Per-seat pricing is dying more slowly than advertised

The best available evidence about how software pricing is changing is not a survey of chief information officers. It is a price list, published, dated, and binding on the vendor that wrote it.

Take Datadog’s. The company charges by host for infrastructure monitoring and application performance monitoring, by ingested or scanned gigabyte for log ingestion, and by million events for indexing and storage. Those are four different units of consumption in one product family, and none of them is a person.

Now look at what the same page does with commitment. Infrastructure monitoring at the Pro tier is $15 per host per month billed annually, or $18 on demand. Application performance monitoring, taken alongside infrastructure monitoring, is $31 billed annually and $48 on demand. Standard log indexing is $1.70 per million events annually and $2.55 on demand.

The premium for not committing runs at about a fifth on hosts and at half or more on traces and log events. That gap is the whole argument in miniature. Hosts are a quantity a customer can forecast reasonably well, because servers are provisioned deliberately. Traces and log lines are emitted by software behaving in ways nobody planned, and they are the units where the vendor charges most heavily for the customer’s inability to predict them.

Consumption pricing is not primarily a shift from people to usage. It is a transfer of forecasting risk, priced explicitly.

The commitment did not go away

The story told about consumption pricing is that it replaces the multi-year contract with something fairer, where a customer pays for what they use and nothing more. The filings of the largest pure consumption vendor say something different.

Snowflake’s fiscal year 2026 closed on 31 January 2026. Product revenue for the year was $4,472.3m, up 29%, with fourth quarter product revenue of $1,226.6m, up 30%. Net revenue retention was 125%, and 733 customers had trailing twelve month product revenue above $1m, a rise of 27%.

Then the number that matters for anyone negotiating one of these contracts. Remaining performance obligations, the contracted revenue not yet recognised, stood at $9.77bn, up 42%.

Contracted backlog growing at 42% while the revenue actually consumed grows at 29% is not what a pay-as-you-go market looks like. It is what a market looks like in which customers sign multi-year capacity commitments in advance and then draw against them. The company describes the mechanism itself: product revenue derives primarily from consumption of compute, storage and data transfer, customers may consume more than their contracted capacity during the term, and they may have the ability to roll over unused capacity to future periods.

Read that as a buyer. You commit to a quantity of future consumption, at a discount, over several years. You may exceed it, which is billed. You may undershoot it, in which case rollover, if you negotiated it, is the difference between a soft landing and a write-off.

That is a two-part tariff: a commitment plus a meter. Two-part tariffs are not new, and neither is the customer problem they create, which is that the discount is earned at the moment of maximum ignorance about future demand.

Per-seat did not die either

Meanwhile the seat continues to earn its living, and the evidence for that is a price list rather than a filing. Microsoft’s UK business plans are quoted per user per month paid yearly, at £5.40 for Business Basic, £18.10 for Business Standard with Copilot and £24.60 for Business Premium with Copilot. The unit is a person, the commitment is annual, and no part of the price depends on how much that person does.

It is worth naming a number that does not evidence this, because it gets used for the purpose constantly. Sage reported subscription penetration of 83% for the year ended 30 September 2025 on underlying revenue of £2,513m, and its own definition of that measure is underlying software subscription revenue as a percentage of underlying total revenue. That tracks how far the perpetual licence has retreated, which is a different transition from the one this piece is about. It says nothing about seats versus meters.

There is a reason categories like accounting, payroll and office productivity have not moved. Seat pricing works where the value of the software scales with the number of people doing a job, where the buyer already forecasts that number for other purposes, and where usage per person is bounded by the length of a working day. Payroll, accounting, human resources, customer relationship management and service desks all satisfy those conditions. Data platforms, observability, communications interfaces, content delivery and, increasingly, model inference do not.

So the honest description of the market is not that one model is replacing another. It is that the industry has sorted itself by whether the value driver is a person or a machine, and priced accordingly. Where a vendor has moved a person-shaped product onto a meter, the usual explanation is that the seat count had stopped growing.

The forecast has no owner

Under seat licensing, the person who could increase next year’s bill was a hiring manager, and the forecast lived in a workforce plan that the finance function already produced and challenged.

Under consumption, the person who can increase next year’s bill is an engineer changing a log level, adding a dimension to a metric, scheduling a backfill, or deploying a retry loop that behaves badly against a rate limit. There is no plan for that, because no organisation has ever produced a twelve-month forecast of its own software’s verbosity.

This is the operational heart of the change and it is consistently underestimated at the point of purchase. The classic failure sequence runs like this. A team commits to a discounted capacity tier based on current usage plus a growth assumption. Consumption grows faster than the assumption, because the platform gets adopted by teams who were not in the room. The overage is billed at list. The renewal then re-bases the commitment at the higher level, which locks in a peak that included a quarter of accidental usage nobody has gone back to remove.

The commercial defence is not clever negotiation. It is instrumentation and ownership: a named person accountable for consumption on each account, per-team quotas that fail loudly rather than silently billing, an alert on rate of change rather than on absolute spend, and a monthly review that treats an unexplained rise as an incident rather than as growth.

The contractual defence is a short list of terms, and they are worth more than the headline discount. Whether unused capacity rolls over and for how long. Whether overage is billed at list or at the committed rate. What happens to the commitment if the organisation divests a business unit. Whether the vendor can change the definition of a billable unit mid-term, which is the clause that has caused the most disputes in observability and is almost never negotiated.

The same shift is arriving in automation, with an extra variable

Metered pricing has spread into automation and model-assisted work, and it brings a complication the infrastructure world did not have. A host is a host and a gigabyte is a gigabyte. A unit of model consumption depends on how a vendor chose to package the work, and that packaging can change between releases without the customer’s workload changing at all.

An organisation that automated a process last year and finds the running cost has moved this year may be looking at its own volume growth, or at a repricing, or at a model change that altered how much work each request represents. Distinguishing between those requires usage data at a granularity most buyers never asked for. That is one of several reasons the second-year bill for automation so rarely matches the business case, and the arithmetic behind it is set out at intelligent automation.

What cannot be measured, and why

Anyone who tells you what proportion of British enterprise software contracts have moved to consumption pricing is estimating, and the estimate cannot be checked.

There is no register of enterprise software contract terms in the United Kingdom. Private sector agreements are covered by confidentiality provisions that survive the contract. Vendor disclosures aggregate at a level that deliberately obscures the mix between committed and consumed revenue, and the metrics that would settle it, such as the ratio of overage to commitment, are not reported by anybody.

Three sources of real evidence do exist, and each has a known bias. Published price lists, like the two cited above, show the structure of an offer but not what was actually agreed, and the negotiated position on a large account can differ from list by a wide margin. Listed vendors’ filings show the aggregate shape of contracts, as with the backlog figure above, but they mix geographies and customer sizes. Public sector award notices and framework price lists show real agreed terms, at real prices, on real British contracts, which is more than the private sector offers, but they describe buyers whose procurement rules and volumes are unlike a mid-sized company’s.

Weighted together, those support a directional claim and not a percentage. The direction is that new categories are priced by machine consumption, that established categories remain priced by people, and that the commitment persists in both.

The negotiating implication

The interesting consequence of all this is that the discount conversation has quietly become a forecasting conversation, and most buying teams have not staffed for it.

A procurement function can negotiate a percentage off a list price. It cannot, on its own, decide whether to commit to three years of capacity at a level the engineering organisation has not modelled, because that decision requires a view about growth in workload, a view about how much current consumption is waste, and a willingness to be wrong in public. Where those views do not exist, the organisation defaults to whatever the vendor’s account team proposed, which is reliably the larger number.

The organisations doing this well have moved the commitment decision out of procurement and into a joint forum with engineering and finance, and they treat the annual commitment as a capital-style decision with an explicit assumption register. That is a modest change in governance and it is worth considerably more than the two points of discount everyone spends the negotiation arguing about.

The team that usually ends up owning consumption in practice, and the reasons it is rarely equipped to, is the subject of the shape of the platform team. How the underlying purchase gets made at all is covered in how UK enterprises actually buy software, and the rest of this desk’s work sits under SaaS and enterprise software.