Which came first: the product or the pricing model?

Ask a company which came first, its product or its pricing model, and the question will sound almost naïve. The product, naturally. One builds something of value and then decides what to ask for it.

Razors and blades

History appears to agree. Even the most celebrated pricing model of all followed this order. Gillette's cheap razor and expensive blades have become the textbook illustration of a business built around its pricing logic, yet the logic was not there at the beginning. As the legal scholar Randal Picker has shown, Gillette did not sell its razor cheaply while its patents held. The model now credited to King Gillette's foresight emerged only later, once the patents expired and the market around him changed. The product came first; the pricing model followed, shaped less by design than by circumstance.

This is what pricing by default looks like. It is rarely a poor decision. More often it is no decision at all, merely an inheritance from the product, from the industry, from whatever seemed customary at the time.

When pricing leads

The more instructive stories are those in which the order is reversed. When Rolls-Royce began charging airlines for each hour its engines flew, rather than for the engines and the parts to repair them, it changed far more than an invoice. It changed what the company was paid for. Under the old arrangement, a failing engine brought revenue; under Power by the Hour, it brought loss. Reliability, monitoring and maintenance moved from the margins of the business to its centre, and the engineering followed where the pricing led.

Michelin met the same logic when it offered truck fleets tyres by the kilometre rather than by the unit. One cannot bill for distance one cannot count, and so the pricing model called for telematics, data and services that a tyre maker had never needed. Hilti, in moving from selling tools to managing a customer's entire fleet of them for a monthly fee, had to learn tracking, replacement logistics and an altogether different kind of relationship. In each case, the pricing model was not a label attached to a finished product. It was a specification the product had to meet.

Built to be measured

Beneath these stories lies a principle I return to often:

A product that does not record usage cannot be priced by usage. A product that does not observe outcomes cannot be priced by outcome. A product whose features are tangled together resists any clean distinction between tiers. The pricing models open to a company tomorrow are quietly narrowed by product decisions taken today, frequently by people who would never describe themselves as having made a pricing decision.

The question of which came first, then, is better left unanswered. Product and pricing model shape one another, and the companies that treat them as a single design problem preserve choices that others surrender without ever noticing.

A forgiving medium

Why did pricing by default serve so many companies for so long? In part because its mistakes were affordable, and nowhere more so than in software. With marginal costs close to zero and value rising with every additional user, almost any sensible model could be made to work. A flat subscription absorbed its heaviest users without complaint. A price per seat grew gracefully as the customer grew. And a model that proved wrong could be corrected quietly at the next renewal. Software was a forgiving medium.

AI removes the cushion

Products built on artificial intelligence are far less forgiving, and this is why I believe the pricing model matters even more for them. The logic has not changed. What has disappeared is the cushion that once softened its consequences.

The first thing to return is cost. Every query, every generated answer, every task an agent completes consumes computing power. An AI product's most devoted users may also be its most expensive, and an unlimited flat-rate plan becomes a liability without a ceiling. More than one AI tool has already had to retreat from generous terms, and customers seldom forgive a pricing model that shifts beneath them.

The second is the quiet separation of value from headcount. Pricing per seat made sense when value grew with the number of people using a tool. When an agent performs the work of several, the customer needs fewer seats while receiving more value, and a seat-based price contracts at precisely the moment the product succeeds. Hence the many experiments with pricing per task, per resolution or per outcome, Intercom's Fin at $0.99 per resolved conversation among the best known. Each returns us to the principle of measurement: to charge for a resolution, the product must first be able to define one, recognise it and prove it.

The third is a peculiar instability. The cost of running AI models keeps falling and competitors keep lowering their prices, so any price point set today may look strange within a year. The pricing model, however, endures. It is what customers anchor on, what procurement writes into contracts, what sales teams learn to explain. A price can be changed at the cost of some grumbling; a model must be renegotiated one relationship at a time. When the levels are in constant motion, the architecture beneath them bears more weight, not less.

The last is uncertainty on the buyer's side. Many customers cannot yet say what an AI product will do for them. Will it resolve a third of their support requests, or two thirds? Will it save an hour each day, or ten minutes? Where value is this uncertain, the pricing model determines who carries the risk, and that often determines whether a hesitant buyer agrees at all.