AI Evolution: An Update

AI is the meteor that threatens to change the SaaS ecosystem and force evolution (and potentially extinction). In early 2023, we kicked off our study of this change by talking to close to 100 companies on practical applications of AI, including hosting two working sessions with 40 executives to understand the impact of AI on the SaaS ecosystem. From those sessions, we came to the initial hypothesis that AI could be neutral or even a tailwind for control point companies, not an extinction event. We published our view in August of 2023 here.

‍

A second wave of adoption has grabbed the hearts and minds of the Valley. This time, the narrative is that AI startups can now offer a product that fully substitutes labor previously performed by in-house staff and third-party software. If AI service providers are able to do that, they can charge on the basis of the value of work and have economics that supersede those of a software company.

‍

Do we think this will happen? Our initial hypothesis is yes, but not in all circumstances. When the conditions are right, if software providers are not nimble, they will be beaten by AI startups. (And we are looking to invest in these native AI companies! Reach out if you want to talk!). However, forward-thinking software companies that occupy control points have strong defenses and may have an opportunity to provide and benefit from AI-powered services instead of facing de facto obliteration.

‍

Writing is learning, both in crystallizing thought and soliciting feedback. By putting this thesis out in the wild, we want your feedback. We want to offer up why we could be right so we can hear back on the more important topic: where and why we are wrong.

‍

AI-Driven Software Extinction (circa early 2023)

The prevailing narrative was that AI is eating software. However, the view we articulated in early 2023 is that AI isn’t the end of software. Rather, it serves as an important step function in automation, forcing evolution but ultimately increasing the importance of a certain class of company—control point software. 

‍

AI infrastructure companies, flush with hundreds of billions of dollars of capital, have made it relatively easy to offer AI functionality. This is especially valuable to control point companies who enjoy data and workflow gravity. This means that control point companies can fine-tune AI models offerings better than any outside provider and offer these automations within the context of their customer's workflow. You’ll have more performative offerings faster than a native AI company can hope to match. The underpinning market assumptions are that the infrastructure layer continues to be competitive—something that seems likely with the rise of open-source and the amount of investment across cloud providers—and data privacy infrastructure continues to evolve to protect app data from being sucked into the generalizable models. 

‍

That’s not to say that an incumbent control point company can fall asleep at the wheel. It needs to be nimble and forward-looking to take on its AI challengers. We detailed these challenges in our “Practical AI Series,” but the key takeaways are:

‍

• Recruit AI expertise and power through initial friction: AI talent is expensive and hard to find. Then, even once you recruit the engineers who make 3x what the founder does, you have to find the time to build a whole new offering that cannibalizes the core product suite. As a fun bonus, this new AI product has zero gross margins, and your sales staff have no idea how to sell it. Then, once you sell the damn thing, many of these AI products are initially service-intensive, which requires yet another net new capability and skill set for your company. Plus, there is a decent chance this move will confuse your investors and impact your valuation multiple in the short term.
• Transitioning pricing models from seat to value-based: Easy to say, hard to do. After all, pricing models are business models. And business model changes have a way of collapsing category leaders. For example, you can think about how the on-premise leaders didn't evolve when the smaller, nimbler SaaS players arrived on the scene (RIP Siebel). 
• Protect your data: You’ll need to avoid LLMs slurping up your data or having your customers contribute your data to a data lake where you start losing control of the UI.
• Lean into natural language/chat and voice interfaces: AI is a step function improvement in these engagement styles particularly for ad hoc analysis.
• Own the workflow and empower the human in the loop: Many AI use cases provide 80% of what you need. However, many tasks require more than 80%, and if your workflow brings the human in the loop to verify, you continue to own the control point and the unfair right to be the provider of AI.
• Be wary of the integrate and surround: Native AIs will come at you trying to first integrate and then subsume your control point software. Watch out, particularly for those that generate revenue for a merchant or dramatically change labor cost or availability.

‍

Many founders are trying to shortcut this journey through tech and talent M&A or skunkworks teams. Even these smaller efforts can hold off overfunded AI startups. We’re seeing control point companies with built or acquired AI functionality catch up to third-party AIs. We encountered a recent example in call centers, ground zero for AI adoption. Recently, a call center software company (as the control point) bought a conversational AI company, and the word on the street is that they are starting to displace third-party AI companies, even those with superior technology.

‍

Full Substitution Risk (circa summer 2024)

The longer-term risk is full substitution, where companies go from buying software for their employees to outsourcing the function completely to AI-powered service providers. In this scenario, these AI-powered service providers may be able to price relative to the cost of labor and, as a result, have superior value capture to that of a software vendor. The modifier “full” is probably misleading in that if a competitor can substitute a significant chunk of the labor that changes the value equation altogether or delivers revenues, that’s a big threat. And we’re seeing some hybrid software + AI service models gaining traction.

‍

What Can We Learn from Our PE Brethren?

Admittedly, the idea of full substitution is in its infancy, but we’re seeing the most organized and aggressive push within the operating and acquisition teams of private equity (“PE”) firms or their roll-up companies. PE firms are less focused on sparkling technical capabilities and more on taking out costs. AI is great, but EBITDA is better! Perhaps this is not surprising, as PE firms have always been cost optimizers and catalysts for change. And when PE comes for you, they come in force with their billions of capital and legions of operating partners.

‍

To that end, I’ve spent time with PE owners who are working through this opportunity in real-time. Not just your typical finance jockeys, but finance jockeys who are also deep tech wonks. It’s early, but I would characterize the PE approach as follows:

‍

1. Focus on manual services that can be digitally delivered. Common use cases include data transformation or audio transcription.
2. Deliver in a labor-constrained market, such as accounting, where labor availability impedes growth and service delivery.
3. Assess which tasks can be automated (more on this in a second).
4. Map people to the tasks. This will help you understand the cost-reduction opportunity and the interdependencies of workflows.
5. Manage to the technology's capabilities. The PE investors I spoke with couldn't care less about AI—they only cared about cost reduction. Their perspective was that the biggest opportunity was in information ingesting, mapping (showing the relationship between documents and data points), and retrieval. Almost all AI deployments have humans in the loop and can direct humans to the source data for easy quality control. Fully automated analysis was too high stakes for the current state of technology, especially in professional service fields.

‍

These early attempts by PE hint at broader principles around AI deployment. Ideal markets have labor constraints, utilize digital delivery of their goods, and are lower stakes, so the occasional hallucination isn’t a big deal. Beyond that, we have more detail on areas where AI can be successful: 

‍

• Isolated workflows: Look for workflows that don't need to rely on lots of cross-functional collaboration.
• Narrow data: Most AI-powered services use both an LLM and industry or company-specific data. If that data resides in multiple systems or even companies, an LLM-powered AI might be fundamentally disadvantaged. If that data is all in Joey’s head, the guy who has run the company’s manufacturing process since the 1970s, forget about it. The narrower the scope of data, the easier for a software startup to neatly take a job over.  
• Human in the Loop

‍

Evolution is Coming

Like it or not, the AI meteor has hit Earth. We’re seeing native AI startups gain share and dollars. Native AI startups will be knocking on your door (and we’ll be backing the ones that fit our framework). Vertical SaaS control point CEOs need to be prepared.

‍

The good news is that as the control point, you have an unfair right to cross-sell AI services or at least facilitate AI services and benefit because of your data gravity and workflow gravity. 

‍

Even the early substitution approaches that we highlight above offer opportunities to the paranoid, nimble, and astute. PE firms are an opportunity, not a threat. If you match tasks to people, you are the tool to bring AI into the world. If you own the workflow, you can also ensure you are the human in the loop.

‍

Beyond that, look to facilitate AI and multi-homing. If AI is going to transform your customers, you can provide the AI and price to value. Go own the AI spoils! For the service areas you think you absolutely need to own, get over your software purity sensibilities and be willing to have a separate service offering. While building this, you can facilitate third-party AI offerings, but make sure that your customers can multi-home and price compete. Ideally, you can reduce switching costs so you can insert your offering when you’re ready. Doing all of this will let you fend off attempts by the native AIs to own crucial parts of a merchant’s workflow and data. 

Yes, the world is coming for you, but this can be an enormous opportunity. In a world where labor substitution is possible, no one is better positioned than control point companies to capture that market. If you’re looking for a capital partner to help you on that journey, give us a call or drop us a note at knowledge@tidemarkcap.com!