance Market

South Africa dominates as the largest market, accounting for ~65% of premiums, ~80% of which are for life insurance.

Source: The Geneva Papers on Risk and Insurance — Issues and Practice — Institutional determinants of insurance penetration in Africa

The significant opportunity lies in the fact that 97% of the continent remains uninsured, offering a vast untapped market. Technology has the potential to bridge this gap, providing affordable insurance as a formal and efficient financial safety net. However, challenges persist, including low demand due to lack of awareness and trust in traditional insurance companies, as well as regulatory inconsistencies among countries, highlighting the need for innovative solutions in the African insurance landscape. Moreover, distribution continues to pose a significant challenge for the industry, with obstacles stemming from various factors, including cultural habits. Beyond addressing distribution issues, another crucial factor is the generation and access to vital data for underwriting and other purposes.

According to World Economic Forum, the insurance sector is leading Africa’s sustainability push.

African nations have embraced a strategy similar to that of Latin American countries through the initiation of ‘regulatory sandbox’ initiatives. However, the outcomes of these initiatives, designed to foster innovation, have generally fallen short of expectations as of today.

In the realms of healthcare and insurance, insiders in the industry suggest that many individuals still prefer to rely on God as the ultimate — or only — physician, or allocate their financial resources to address other needs first. The tendency among Africans is to neglect their health until it becomes an emergency, at which point they start seeking pharmaceuticals and consulting doctors, hindering the necessary innovation.

As per Brighter Bridges, there are more than 70 startups focused on building in and around the insurance and Insurtech sector. Per PitchBook there are more than 100 startups headquartered in Africa, operating across different verticals, from API development to car and crop insurance, with a cumulative funding of approximately $250 million (estimated based on various sources).

Microinsurance — An opportunity within the thriving African ecosystem

Microinsurance has a concise yet influential history, evolving as a mechanism to offer financial protection to low-income individuals traditionally excluded from conventional insurance services. Rooted in the microfinance movement, the concept of microinsurance gained substantial traction in the late 1990s and early 2000s. This momentum can be attributed largely to the groundbreaking efforts of Muhammad Yunus, the Nobel Prize-winning banker and economist from Bangladesh. The microinsurance landscape, therefore, stands as a testament to the extension and impact of the broader microfinance movement.

The International Association of Insurance Supervisors (IAIS) defines microinsurance as “protection of low income people against specific perils in exchange for regular premium payments proportionate to the likelihood and cost of the risk involved.”

Microinsurance plays a crucial role in safeguarding low-income populations through the provision of affordable, basic coverage featuring low premiums and policy sums insured. This form of insurance addresses a vital need in emerging markets where government social safety nets are limited, leaving individuals susceptible to income shocks. However, the global market remains largely untapped, with only approximately 230 million people covered, according to recent estimates. Offerings encompass life, accident, health, property, and agriculture lines specifically tailored to the risks and cash flows of the low-income market. Small farmers for example produce up to 80 percent of the food supply in Africa and Asia and face a 90 percent chance of crop loss due to factors outside of their control, such as pest outbreaks and severe drought.

Source: Amir Kabir — Challenges in Insuring Low-Income Customers

Establishing partnerships with governments, aid agencies, microfinance institutions, and cooperatives is pivotal for microinsurers to reach underserved groups while keeping costs low. Technological innovations, including mobile platforms, user interfaces, and payment systems, contribute to premium growth and enhance financial inclusion. However, profitability remains a challenge due to high distribution costs and low persistency rates. Achieving sustainability hinges on effective claims management and achieving scale. Regulatory development, consumer education, and improvements in product distribution and servicing are essential factors that can significantly expand microinsurance markets.

Source: Micro Insurance Network — The Landscape of Microinsurance 2022

Startups building in that space for example are Pula Advisors, an agricultural insurance and tech company that designs and delivers innovative agricultural insurance and digital products to help smallholder farmers endure yield risks, improve their farming practices, and bolster their incomes over time. Pula Advisors, which works across Africa and Asia, has insured more than 15 million farmers so far, and surpassed US$80 million in gross premiums.

Others like Jamii.one operate a digital finance platform intended to transform data into financial opportunities. Jamii’s platform also provides affordable and accessible micro-insurance products to low-income individuals and communities in Ethiopia. The company’s products include crop, livestock, life and health insurance.

OKO brings affordable crop insurance to underserved smallholder farmers, giving them access to the finance that keeps them sustainable and allows them to grow. As of July 2022, OKO has insured more than 18,500 farmers, making OKO the most popular crop insurance product in the country.

Turaco is pioneering microinsurance solutions to make insurance accessible for those living on less than $5 a day. The company offers affordable life and accident coverage, priced at an average of $2 per policy, specifically designed for urban and rural populations. In its first year, Turaco insured more than 40,000 customers, providing peace of mind and coverage in times of greatest need. Since 2019, Turaco has delivered customised policies covering over 1.3 million Africans.

Middle East & North Africa (MENA) — The region with the lowest insurance penetration

The insurance penetration in the Middle East and North Africa (MENA) region is the lowest globally, with less than 1% market penetration for insurance companies. This can primarily be attributed to fragmented market structures, lack of awareness, cultural and religious beliefs in the region that play a significant role in the low insurance penetration rates. Furthermore, impediments to market growth include competitive distribution strategies and a scarcity of talent possessing extensive insurance experience. The primary lessons derived from the initial wave of Insurtech’s in the US, as highlighted in Part II, emphasize that technology alone is not a winning factor; a solid understanding of insurance fundamentals is paramount. Notably, the aggregation model has emerged as the driving force behind Insurtech growth, earning recognition from regulators throughout the region. This underscores the observation that the MENA region is still in the early stages of the insurance innovation cycle, mirroring the efforts discussed in Part I regarding the US market.

The markets of Saudi Arabia, the United Arab Emirates, Kuwait, and Qatar represent the biggest insurance markets in the Gulf Cooperation Council (GCC) region. Non-life insurances are the more significant segment of the insurance industry in this region. Most insurance policies have been adjusted to Islamic law requirements, in the same way as other financial products, to avoid interest-related income.

Source: Swiss RE Institute — Emerging Middle East Insurance Market

Determining the comprehensive market size of insurance in the Middle East poses a challenge. However, according to Swiss Re’s 2022 report, the insurance market for the Emerging Middle East region (United Arab Emirates, Saudi Arabia, Iran, Iraq, Bahrain, Jordan, Kuwait, Lebanon, Oman, Pakistan, Qatar, Yemen, Syria) amounted to approximately $43 billion. This amount is divided into 83% for non-life insurance and 17% for life insurance. Considering Morocco (~$3B), Egypt ($3.5B), Algeria ($1.5B), Israel ($11.5B) and Tunisia ($1B), the collective market size for insurance in the MENA region approximates $64 billion.

According to Alpen Capital, the GCC insurance market (Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and United Arab Emirates) is projected to grow at a CAGR of 5.3% from $34.3 billion in 2023 to $44.4 billion in 2028. The non-life insurance segment is anticipated to grow at a CAGR of 5.4% between 2023 and 2028 reaching US$ 39.6 billion, comprising 89.2% of the region’s GWP by 2028. However, the GCC insurance market is fragmented and highly competitive. Price competition to secure business in the fiercely competitive motor and medical insurance segments threaten the profit margin of insurers.

Furthermore, the complexity level varies across different insurance segments and regions; for instance, health insurance is highly regulated, while motor insurance has become a prominent focus in MENA due to its growing significance. Health and motor insurance together constitute the majority of written premiums collected by insurance companies in the region.

Source: Best’s Review— February 2024 Edition

In the UAE, which has over 550,000 SMEs, and Abu Dhabi, health insurance is mandatory for its residents and all employers must provide health insurance to employees and their dependents. However, Abu Dhabi and Dubai are the only two emirates of the UAE that have mandatory health insurance laws. Another interesting fact is that motor insurance in Qatar insures the vehicle rather than the driver. Therefore, once a vehicle is insured, anyone with a driving license can drive it.

According to CBInsights there are around 70 companies transforming Insurtech in the Middle East and North Africa. Per PitchBook there are 60 startups headquartered in MENA (excluding Israel), operating across different verticals, with a cumulative funding of approximately $200 million (estimated based on various sources and excluding Israel). Startups such as Dubai-based Yallacompare let consumers compare and buy insurance policies — including auto, life, health, pet, and travel — in 9 markets in the region. Others like Democrance, have a mission to enable partnerships that make insurance accessible and affordable for those who need it most.

The ascent of Takaful as a solution influenced by cultural factors

Access to insurance is further restricted in most countries with sizeable Muslim population as people remain reluctant to buy insurance due to socio-cultural and religious reasons. In general, the conventional insurance mechanism involves elements that are prohibited by Islamic rules and principles.

Takaful is an Arabic word that means assuring each other.

Takaful, as an Islamic insurance concept, resonates with the Islamic heritage of the Middle East, making it a natural choice for individuals seeking insurance without compromising their faith. This cultural and religious alignment has contributed to the remarkable growth of Takaful in the region in recent years. The global Takaful insurance market was valued at $31.7 billion in 2022, and is projected to reach $126.8 billion by 2032, growing at a CAGR of 15.2% from 2023 to 2032. Saudi Arabia remains the largest Takaful market in the GCC with overall GWP of US$ 14.2 billion in 2022.

Takaful and conventional insurance exhibit certain differences. Takaful operates on principles of mutual collaboration, solidarity, and indemnification, where participants protect each other from potential losses or catastrophes by sharing the financial burden. In contrast, conventional insurance involves the transfer of risk from the insured individual to the insurer firm.

Source: The Actuary — Takaful’ models of Islamic insurance

Takaful companies invest in Sharia-compliant instruments, promoting ethical financial practices and alignment with Islamic teachings. Additionally, Takaful’s customization process takes into account the cultural values and beliefs of the target communities, ensuring that the products are not only financially effective but also culturally sensitive. Furthermore, Takaful’s emphasis on community solidarity is well-suited to addressing complex, cross-border challenges, reflecting its cultural adaptability and resilience.

Among the 20 countries that contribute to 90% of the world’s insurance premiums, none are recognized for having significant Takaful market share, and the majority lack notable availability of Takaful products.

It is crucial to emphasize that Takaful is not limited exclusively to Muslims.

It is fascinating to see that such an old model of peer-to-peer insurance has strong relevance today within a society with the lowest insurance penetration.

How Incumbents are leading change with Insurtech

Within the GCC countries, the United Arab Emirates (UAE) and Saudi Arabia stand out as primary hubs for Insurtech innovation. Their robust economies, expanding middle class, and favorable regulatory landscapes make them appealing investment destinations for Insurtech startups.

Saudi Arabia has probably the most developed frameworks as part of its efforts to push insurance forward. The Insurance Authority in Saudi Arabia aims to regulate, supervise and control the insurance sector and provides specific guidelines and rules — here are the guidelines and rules for licensing and supervising Insurtech companies operating in the insurance sector. An interesting highlight is that Insurtechs must have a physical presence in Saudi Arabia and employ Saudi nationals.

In the UAE, Dubai, in particular, has positioned itself as an innovation hub, drawing in a diverse array of startups. It encourages collaborations among Insurtech firms, regulators, and traditional insurance providers. The Dubai International Financial Centre (DIFC) has initiated several programs, such as an innovation hub and accelerator, to nurture Insurtech startups and aid in their development.

Established players have been actively engaging in partnerships with Insurtech startups to cater to their clients. A notable instance is Sukoon Insurance, a prominent multi-award-winning provider of health, travel, and car insurance in the UAE. Founded in 1975 and majority-owned by Mashreq Bank, Sukoon stands as one of the largest publicly listed insurers in the country. Renowned for innovation, it serves over 800,000 customers and offers various digital solutions, including mobile apps, online quoting, and portals for existing clients, as well as its network of agents and brokers. Recently the company entered into a strategic partnership with WAX, one of my portfolio companies here at AV8, to launch the first collector App in the UAE.

Another example is XA’s Group roll out of a Blockchain-based motor insurance platform. The platform facilitates insurance companies in reconciling motor recovery receivables and streamlining the claims recovery process.

The GCC’s emphasis on digitalization opens doors for Insurtech startups to revolutionize the insurance sector through technology, data, and customer-centric strategies. Insurtech facilitates the creation of tailored products by insurers to meet individual needs, contributing to a heightened demand for insurance.

In Part IV of Insurtech and the ongoing expansion of the E&S insurance market, I will cover the opportunities within the E&S sector and how Insurtech’s are playing a pivotal role. Whether you have valuable Insurtech insights or wish to be part of the conversation, feel free to reach out at amir@av8.vc or via Twitter at @AmirKabir99.

Amir is a General Partner at AV8 Ventures where he leads Fintech and Insurtech investing and has been one of the earliest investors in the Insurtech space. https://av8.vc/amir-kabir/

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future. (I have copied the disclaimer used by Jamin Ball in his great newsletter Clouded Judgment)

We formed Axuall in 2019 to address modern history’s biggest supply chain problem: the clinical workforce supply chain. Coming off the heels of my last big data company in healthcare, Explorys (sold to IBM in 2015), it became clear that there was a serious information gap in healthcare related to clinician data. Timely, accurate, and comprehensive data about providers is critical to making informed decisions about planning, recruiting, and quickly onboarding resources, particularly in an era of increased demand and workforce shortages. Clinicians represent the highest cost and primary source of revenue to healthcare organizations, all struggling with rising costs, high attrition, and burnout.  

Axuall is a workforce intelligence company developed with leading healthcare organizations built on a national real-time Clinician Data Network. We connect health systems, staffing firms, and digital health groups to a vast array of data, providing insights for network planning, analytics, and reporting while dramatically reducing onboarding and enrollment time via provider-enabled digital credentials.  

2.  While you’ve spent time in larger companies, you have co-founded and grown several start-ups.  Was there something in your past, or growing up, that planted the entrepreneurship seed in you?

I grew up in an immigrant family with entrepreneurial parents, grandparents, and siblings, so I guess it’s in my blood. There is a thrill in creating something new that solves problems for others without the constraints of legacy inhibitors. I started my first computer company at 15, and I’ve been blessed to work with customers and partners who share our visions over the years. That said, my time in big companies was also instrumental in my professional development. Seeing how large “organisms” make decisions, manage execution risk, and navigate cultural challenges taught me much about the perspective and thinking of the enterprise customers we sell into and serve today.   

3. You’ve worked in both banking and healthcare, two of the most regulated (and important) industries there are. How would you compare the challenges of each? 

When I entered healthcare with Explorys in 2009, we had no choice but to make comparisons to the financial space. As crazy as it might sound, my co-founder, Steve McHale, and I had no experience in healthcare. We were fortunate to meet some innovative people at the Cleveland Clinic who knew they had to apply out-of-the-box thinking to address the actuarial challenges of value-based financial models, and given my experience in leveraging big data to manage consumer and organizational risk in banking, it was a way bridge that gap between us and industry.   

The financial industry constantly measures and manages risk, often driven by seismic economic shifts and consumer behaviors. Healthcare must also deal with economic, pandemic, and consumer/patient behavior risks, not to mention a workforce that sees the world much differently than five years ago. Objective data should be at the heart of how healthcare addresses these challenges, lest they be doomed to repeat past mistakes.      

4. If there’s one thing you could change about the US healthcare system, what would it be? 

The billions of dollars associated with administrative waste and resource mismanagement are significant problems today. It robs everyone of the care and careers they deserve. We want to help change this.  

5. What are some of the more exciting trends you have seen in healthcare workforce development recently?

We’re very bullish on workforce intelligence. I was lucky to be part of a wave of patient intelligence and big data a decade back that gave rise to new ways to optimize outcomes and reduce costs. The parallels today with the workforce are just as exciting. The industry cannot address the workforce supply and demand gap only by hiring more doctors and nurses. There must be a data-driven approach to understanding care delivery and utilization trends to develop more efficient care team models and orchestration. Big data, analytics, and AI will play a prominent role in this, and we’re excited to be part of it.   

6. As a serial entrepreneur and founder of multiple companies, what advice do you have for new founders; what do you know now, that you wish someone had told you when you first started?

I suggest drawing a line in your mind from where you are today to where you want to take your team, customers, and investors in the future. This should go beyond the standard formulaic multi-year proformas that entrepreneurs start with. It should encapsulate your goals for how your team develops their career, how your customers will find you indispensable, and how you’ll return the capital your investors deserve for making their commitments to your company’s growth. With that timeline in place, catalog the deliverables and milestones to get there and the risks you’ll have to manage carefully along the way. Regarding risks, be honest and maniacal about how you rate and track them, along with the degree of probability, control, and impact they can have on your business. Great businesses are not built on models where everything must go right to make it work. Master this, and you’ll have happy employees, customers, and investors!   

7. What’s your go-to interview question for new hires?

What was the biggest mistake you made in your career, what led up to it, what impact did it have, and what did you learn from it? The answers to these questions tell a lot about someone’s thought process, self-awareness, and ability to adapt — all critical traits for being in a startup/scale-up.  

8. What are some of the unique challenges and opportunities launching and scaling your business in Cleveland?

Five years ago, I would have had a different answer, but to be honest, nowadays, there aren’t any – at least no more than any other city. Since COVID, much of the tech workforce has gone remote, so drawing from a limited metropolitan area is less of a factor. We’ve found ways to foster culture via creative video conferencing and regular in-person events. Having done several startups in Cleveland, I had a strong network of seed investors eager to fund and support us, some of whom were pilot customers. All to say, your local community can be a significant asset as you grow your business – even in the Midwest.  

9. How have your investors like AV8 been most helpful to you?

Our investors have been helpful in three key areas: metrics, contacts, and strategy. Key performance metrics such as ARR, burn multiples, EBITDA roadmaps, and expense ratios are critical to short and long-term tactics and strategy. Regularly asking yourself where the company is today and where it will be a year from now, given market conditions and run rates, allows leaders to adjust and adapt early and maximize capital.  

Introductions to industry contacts are another benefit that AV8 and others have provided. Whether these referrals are to customers, partners, or other portfolio companies, these relationships and insights offer essential guidance and sometimes translate to large revenue deals.  

Lastly, having industry-savvy investors to test new ideas with, particularly as it relates to revenue opportunity and competitive positioning, is something they are best suited to help with.    

10. What have you recently read or watched (books, TV, movies, etc) that others should know about?

While on vacation, I just read The Road Less Stupid by Keith Cunningham. It’s a book for entrepreneurs about managing risk and opportunities by applying facts over emotion and unfounded optimism. These lessons are particularly important for startups during the good and bad times they experience.  

Source: Canva — Magic Media Editor — The Future of Insurtech

While there has certainly been disruption in insurance in the past decade, there are still many hard problems to solve in the industry, and incumbents have proven to be more resilient than many initially thought

Source: Amir Kabir — The Evolution of Insurance & Insurtech (Updated)

The rise of Insurtech can partially be attributed to traditional insurance companies’ lack of focus on customer experience. Historically, these companies have classified their clients as “policyholders” instead of “customers”, demonstrating a disregard for their needs and satisfaction. The first wave of Insurtech startups aimed to improve the customer experience by focusing on user-friendly interfaces, advanced data sources, and digital solutions. While these improvements were significant and a great start, there is still a lot of room for growth, particularly in the specialty insurance sector.

Lessons derived from the first wave of Insurtech’s

It’s important to note that the first wave of Insurtech did not completely revolutionize the insurance industry, but rather made incremental advancements towards better serving customers.

However, Insurtech v1.0 companies successfully navigated the intricate task of convincing incumbents, service providers, consumers, regulators, and the broader financial markets that digitization of insurance is inevitable.

Their mission was to underscore the imperative inclusion of advanced technologies in the future of insurance while outlining a viable path for venture backable companies. Despite this, the performance of few Insurtech v1.0 ventures, assessed through exit metrics, might be perceived as underwhelming by the insurance or venture community.

Source: Amir Kabir — Insurtech v1.0 <> Insurtech v2.0 (Updated)

Source: McKinsey Global Insurance Pools

Source: The Geneva Association — The root causes of insurance protection gaps, main areas of relevance

There is also a huge protection gap between the amount of insurance that could have been purchased and the insurance that is actually bought, especially in the risk pools of natural catastrophe, mortality/life, and health care. The value of unprotected risk exposure has risen steadily in the past five years. According to the Swiss Re Institute’s research, the global protection gap is at USD 1.8 trillion in premium equivalent terms for 2022, a cumulative 20% increase on the comparable-terms USD 1.5 trillion estimate for 2018.

While embedded insurance will not fully mitigate the gap, it can act as a catalyst

Source: Swiss Re Institute — Economic Resilience

Speciality Insurance — The next frontier in digitizing business lines, embracing successful strategies from other sectors of insurance

Source: Amir Kabir — Some of AV8’s investment in the speciality insurance category

Here are some other speciality insurance categories that have exiting potential in the future:

Medical Malpractice — An ancient market that needs technological help

Medical malpractice can be dated back to the Mesopotamia time, where the concept of holding medical professionals accountable for deaths or injuries that could have reasonably been prevented. In todays world medical professionals including physicians, clinics, nurses etc. have to carry some sort for medical malpractice insurance that covers them for injury related to medical negligence, resulting in a patient needs to prove that substandard medical care resulted in an injury. Similar to other lines of insurance in the U.S. medical malpractice law has traditionally been under the authority of the individual states and not the federal government, in contrast to many other countries.

Evaluating and underwriting medical malpractice insurance is a highly manual process, requiring considerable time. Hospitals and doctors must complete a PDF application, submit it to an agent along with their resume, credentials, financial projections, etc. An underwriter then manually inputs most of this information into a rating table. The underwriting process commonly considers cost determinants and data such as the frequency of claims, the severity of claims, required coverage amounts, local regulations, and the geographical practice location. Typically, physicians wait around one week for a quote, home healthcare systems take 3–4 days, and hospitals usually face a 2–3 week turnaround time.

Medical malpractice insurance premiums are 1% of national health care costs (Congressional Budget Office)
Medical malpractice insurance premiums are the 3rd largest cost for physicians, after payroll and space costs

Innovative players in this field include Indigo and Now Insurance. Even major established companies like Berkshire Hathaway are actively involved in this sector. For instance, Berxi, a segment of Berkshire Hathaway Specialty Insurance, offers medical malpractice insurance through a direct channel.

NDBI — Catastrophic risk is the hardest risk of all for a business to protect itself against

NDBI, which stands for Non-damage Business Interruption, represents the next wave of innovative insurance solutions that have been in existence for a considerable period. However, the recent pandemic and subsequent shutdowns in various economies have significantly amplified interest in this area.

While traditional Business Interruption (BI) insurance is centered on covering actual physical damage and necessitates evidence linked to that damage, as typically outlined in the policy, NDBI deals with scenarios where there’s no physical damage. These scenarios might encompass business interruptions stemming from factors like political unrest, pandemics, cyber attacks, war, etc., situations that typically result in business interruptions. Most of these events lead to medium to long-term economic repercussions that demand a sophisticated insurance, risk transfer, and resilience solution.

While traditional BI relies on standard, tangible risk identification and assessment methods, NBDI necessitates an entirely different and sophisticated approach. Current NDBI products in the market are more tailored than mass market applicable. For instance Munich Re offers EQuIP (Earnings Quality Insurance Protection), a Non-Damage Business Interruption solution specifically designed for the Pharma industry. This specific solution provides a bespoke risk transfer solution for the Pharma industry covering regulatory related production shutdowns. AON on the other hand has been in the market since 2018 with a more broad NDBI product using data, analytics and parametric triggers to cover a broad range of businesses, including hotels, retailers, pharmaceutical firms and transportation companies.

Source: Swiss Re Institute

My passion for innovating in areas that directly support small businesses’ success has been a driving force. Small businesses are not just the backbone of the economy; they embody the essence of the American dream.

Glow came to life because I saw firsthand how tough it was for my family members running small businesses to get the right insurance coverage. They were stuck with outdated tech and solutions that just didn’t match their needs. Having personally observed these struggles, made it clear: this needed to be fixed.

2.   You didn’t spend your career in the insurance industry, what drove you to start a company focusing on the insurance needs of small businesses?

I’ve had a lifelong passion for technology and design, particularly in making technology both accessible and simple. However, building another vertical SaaS application or an AI feature didn’t feel fulfilling enough and definitely not as far-reaching in terms of human impact.

I recognized a critical gap in the insurance sector, especially concerning small business needs where they have been left underserved and overcharged. Using technology to solve this is a massive opportunity to make a real difference in an industry that affects so many small businesses and their employees.

3.   What are some of the challenges unique to serving small businesses and how is Glow handling them?

The biggest challenge is that it is often unprofitable to acquire and service small businesses. Addressing this problem requires automating and driving down cost for every part of the business. This creates a large product surface area for both customer-facing and internal tools. Rather than piecemealing disparate systems, we are building the Glow platform from the ground up, orienting around the ability to react, respond, and adapt to continuous data and process changes in real-time.

4.   How do you see AI impacting the future of insurance?

Throughout history, insurance has led the way in understanding risks through data. Back in the day, companies like the East India Company and Lloyd’s of London relied on shared knowledge and historical data to manage risks, laying the foundation for modern risk assessment.

Looking ahead, AI will transform how insurers use data, offering more personalized policies, underwriting will become incredibly precise, claim processing will become highly efficient and fraud will be more easily mitigated helping further drive down insurance cost.

5.   As a serial entrepreneur and founder of multiple companies, what advice do you have for new founders; what do you know now, that you wish someone had told you when you first started?

Be unreasonable. I am always reminded of my favorite Bernard Shaw quote: “The reasonable man adapts himself to the world: the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man.”

Go all in. It won’t be easy is an understatement, nothing worth achieving ever is. The burnout is real, disappointment is real, the pressure is real and the only thing that will get you through it is if you are passionate enough to “burn your boats behind you.” Success is not guaranteed but you can only do one of two things, go all in for success or brace for failure, no matter the odds.

Trust your intuition. Another advice I would give new founders is to trust your intuition, and I’ve gotten better at this over the years as I’ve discovered that every time I don’t listen to my gut I make the wrong choice.

6.   What’s your go-to interview question for new hires?

I start by asking, can you share an example of a complex challenge or project you faced and how did you go about solving it? This question gives me insight into how they approach problems, especially whether they apply systems thinking. I like to go deep and see if they can clearly articulate details and demonstrate their capacity for handling complex tasks, which is crucial in a startup environment.

I spend most of my time determining cultural fit and communicating clear expectations. I ask questions like: “What excites you about working in a startup environment, and how do you handle demanding situations or tight deadlines?” These questions help gauge their enthusiasm for the startup culture, and their willingness to commit to the demands of a startup.

7.   How have your investors like AV8 been most helpful to you?

Raising capital isn’t easy, but finding the right investors is an even greater challenge. Our Series A investors, particularly Ben Malka from Cota Capital and Amir Kabir from AV8, have been incredibly valuable. What sets these investors apart is their partnership mentality and the qualities they bring to the table.

Having partners who genuinely understand and support our mission has been a game-changer. Their industry and adjacent experiences have provided us with invaluable insights, allowing us to leverage their expertise in navigating challenges.

But beyond their expertise, what truly sets them apart is the relationship we’ve built based on mutual respect and trust. It’s not just about the investment; it’s about forming a partnership where their guidance and support go beyond financial backing. Their commitment to our success and the genuine partnership they’ve cultivated has been instrumental in our growth.

8.   What have you recently read or watched (books, TV, movies, etc) that others should know about?

I am currently immersed in “Amp it up: Leading for Hyper growth by Raising Expectations, Increasing Urgency, and Elevating Intensity” by Frank Slootman. I highly recommend it. This book delves into topics that resonate with me and I find Slootman’s approach both refreshing and insightful.

AV8 has been working with Jack and the Measured team since 2021.  Given the ever present and growing risk from cyber threat actors, we asked Jack to provide his insights and views on the state of cybersecurity and cyber insurance. 

Global Cybersecurity Threat Landscape
AV8: Attacks, especially ransomware, seem to be perpetually on the rise (some surveys suggest 3/4 of orgs have experienced some type of attack).  What can you tell us about the nature of the threats today – but also threat vectors that are on the horizon that we should be watching?  

Jack Vines:  Ransomware is certainly top of mind for everyone. At Measured AI we have one of the industry’s best repositories of ransomware attack data.  Severity and frequency of attacks have been on the rise for at least 4 years. There are growing numbers of events, with more ransom demands but also higher consequences if ransoms aren’t paid.  For the cyber insurance industry, this was a significant driver of losses into 2022.    

However, something curious happened coincidental to the Russian invasion of the Ukraine – ransom attacks seemed to slow down – no doubt due to threat actors for hire being occupied.  Also the insurance industry tightened their requirements given the loss history.  Another important driver has been law enforcement’s emerging ability to track crypto-wallet ownership and seize assets in some situations.  As such, this year we have seen a dramatic increase in financial fraud i.e., convincing a target to wire funds fraudulent bank accounts which are then liquidated quickly.  

But in late 2022 and in the first half of 2023 year both the frequency and severity of breaches began to accelerate again.  The frequency, magnitude (incl. types of systems and industries being targeted, plus ransom demands) and severity of these events – e.g., data being destroyed, new attacks on the same target 1 day later – are all going up.  Unfortunately, it’s an arms race between threat actors, their targets and insurers – the latter two who are running to catch up with the former.  That said, we haven’t really seen big systemic attacks from state actors recently, thankfully, even though the threat of all out cyber-war risk is certainly top of mind for carriers.  

As the threat actors hone their processes they are becoming more and more sophisticated operationally and acting more “business-like”.  For example, there was a situation recently where a target paid a 7 figure ransom but was not provided encryption keys to unlock their data.  After some time, the company finally heard from a “supervisor” at the threat actor saying “we are so sorry…so and so was on vacation and no one was checking his email…if you ever have any questions or issues please contact me directly” and provided them their keys.  So in other words, the threat actors are getting more and more professional and operationalized. 

Enterprise Security Posture
AV8: What are some common mistakes companies (and public sector entities) are still making that are putting them at risk?

JV:  Unfortunately, people are generally still the weakest link in the chain. For example, companies are operating on such short timelines today it’s common to hand out credentials to people who shouldn’t have them or more challenging – granting users way too many rights than they should have.  We still see a lot of challenges with email security where security protocols aren’t configured properly, allowing lots of phishing emails through.    

Further, the “shadow IT” problem where a group within a company quickly spins up servers in a public cloud environment, putting it on a credit card, but bypassing corporate security controls yet still accessing other corporate data repositories or systems can be a substantial threat vector. While processes and security are certainly getting better over the years, despite security teams and risk carriers banging the drum for better employee training, MFA, endpoint security, email phishing there are still large measurable gaps in these basic areas.   

Another challenging area is the growing supply chain of vendors and services that companies rely on.  It’s now not uncommon for even small to mid-size enterprises to have hundreds of vendors, and of course large global entities will have thousands all with so much more digital connectivity and access to data than before.  It is very challenging to stay on top of all these 3rd parties from a security perspective and we expect a growing number of attacks coming through these 3rd parties.  (Editor’s note:  There have been some well publicized breaches that have come via a vendor like Toyota who had an automotive supplier compromised and Home Depot who had a vendor’s credentials harvested)   

Regardless of these challenges and emerging threats, at Measured we believe that speed is of the essence and have made that part of our DNA.  You must have real-time data coming from the insured to understand where they are vulnerable when a new threat is discovered elsewhere.  It’s a game of “whack-a-mole” where you have to move as fast as the threats move to close potential gaps.  Additionally, speed of remediation is vitally important when key corporate systems are at risk, another area where Measured shines.  Adapting to these changes so quickly can be a challenge with legacy risk providers.   

Current Challenges for the Cyber Insurance Industry
AV8:  Let’s start with the customer base.  Where do you see the customers for cyber-insurance today, where are their mindsets and what are their initiatives?

JV:  Historically, a lot of new policies were small and medium size businesses with 1st time cyber policies (as well as very large enterprises who tended to self-insure).  However, risk awareness is growing in all types of customers.  At Measured we are seeing more sophisticated buyers than in the past, who started with basic coverage that perhaps was too expensive, didn’t meet their needs, or wasn’t providing them with real time insights regarding threats and risk, who are now looking for a more sophisticated and holistic risk partner.   

Another interesting dynamic is that historically, cybersecurity tools were purchased by CISOs and CIOs while cyber-insurance coverage was purchased separately by CFOs (often through insurance brokers who were non-technical and couldn’t really assess risks).  We are starting to see more integrated and holistic risk management programs that combine both, where there’s a comprehensive assessment of risk that can be mitigated by investment in security tools along with insurance required to cover risk that can’t be mitigated by these security initiatives.  

It’s also important for the C-suite to understand the full cost of a cyber breach.  For example, they understand that all parts of the business are impacted during a ransomware event and just paying the ransom is the tip of the iceberg.  We need to help them understand all the other costs – like the costs of business interruption, reputational damage, fines and penalties if they can’t access servers for a certain period of time, or true costs of remediation if there is an incursion.  And then take all this into account when thinking about appropriate coverages.  

AV8: As part of this more holistic approach, how do you see companies partnering providers like Measured today?

JV:  The industry has generally developed more skills and discipline in writing policies.  We are starting to see the industry now pushing more requirements on customers i.e., for us to provide coverage you need to demonstrate not just MFA is in place but on which key systems; or not just endpoint protection, but how many endpoints are under protection vs. not.   This is much different from a few years ago when all you had to do was answer 5 questions on a simple form to get coverage.  Now, companies are more willing to share this information and collaborate more with companies like Measured who can give them real time threat information.    

AV8: What are some unique approaches Measured is taking to respond to these changes in the landscape and industry?

JV:  Speed and monitoring, which go hand in hand, are some of our core principals. We partner with industry leading security tools providers like Tenable and Sophos to ingest their data, analyze it in real time, quickly notify clients if they are at risk, and help them mediate threats.  This takes a significant amount of technical expertise, systems interconnectivity, but also trust from our customers to partner with us in this way.  For example, earlier this year we saw the MOVEit attack vector which exposed hundreds of companies using this common file transfer service.  We saw it first on a Wednesday, assessed the impact on all our insured by Friday, and by the following Monday/Tuesday we helped our clients remediate the threat.  You need to have systems, tools, capabilities and trust to make this happen.    

AV8:  And what about cyber-insurance pricing?  Premiums were rising for some time. Where are things now?

JV:   Industry pricing is a lagging indicator and well behind where actual risks are based on just looking at historical data.  At Measured, because we see risk in real time, we feel that we have an advantage with regard to accurate and timely pricing of risk.  That said, with the rise of ransomware the last several years, demand for coverage dramatically outpaced capital available from risk carriers which drove up prices markedly.  However, in the last 6-9 months prices have stabilized as carriers have generally gotten better at pricing risk.  But ability to price risk can change quickly and we still see true risk not being priced into the market.  Currently the frequency and magnitude of cyber events are rising dramatically, and that risk is not being priced into the cost of insurance. As such, we expect to see a rise in claims in the future which will have upward pressure on pricing.  Otherwise the market is not sustainable. Given this, we also expect to see more companies moving to self-insured given where pricing can be especially for companies with substantial risk. 

Future of the Industry
AV8:  To close out, give us your thoughts on what the cyber-insurance industry looks like in the coming years?

JV:  As we mentioned already, speed and the growing interconnectedness of enterprise data and risk/insurance efforts is a trend from which there’s no going back.  It will be difficult for companies who haven’t invested in these skills or can’t move at the pace required by the ever-changing threat landscape to thrive.    

I also feel the industry will become much more global.  Geographically, today the bulk of cyber policies are in the US and Europe.  However, companies are operating globally.  More importantly cyber events span across geographic boundaries, and the threat actors are global as well.  We anticipate accelerating industry growth in Asia and Latin America including policies that span multiple countries, and regulatory frameworks that support this given how interconnected businesses are today.   

2. What got you interested in the manufactured housing industry and in particular insuring those types of homes?
   24M Americans live in manufactured homes (7% of the US population). It’s core to the American Dream of American homeownership. The pandemic taught us it’s the future of how we’ll live. It’s difficult to build homes onsite (1) getting materials there: supply chain (2) hiring talented labor on-demand. On top of this, housing costs and rent spiked. We saw people moving out of major cities and into more rural/remote areas.

   It makes sense to build homes in an air conditioned factory on an assembly line–efficiency with material waste and production is more sustainable. In 2022, 10% of all new home builds were manufactured. Traditional manufactured home production will continue to boom, but there’s an emergence of a new segment of modular homes, tiny homes, ADUs, etc. While manufactured homes are the future of the way we’ll live, consumers can’t get a traditional mortgage or insurance policy for these homes. To move the industry forward, we’re supporting financial services by building a fantastic and easy insurance product.

3. You founded the company in Detroit. Tell us about that decision and both the unique opportunities and challenges it presents.
   There are 250,000 manufactured homes in Michigan–we are big consumers of manufactured homes. It’s important to be close to our consumers. The industry is based here: within a 30 min drive sits 20% of the top community owners. 1 in 3 major manufactured home communities in the country is owned by a Michigander. We have great insurance talent in Michigan.

4. You’ve spent time visiting manufactured home owners across the country. What have you learned from these visits to “middle America”?
   In starting CoverTree, I’ve stayed in numerous manufactured homes across the country. Additionally, we’re in communities on a weekly basis. The biggest thing we’ve learned is the sense of community and pride in living in a manufactured home. Additionally, in these homes, people can truly live more “off-the-grid. ” manufactured homes are contained and can be more energy efficient with energy usage.

5. How do you think about and approach building your team, given your location?
   We enjoy having shared experiences with our teams. As a result, we’re from similar cities and states. We grew primarily through our personal and professional networks.

6. What’s an under-appreciated talent you look for in new hires?
   The ability to get obsessed with anything!

7. What’s an important trend or technology in your space (i.e., Insurtech) that you are following?
   Continuing to get more and more current data for underwriting and rating. Additionally, more complex weather data and modeling—i.e. how can we predict the weather?

8. What advice would you give aspiring – yet first time – entrepreneurs who are thinking about starting a company?
   Look where others aren’t and be obsessed.

9. How have early investors like AV8 been most helpful to you?
   Amir has been a great friend through building CoverTree. He’s often the first person I go to for opinions and bounce ideas off of. As we continue to evolve, AV8 continues to be a huge supporter of our team and vision for the company.

10. What have you recently read or watched (books, TV, movies, etc) that others should know about?
    I enjoy watching documentaries on the history of the United States. We’re a young country, but history seems to repeat itself.

Embrace is built out of what my co-founder and I, and a bunch of investors in first generation mobile games and tools like MoPub and Testflight had seen firsthand. A mobile device, whether a phone, point of sale system, or appliance, represents a user and their experiences, behaviors, and pains. The existing vendors treated mobile like a website or server side application, even though it is compiled code run on disparate devices, in different countries, and through different carrier connections. While a ton of tools existed pitching mobile, no single tool had the end user as its index with every one of their session’s at its fingertips.  Instead my team hopscotched through product analytics, APMs, customer experience, logging, and crash reporting tools with no guarantee an issue would be found or the data would be collected and accessible to resolve. In other words, I had an extreme lack of tooling focused on the uniqueness and extreme potential of mobile.

Embrace was born to help all mobile teams, especially engineers, build the best experiences.

When starting Embrace, I believed that mobile would become more integrated into our daily lives, way beyond just games. Now, mobile experiences travel with us on our phones, wrists, cars, and devices. They shape the rhythms of how we work, live, and play as we weave them into our environments and processes. More than any other technology, mobile is transforming us by instantly and meaningfully integrating technology and people. Embrace allows developers to bring their boldest visions to life using the only platform created for building a better future.

2. You have a unique educational background:  Undergrad EE/CS degree plus a JD/MBA.  Which of those skills do you think helped you the most in your career as an entrepreneur?

All of them?  When starting a company, skills are required across product, engineering, fund raising, and legal. While many founding teams derive that value from different people, I am the ultimate jack of all trades.  With hands-on experience, I can accelerate different aspects of the company to quickly push through what would be blockers or slowdowns for other founders.

Although Fredric, my co-founder at Embrace, would argue that one degree I’m not allowed to use is coding.  That one I have to admit has atrophied although I may have submitted some website changes a long time ago and plan to do a PR to provide the team an unexpected laugh.

3. In what ways did the pandemic and all that followed impact Embrace?

For our business, the pandemic was a tailwind in the making and one we took advantage of. Mobile devices and apps were an increasingly integral part of everyone’s lives and Covid hastened this integration as the work from home movement happened overnight. Our customer base grew QoQ during the pandemic because mobile was a necessity for businesses with grocery store pick-ups to telehealth.

For our team, we adapted quickly to safe-guard our amazing culture we built over the years. We shifted  meetings, added virtual happy hours and donut buddies to the mix. The common-truths is that we as humans enjoy working with people with common values with opportunities for in-person contact. When it was safe to meet in person, we were excited to come back together because as humans, we cannot solely operate over Zoom (or Google Meet).

4. What Advice Do You Have For Entrepreneurs Who Are Just Starting Out?

Go fast and break things. Making mistakes is the key to learning and achieving success. These mistakes mean you discovered something and, when the discovery is shared, everyone learns. Learnings compound week over week into growth. Making, openly discussing, and stopping mistakes is required to grow quickly. Only by growing and failing together do you achieve success.

Create a great culture. No opportunity cost is higher – it is always better to wait for the correct hire as opposed to stopping your search and then restarting it when that poor hire inevitably fails or, even worse, underperforms and lingers in your company. We are a kind team that encourages blunt feedback – this is the best environment for your entire team to grow and experiment.

5. Outside of functional skills, what do you look for in new hires for Embrace?

When we hire at Embrace, we evaluate culture fit and technical acumen separately. Whenever companies mix the two together, they inevitably bias to technical over culture. Culture fit is so much more important! We will hire a partial thumbs up on technical with a full thumbs up on Culture every day of the week. But we must never hire a candidate with a partial thumbs up on Culture. We interview culture based on our culture tenets of: Perspective – we seek to understand your perspective, Investing – meaningful insights are found by investing time, Honesty – brutal honesty delivered kindly, Simplest – Simplest solutions are identified by focusing on outcomes, Ownership – Growth is achievable only by empowering with ownership, and Dark Humor – finding joy and humor in any situation.

6. How have your early investors like AV8 been the most helpful to you?

When I look to lead a round of funding, I look for a partner. By “partner”, I mean a person who understands their role in starting and growing a company, and is also empathetic to the founder.  The alignment does not exist when startup founders often lose focus on activities that do not drive true business value.

I have been fortunate to be in a position at each round of funding to be selective of my VC partners. Hadley Harris at Eniac, Baris Aksoy at AV8, and Aaron Jacobson at NEA are three partners that I’m glad to have on the Embrace journey.  Specific to Baris, as a recent operator he really understands my day-to-day and how not to step in directly but instead to lend a healthy ear or provide external driven insights on the market and category dynamics.  I am lucky to now consider Baris a friend even beyond Embrace.

7. What are your daily routines and how do you allocate your time? Walk me through a typical workday for you.

I try my best to workout every morning but life takes over a lot of the time, this is something I am actively working to get better towards. Eating healthy with morning smoothies and a salad + protein at lunch keeps my energy levels up during the day. I’ll reserve a night in the office once a week to get into my flow state and be able to think creatively on vision and larger projects.

8. What have you read recently or watched (TV, movies, etc) that you can’t stop recommending to others?

A few weeks ago as traffic picked up in LA, I started to listen to podcasts – sometimes with my kids.  The primary two are

• Stuff You Should Know is my 10 year old and my favorite. Covering topics that we have all wondered about but never taken enough brainspace to truly research. With a whimsical and nerdy approach, the hosts tackle these topics with enough depth to hold your attention and teach you something. Think Loch Ness Monster, the effect of picking your nose, and the history behind the Delorean
• All In is a cliche one to choose as its hosted by 4 prominently known VCs. Even so, I listen to it for perspective both in terms of VCs vs founders, for which I dont always align, and outside of the venture world.  Their takes on the Ukraine-Russia war are particularly enlightening and have changed my perspective on the world.

In recent weeks an incredible amount of attention has focused on generative AI technologies like ChatGPT, and the recently launched GPT-4.  However impressive these models are, they are still somewhat rough around the edges.  To wit:  just ask ChatGPT to explain long division with decimals to a grade schooler.  The answer will be incorrect and full of inaccurate examples that would easily confuse a novice trying to learn:

Chat GPT:  For example, if we are dividing 10 by 3 using long division, and we get the answers 3, 3, and 1 above the digits, we know that the divisor (3) goes into the dividend (10) three times, with a remainder of 1. So the quotient is 3 + 3 + 0.1 (which is the same as 0.3) = 3.3.

Microsoft poured $10B into the company, and rushed its ChatGPT-powered version of the Bing search engine to market.  After initially receiving glowing reviews, it became clear that more testing and refining would be needed for the new AI-powered search engine to live up to the hype.  Like the Lenza generative AI that came before it, which struggled to keep the avatars it output PG, Bing’s implementation also had issues that even tech giant Microsoft couldn’t smooth out in time to avoid a somewhat embarrassing series of reports about unsettling interactions with the chatbot.  Indeed, ask Bing for help scheduling a flight, and the answer you get might be far from correct:

Modern AI systems are truly remarkable, but it’s easy to get carried away when the system performs well on “happy path” scenarios, and become overconfident it’s ready for prime-time.  This kind of optimism is all too familiar to any practitioner who has deployed AI systems in real-world settings.

The right deployment context matters

Even in its current, imperfect form, generative AI can still be very useful.  For example, generative AI has been reliably used for years to create more realistic speech from text, to draft marketing copy that can be checked and edited before use, and more recently, to interactively recommend the next line of code which a developer can decide to use or not.  Generative AI is being adopted by Hollywood to bring a younger version of Tom Hanks to life in the upcoming film Here which undoubtedly will be edited by dozens of humans to correct any errors that might crop up in the process.  Even Lenza put a human in the loop when they allowed users to select which avatar they liked the best.  “Human in the loop” is one solution for AI’s rough edges.  The other is vertical or domain specific applications.

Consider vertical applications that are more easily verified

Vertical applications of generative AI hold much near-term promise, because the AI need not be an expert in everything, and a narrower use case is naturally easier to test.  For example, large businesses have already adopted AI chatbots to reduce customer service costs, and the next generation of these bots will undoubtedly be powered by generative AI, to provide a more satisfying and realistic user experience.  Because the questions you can ask a bank’s customer service chatbot or call center are relatively limited in scope, the answers can more easily be verified.  And because the subject matter your virtual customer service rep is expected to cover is relatively limited, it’s easier to put in guardrails to prevent embarrassing situations when the user asks questions that are far afield from the AI’s designed purpose.  The same holds for sales productivity reports, business financial reporting and summarization, etc.

Any experienced AI practitioner will tell you that the last 20% of quality or accuracy takes 80% of the time and budget (like in any engineering project).  For Ai projects though, the key question is 20% of what?  If the problem is too big, or too high stakes, that 20% can quickly become unmanageable.  We have witnessed this in the self-driving car hype that is taking years longer than originally promised to come to fruition.  In high stakes scenarios, e.g., when human lives are at stake, or big financial risks are being made, it’s really important the system be thoroughly tested.  Yet, more thorough testing typically generates more issues that need to be fixed, resulting in delays, which drive up costs.  Of course, fixing the self-driving car so that it doesn’t suddenly brake in the middle of an intersection for no apparent reason is necessary, but the steady stream of unanticipated problems like these can lead to a money pit of engineering cost, especially when the tooling required to debug and correct AI models is still relatively underdeveloped.

While ambitious, high-stakes, general-purpose AI problems like these are worth solving, and much progress has been made, a less expensive and less risky path is to focus on better constrained, vertical applications that can be enriched with domain knowledge, and for which a more comprehensive suite of test cases can be more easily produced, even when the stakes are high.  This is the approach I took at Zest AI, where we built AI systems that automatically generate credit underwriting models that move 100s of billions USD.  Because the credit underwriting problem domain is relatively constrained, it was possible to enrich our automation tools with domain semantics.  This domain knowledge complemented algorithmic learning algorithms to ensure the models our automated pipelines produced learned what we wanted them to learn, and could be made transparent to human reviewers so they could be validated. 

Apply a risk management framework to your AI investment decisions

The more constrained the problem, the easier it is to get the AI to work, and to verify that it is indeed working.  The less constrained the problem is, the more investment will be required to gather data, design the algorithms, train and retrain the AI, and all that effort drives up data science headcount and timelines.  Moreover, testing costs balloon with less constrained problems, because a human has to verify the outcome is correct, and in an unconstrained problem domain, even just a 1% sample of potential outcomes is significant (after all, infinity times a constant is still infinity).  When stakes are high, and the cost of a bad outcome is high, these costs compound.  Add in regulation that requires independent validation and verification, and the costs can get very large very quickly.

The risk framework below can help assess AI-driven use cases:

Most successful AI applications hang out in quadrant IV, where the stakes are low and the problems are well-constrained.  The board game checkers, for example, which has no-cost bad outcomes, and is highly constrained, was solved by Arthur Samuel in 1959.  AI has tackled successively harder problems ever since, and there is still plenty of value to be created solving constrained, low-stakes problems, where each individual AI-driven output doesn’t matter that much, but the aggregate outcome can be improved as a result of incremental accuracy.  For example, when you upgrade from basic rules-based or linear modeling methods to machine learning models, it can generate significant improvements in business outcomes.  This has led to successful applications of AI in progressively higher stakes problems like underwriting and computer-assisted radiology, where AI can make more accurate predictions on its own than a human could (as in the case of automated underwriting) and in high stakes decision support scenarios where there is a human in the loop (as in computer-assisted radiology).

In recent years, practitioners have started to focus on less constrained problems, like the kinds of applications we see in quadrants II and III.  Novel deep learning, attention mechanisms and adversarial training methods allow for the construction of generative AI models that produce human-like outputs.  When getting it wrong has insignificant consequences, generative AI products have already been brought to market with varying success. For example, generative adversarial networks (GANs) are already built into widely-used products like Google Pixel where they are used to de-blur, de-noise, and correct the lighting in photos you take, and users love the result.  If the process doesn’t work, the original photo is still available.  But more ambitious applications have mixed results, which is a problem when your reputation is based on the accuracy of your answers.  Even though the consequence of one bad search answer is relatively small, factual errors in Google’s Bard announcement led to $100 billion in market cap disappearing overnight.

There is plenty of money to be made where AI can be used to create unique experiences people will pay for or that can be monetized by advertising, even when the stakes are relatively low.  For example, we are sure to see a slew of video games announce they have LLM-enabled NPCs in the coming months, and there will be an ecosystem of tools developed to make adding avatars and bots to various low-stakes experiences easier.  I am sure that people will be delighted by these developments, and over time, the solutions may migrate up to Quadrant II.

In Quadrant II, the stakes are high and the problem is unconstrained.  This quadrant is the hardest, the highest risk, and the most expensive.  Model complexity goes up dramatically.  There are networks of models that all need to work together to create the desired outcome.  Debugging one model is hard but when you have a complex system of models, things become even harder.  For example, consider applying LLMs to a high-stakes task like summarizing medical test results for patients, or drafting legal agreements.  For certain types of test results and certain types of legal forms, I bet an out of the box LLM would be pretty good.  But when it makes an error, how do you correct it and make sure it is corrected?  As a novice, how do you know whether the answer the AI gave is correct or is an error?  Hard to say, and I cheated – I moved the problem from Quadrant II to Quadrant I.  Until we build up enough confidence and trust in models, there will be a human in the loop reviewing the output, and many AI solutions will be “suck” assisting humans, versus replacing them.  That’s not all bad.  AI systems in a decision support role, generating first drafts, or in other “human in the loop” scenarios, can still improve efficiency and drive down costs.

Because of the high cost and uncertain timelines, we will likely mostly see tech giants with deep pockets attempt applications in Quadrant II for now.  But what will it take to see these high value use cases come to fruition?  I will tackle this question in my next post.

While we have seen much progress in generative AI, we still have work to do.  General purpose generative AI isn’t ready for prime time yet, but with the right risk management framework, you can pick applications that avoid runaway expenses and are more likely to succeed.

What will it take for general purpose AI to impact high stakes problems? 

So far we’ve outlined a risk management framework for AI and encouraged a focus on vertical, constrained applications.  This is because general-purpose AI is very hard:  humans are very good at qualitative reasoning, “common sense”, and emotional intelligence; computers simply are not.  Humans have a notion of what is socially appropriate, about how confident they are in their answers, and they can explain where they learned what they learned and why they think they are right.  Most AI can’t do any of this yet.  Even just controlling what an AI learns from data is very difficult, but not impossible.

More Progress is Needed on AI Fundamentals and Better Tooling

To make general purpose generative AI a reality for all of the use cases we can dream up for it, without humans in the loop, we need three fundamental technology leaps and some good engineering:

1. Introspection, the ability for the AI to understand why it makes decisions and explain them to humans
2. The ability to incorporate and enforce logical and qualitative constraints on models
3. The ability to better influence what models learn
4. Better tooling for simulation and model verification

The good news is, progress is being made in all of these areas, and it’s only a matter of time before effective techniques move out of the lab and make it into industrial applications.  These problems create opportunities for companies seeking to solve challenges that will allow next generation AI solutions to flourish.

AI Introspection

Understanding what a model has learned is a fundamental challenge.  Model explainability methods based on the mathematics of game theory (e.g., SHAP and Integrated Gradients) have opened up “black box” models like XGBoost and Deep Neural Networks.  In the context of tabular data, where each data attribute has known semantics, these techniques are already finding useful applications in industry.  But when applied to LLMs and convolutional nets, the results are still difficult to interpret.  More progress is needed on LLM and vision model explainability, to summarize the internal representations deep nets have inside them so they can be more easily reviewed in advance of deployment, and in the context of generative models, to explain why a particular output was generated.

While a model may have learned the right thing, getting a vision model to summarize what it learned in a way humans can easily understand remains a significant challenge. Moreover, understanding how data impacts what a model learns is still an open area of investigation.  Further research on influence functions is needed to summarize how model outputs relate to training data instances, so models can be more easily corrected by revising the training data.  Likewise, more progress is needed to quantify the extent to which a model is confident in its answer, and that it has adequate support in the data for the predictions or outputs it’s generating.

User:  You seem pretty confident about that 

ChatGPT:  Yes, I am confident in my explanation of the long division process for dividing decimals. Please let me know if you have any further questions or if there’s anything else I can help you with! 

User:  How can you be sure? 

Chat GPT:  As an AI language model, my responses are generated based on patterns and rules derived from analyzing a large amount of text data. I am designed to provide accurate and consistent information based on the rules and knowledge that I have been trained on.

When a model prediction or output comes with a real (not canned) estimate of confidence, and the reason a model generates an output can really be interrogated, the user can more easily develop an opinion about whether they should trust the answer.

Better model introspection tools will allow the systems these models are embedded in to migrate from low-stakes to higher stakes applications, and allow model developers to more easily correct mistakes made along the way.  Solving these problems may also allow the next generation of ML ops companies move beyond the relatively mundane (yet still important) artifact management, experiment tracking, deployment, inference infrastructure use cases.

Incorporating causal and qualitative constraints

While LLMs and GANs can generate some pretty incredible outputs, to operate safely without a human in the loop, model developers need to be able to enforce constraints that ensure they won’t go off the rails.  Additional work is needed to add complementary semantic and qualitative reasoning models to statistical learning algorithms in order to prevent bad things from happening.  The ability to add simple constraints to ML models is widely available today.  For example, directional constraints that control how an input should change a model’s output (known as monotonic constraints) are already commonplace.  While this is very useful in the context of tabular data, where the meaning of an input attribute is well-known, and the desired direction of the relationship to the output is also known (e.g., higher prices lead to lower take rates), when we consider more complex models and constraints, things get a bit more difficult.  For example, consider how to prevent Lenza from generating sexualized images, or how to ensure ChatGPT gets its arithmetic right.  This is not as straightforward as it may seem, and incorporating such qualitative and quantitative models into a statistical learning process is an area of active research that is already showing promise.  Tools that allow model developers to easily incorporate such constraints into the model training process will become increasingly important.

Influencing what a model learns

Machine learning practitioners are all too familiar with the age old adage “garbage in, garbage out.”  The quality and quantity of training data can have a dramatic impact on model performance, whether the model is used for classification, regression or generation.  However, available training data often reflects human bias that needs to be corrected, or simply distributional anomalies due to the particular sensors used to gather the data that won’t generalize to the real world.  Influencing what a model learns from data is difficult but not impossible, as a model’s objective function can be modified to incorporate feedback from a critic or adversary.  At Zest, we successfully applied adversarial training to generate models that achieved fairer outcomes for women and people of color, while sacrificing very little, if any, predictive accuracy.  Other approaches include re-weighting input data, and filtering or correcting model outputs with other models.

Techniques like these can be applied to generative models to ensure the generated outputs meet a desired qualitative bar, and don’t get called “hyper-sexual” “creepy” or “unhinged.”  This creates an opportunity for infrastructure companies that allow model developers to more easily incorporate feedback from secondary models to make generative AI applications safer and easier to correct, or companies that provide datasets or canned models that generative AI model developers can use to filter outputs.

Next-generation tooling for reinforcement learning and model validation

To make more rapid progress on AI for high-stakes and less well-constrained problems, we need better tooling to ensure the AI will do what we want it to do.  Feedback mechanisms are important not only for verification but also for generating new data that can be used to train the next iterations of models. Simulation allows reinforcement learning systems to explore potential outcomes and receive feedback that enables models to more rapidly learn, but all RL frameworks require a simulator and an outcome measurement.  Creating these today is expensive and hard, which creates another opportunity for solutions.

Likewise, without tooling to validate models, it’s difficult to be confident they will perform as expected.  This is compounded when you consider generative AI applications in open-ended problem domains.  Tools for managing the testing process, generating test cases, tracking regressions and enabling rapid large-scale evaluation of new model iterations will reduce cycle times and speed the generative AI review process, enabling more LLMs and latent diffusion models to make it out of the lab and into production.

Conclusion

While general purpose generative AI isn’t ready for applications that are open-ended, high stakes, and without a human in the loop yet, with further investments in fundamental technologies and tools to help manage AI risks, we will begin to see more and more AI-powered applications have commercial impact.