Patterns that signal long-term market fit

This week, we explore the patterns that signal long-term market fit.

If you’re looking to build real conviction around which companies will sustain product market fit, I’d like to invite you into House of Arāya - the home for angel investors. 

One of the questions I come back to most often when reviewing an early-stage company is not whether they have product market fit (PMF), but whether what they have will still be there in three years. Those are very different questions, and in my experience, most decks only answer the first one. 

The companies I have watched most carefully have taught me that fit is not a milestone you reach and then build from. It is a condition that requires ongoing structural support. And in the current AI landscape, that condition is being stress-tested faster than ever before. The pace of change means that what constitutes fit for an AI company today may look very different in three to four months.

A competitor launches, a foundational model improves, user behaviour shifts, and suddenly the problem a product was solving is either no longer a problem or is being solved in a completely different way. For founders building in this space, and for investors evaluating them, that compression of the cycle is one of the most important things to get comfortable with. 

Before reading further, pause and answer this:

The clearest illustration of this I have encountered in the market is Chegg, an AI-powered educational technology company designed for students to improve learning efficiency. Although I haven’t invested in Chegg, for years the business had what looked like genuine, durable fit. Students subscribed, engaged, and returned. The core growth loop was a content engine: curated human answers distributed through SEO, which attracted more subscribers, which funded more content.

It worked until ChatGPT launched and students could get an instant, personalised answer to any homework question at no cost. The loop did not slow. It broke. In January 2024, the company was valued at 1.2 billion dollars. By October, nine months later, that figure had fallen to 150 million dollars.  

Stack Overflow followed the same logic. Its loop was built on developers asking questions and answering questions for social capital, distributed through search. When GitHub Copilot launched, developers could get answers instantly, directly inside their coding environment. The incentive to post on Stack Overflow eroded, fewer questions got answered and engagement declined across the platform. 

Both cases reveal the same underlying pattern. Product market fit built on a growth loop that a faster, cheaper, more embedded alternative can short-circuit is structurally fragile. The question worth asking of any portfolio company is not whether the loop is working today. It is what would break it. 

For a long time, product market fit meant building something people wanted and then scaling it for years. Traditional SaaS companies could find fit and largely hold it, with growth following a relatively stable and predictable curve. That assumption no longer holds for AI companies. 

What the data shows is that AI startups experience PMF very differently. Confidence in fit tends to spike early, often driven by the novelty of the technology, and then pulls back sharply as the market catches up, competitors emerge, and foundational models improve. The companies that recover and build something durable are the ones that treat that pullback not as a failure but as a signal to re-earn their fit at a deeper level. 

This means that for AI companies, fit is less a destination and more a practice. The companies worth backing are not necessarily the ones who found it first. They are the ones who are structurally positioned to keep finding it, and who have the team velocity and proprietary data to do so faster than anyone else. 

The first thing I examine is where the product sits in the user's working life. Products at the centre of how users create and operate are far more defensible than those that assist from the periphery, and that distinction becomes decisive when a new tool enters the market. 

The second is whether the company owns data that cannot be replicated. What remains defensible is private, contextual, structured data that foundational models cannot access. The question I ask directly: what does this business know that nobody else can get to? 

The third is growth loop integrity.  Loops anchored in direct customer relationships, embedded workflows, and proprietary data are structurally more resistant than those dependent on SEO distribution or user-generated content. The Chegg and Stack Overflow cases are not outliers. They are a preview of what happens to any loop whose core behaviour, AI can perform faster and cheaper. 

Take a company you are currently reviewing and map its core growth loop. Identify the single behaviour that holds it together, then ask what would happen to that loop if a faster, cheaper, zero-friction alternative addressed the same use case directly inside the user's existing workflow. If the loop would break, the fit is conditional. Then, put three questions directly to the founder:  

Their answers will tell you more than anything in the deck. 

Inside House of Arāya, we work through this in practice. Members use our Market Fit Radar to break down real companies, stress-test their growth loops, and understand where fit is durable versus where it’s likely to fail under pressure.

We built it because this is one of the hardest judgment calls in early-stage investing, and one of the most expensive to get wrong.

We’re opening on 23rd March.

If you want to develop the ability to spot durable fit before it becomes obvious, join the waitlist.

Warmly

Rupa Popat with Team House of Arāya