Understanding the GenAI Tech Stack : Part 4 — Application Models

Raphaëlle d'Ornano
5 min readNov 7, 2023


Google DeepMind

Although tech stocks have seen recent volatility following a low earnings season, generative artificial intelligence (GenAI) continues to inspire investors, with Sequoia calling it “our generation’s space race.” Even after a cooling-off period in Q3, 2023, GenAI investment is still on a long-term upward trend. This sustained optimism stems from the belief that GenAI is not another tech fad, but instead represents a profound technological paradigm shift that will reshape industries and disrupt conventional business models. The real difference between 2023 Q2 and Q3 is that investors and founders are coming back down to Earth after realizing that not every interesting AI idea can be translated into a strong business. Growth at all costs is no longer a recipe for success, even in GenAI.

While short-term market fluctuations are inevitable, the long-term potential of GenAI to revolutionize various sectors, enhance productivity, and unlock new revenue streams remains undiminished. GenAI has already had a more successful launch than software as a service (SaaS) did in its time, reaching in months what it took years for the SaaS industry to achieve. In short, today’s investment mandate is about analyzing the intrinsic value in GenAI companies.

As part of our series on understanding GenAI, this month, we’re discussing the final model in the tech stack: Application Models. (See more about Foundation Models here and MLops Models here).

Understanding Application Models

Application Models focus on user-facing applications where GenAI is a core value proposition. These companies fall into two buckets: brand new, native GenAI companies and SaaS businesses that have been completely transformed by AI to their core. This model does not include companies that have merely tacked GenAI onto a few isolated processes.

Both buckets have two possible growth axes: Vertical and Horizontal. Vertical Application Models are industry-specific; examples here include abridge, Harvey and Wonder Dynamics. The medical and legal verticals in particular are two core fields of application, and where we see a lot of potential. Horizontal Application Models are workflow/function specific, and include companies like Jasper, copy.ai and Tabnine. Winning Application Models demonstrate clear differentiation, good customer retention, and strong margins/profitability (at least in late stage).

Whether these applications are built from the ground up or rely on third-party APIs, this category is akin to the state of mobile phone apps 15 years ago. On the positive side, there is tremendous opportunity and significant room for expansion. But like phone apps, the Application Model category is experiencing similar challenges and an abundance of low-quality vaporware. Even today, 95% of downloaded phone apps are abandoned after one month and 25% are used only once.

After the initial tsunami of interest in GenAI apps at the beginning of the year, we’re already seeing disappointing products that fall far short of expectations — most visible in product-market fit and declining user retention. Though consumers are breaking down doors to try GenAI — ChatGPT took 5 days to reach its first million users; Spotify took five months — they’re finding the value isn’t yet there. The best consumer companies have 60–65% daily or monthly active users (DAU/MAU); GenAI apps have a median DAU/MAU of 14%. These challenges have been exacerbated by ongoing debates about AI ethics, regulation, and usefulness.

In short, the problem is not demand or distribution, but proving value to the final customer.

So how should investors cut through the GenAI hype to measure the intrinsic value and resilience of Application Model companies? The good news is that many investors will find familiar territory; at the end of the day, GenAI Application Model companies and SaaS companies share the same underlying business model, with a few key distinctions.

Quality of Revenue and Growth

When considering Application Model companies’ quality of ARR, we first look under the hood to see what’s driving growth — new upsell, downsell, churn, gross retention, net retention, etc. The main difference compared to the SaaS business model is managing cost structures.

For example, we expect to see lower gross margins for Application Models because of how they layer on top of Foundation Models. Foundation Model companies like ChatGPT charge fees on a per unit basis, so the largest and most important debate for Application Model companies today is the extent to which those fees can be passed on to the end customer. While this would be normal in other businesses (higher fees resulting from higher COGS), Application Model use cases are not yet completely defined, and neither is the customer benefit. So far, most of the largest software vendors do not pass these fees to customers and just eat the cost themselves. Startups that do choose to pass through these costs run into difficulties because customers object to paying steep (often 20–30%) fee increases for no discernable benefit. Asking customers to pay more without delivering immediate value results in churn, creating a downward drag on revenue.

Until the use cases are there and the benefits of Application Models become externally evident to the customer, with pricing based on value expected to become a more common practice, companies are in a limbo period of higher immediate costs with delayed benefits. The right pricing structure is therefore critical to evaluating Application Model companies’ quality of revenue, whether they’re vertical applications or horizontal applications.

Another area where native Application Model companies diverge from the SaaS playbook is the level of measurement detail required. For new Application Model companies this means building traction; for transformed SaaS companies it means tracking change. In either case, GenAI is on such an exponential growth curve it’s vital to have very granular monthly and quarterly measurement systems. These systems should track the effect on both a company’s user base and its actual contract base. This kind of close tracking is essential to identifying whether the company is truly delivering the value it promises to its most important stakeholder — the end user.

Cutting through the GenAI hype means analyzing the right qualities to see which Application Models are actually changing people’s lives. GenAI’s Act Two is centered around developing a shared playbook as investors zero in on businesses that can demonstrate intrinsic value through the quality of their revenue, growth, and margins. Uncovering GenAI companies that will avoid the app graveyard and stand the test of time requires a disciplined, discerning investment thesis fed by a constant inflow of thorough data.



Raphaëlle d'Ornano

Managing Partner + Founder D’Ornano + Co. A pioneer in Hybrid Growth Diligence. Paris - NY. Young Leader French American Foundation 2022. Marathon runner.