Chegg lost roughly 99% of its value after free AI arrived to do what students had been paying it for. Quentic sold compliance software (the kind of business a glance dismisses as commodity SaaS) and the same wave barely touched it.
The difference is structural: four layers, each one a reason the customer still has to buy rather than build it or do without it.
The other half is time: how long that structure holds.
Chegg: General answers, free elsewhere
This layer is worth something only when the knowledge is proprietary and maintained against a moving target. Chegg's answers were general and static: exactly what a language model now generates for free.
Quentic: Regulation encoded, and kept current
To use an AI's answer safely you have to be able to tell whether it's both correct and complete. In a specialist domain the customer usually can't. And if they could, they wouldn't have needed to ask. Picture an industrial site near Paris storing a given quantity of hazardous chemicals: an AI produces a confident recommendation in seconds, but the operator has to be certain it's compliant with both the EU Seveso III Directive and the relevant French rules, and typically has no way to verify that. Quentic's durability was the regulatory logic encoded into workflows and kept current across jurisdictions. Knowing EHS was never enough on its own. Whether that ran on Java or PHP was irrelevant.
Chegg: Nothing to stand behind
No liability anyone needed transferred, nothing to stand behind.
Quentic: The name on the audit
In EHS, being wrong is expensive: fines, shutdowns, real safety exposure. So customers had little appetite to insource that risk. Being the system of record meant a customer could walk into an audit and pass it easily, the documentation and trail already in defensible form. An LLM won't stand in front of the regulator on their behalf.
Chegg: A library, not an asset
The answer library looked like accumulated data, but it wasn't proprietary or load-bearing. An LLM produces equivalent answers without it.
Quentic: Real friction, but exportable
Years of incident records, audit trails and compliance history raised the cost of leaving, but customers had the right to export all of their own data, which deliberately capped how much lock-in it created. Friction, not a wall.
Chegg: A channel that closed
The funnel came from students typing questions into Google and clicking through. When Google began answering those questions itself, the channel closed at the source.
Quentic: Strong at home, ceded elsewhere
Quentic held this layer, but only within its borders. In Central Europe the footprint was strong enough to be self-reinforcing. Elsewhere, there were whole industries we never entered, because an established competitor already owned the attention there — the experts, the relationships, the standing. In those verticals, deals closed before we even knew they existed. Not a technology gap we could engineer past; a distribution-and-authority gap, and close to unassailable.
The layers of the stack don't move. Regulation doesn't retreat, encoded knowledge doesn't decay, first-party data keeps compounding. What moves is everything outside them: how much AI can already do on its own, and therefore how many customers still have to buy.
So picture the market as a pyramid. The most demanding customers sit at the narrow top; the least demanding at the wide base. Now draw a waterline across it. Below the line, the need collapses: AI does the job directly and nobody has to buy anything. Above it, the complexity still justifies buying.
That's the erosion people expect. There's a second one, running the other way. The most demanding customers are also usually the largest and the most expert: some understand the domain better than any vendor serving them. They were never buying knowledge, only the capacity to turn it into software, and that capacity is what AI hands them. They don't leave because their needs got simple. They leave because they can finally have exactly what they want instead of a product built for everyone.
At the apex: customers who know the domain better than you, and build it themselves.
In the middle: demanding enough to need you, not sovereign enough to replace you.
At the base: needs simple enough for AI to absorb them.
Neither boundary is fixed. AI lifts the waterline from below and lowers the cost of building from above. The layers themselves don't weaken. The set of customers they protect narrows from both ends, and what's left is the middle: demanding enough to need you, not sovereign enough to replace you.
“The stack gives you altitude. The pyramid gives you rate. You need both to answer the only question an owner has: how many years are left, and which layer goes first.”
Chegg: The base was the whole market
Chegg's customers sat at the base of its own pyramid: students with routine questions, the widest and least demanding layer in that market. The waterline didn't cut through Chegg's category. It rose past the whole thing in one step.
Quentic: Clear of both lines
Quentic's customers sat in the middle: industrial operators tracking incidents and audits across multiple sites and jurisdictions, where being wrong means fines, shutdowns, or someone getting hurt. Demanding enough that AI alone wouldn't do it, and none of them held the cross-jurisdiction regulatory picture Quentic had encoded. That's the zone that holds.
“Which of your customers could do without you in eighteen months, and what share of revenue sits with them?”
Then ask it again about the market beyond today's customers. That answer is the one that moves the multiple.