What Is CAP, and Why Investors Should Care

This is basically what Michael Mauboussin - adjunct professor of finance at Columbia Business School since 1993 - called the Competitive Advantage Period (CAP), the length of time a company can earn returns above its cost of capital. Not just how much value you create, but how long you can sustain it. That framing feels extremely relevant right now, especially as AI reshapes what software even is.

The Shrinking CAP in the AI Era

The more interesting question is what happens to CAP when building software becomes dramatically easier?

Traditional SaaS had very clear drivers of durability. High switching costs from data and integrations. Workflow lock-in. Distribution embedded inside enterprises. High net revenue retention that made cohorts compound over time. That’s why elite SaaS companies historically commanded huge multiples, markets assumed their advantages would last 15–20 years.

AI changes that assumption.

Model capability is rapidly commoditizing. Interface layers are getting bypassed by agents that execute work directly. Competitive entry is faster than ever, small teams can now build products in months that previously required years. When the cost of building software collapses, the duration of advantage naturally shrinks.

In that world, pure application software starts to look less like a defensible product and more like a service. Easy to start, harder to scale, low need for outside capital, and often limited long-term defensibility. That’s a fundamental shift.

From Software to Systems

If software becomes easier to replicate, then the only way to preserve margins and pricing power is to build real switching costs. Network effects, brand, monopoly-like positioning, or deep ownership of critical workflows become the only reliable sources of durability.

We’re already seeing a shift toward hybrid models, software bundled with services, hardware, or proprietary data. Businesses that absorb operational complexity or risk (infra, security, fintech, compliance-heavy systems) tend to persist because they solve messy real-world problems rather than just providing tools. These “tough” businesses are harder to displace because they sit directly in the operational layer of organizations.

At the same time, it feels dangerous to build companies assuming foundation models will stop improving. The safer strategy is to align with them, treat models as infrastructure and build value on top of their strengths rather than betting on their limitations.

This is why I increasingly believe the biggest opportunity in AI is not horizontal tools, but deep vertical AI.

If CAP is about durability, the obvious place to look is industries with high friction: healthcare, government, insurance, legal, logistics, and other operationally complex domains. These markets move more slowly, have higher switching costs, generate proprietary data, and require deep domain expertise. They also involve real change management, not just software adoption.

Horizontal AI products face competition from every angle. But systems embedded into domain-specific workflows (especially in regulated environments) where the insight is known just by a handful of people, take years to replicate as most founders avoid or miss those categories. That creates longer competitive advantage periods.

Going extremely niche can, counterintuitively, increase defensibility. Even smaller markets can sustain durable businesses when they control critical workflows and create high switching costs. In emerging software categories, depth of ownership often matters more than broad reach, today’s seemingly small or non-existent markets can ultimately evolve into massive ones.

Workflow Ownership, Data, and Orchestration

I also think the future of SaaS, both horizontal and vertical, is about building hyper-personalized, last-mile solutions on the fly. Software is becoming much more adaptive and responsive to real-world needs.

I’ve seen this play out in shipping and logistics platforms like Flexport, which were able to roll out tariff refund calculators almost overnight, likely pushed directly by the CEO or founder and shipped within minutes of a new regulation. That level of speed and adaptability is becoming the new standard. The companies that win won’t just ship features; they’ll continuously generate tailored solutions tightly integrated with customer workflows. That kind of responsiveness deepens switching costs and embeds software directly into labor operations.

The nature of lock-in itself is also changing. In the AI era, the strongest position isn’t necessarily the best UI or feature set, it’s controlling the layer that decides how work gets done. If your product becomes the agent or orchestration layer that coordinates tasks, calls tools, and manages workflows, switching becomes extremely costly.

Below that sits proprietary operational data and feedback loops, where usage improves the system in ways competitors can’t easily replicate. Meanwhile, UI differentiation on top of commodity models is fragile, and simple API wrappers have almost no defensibility.

The hierarchy of moats is being rewritten.

All of this points to what may be the biggest wealth creation opportunity in recent memory: connecting foundation models to high-value, complex real-world problems. But it won’t look like the last SaaS or cloud wave. It will likely be more vertical, more operational, more hybrid, and more opinionated. And sometimes, much closer to the metal too.

We got hundreds of SaaS unicorns over the last two decades. It’s reasonable to think we’ll see hundreds of vertical AI ones next, companies that embed intelligence directly into industry workflows and build durable advantages through data, switching costs, and domain expertise.

The most useful question for founders and investors now is: What would need to be true for this company’s advantage to still exist in ten years?

Does it accumulate proprietary data? Does usage make the system better? Are switching costs structural or superficial? Is the industry stable enough to support long-term advantage? Do returns on investment expand or compress over time?

If those answers aren’t clear, the competitive advantage period (CAP) is probably short, and valuation should reflect that.

AI is making software easier to build than ever. That means durability increasingly becomes the only thing that matters. And durability, at least for now, seems far more likely to live in vertical depth than horizontal breadth.