Why software pricing is moving from seats to usage, and what it does to predictability

What a price based on consumption changes

A seat price is a bet on headcount. If a company hires ten people, the bill rises by ten seats whether or not those people use the product. If it cuts ten people, the bill falls.

A usage price is a bet on activity. The bill rises when the product does more work for the customer and falls when it does less, regardless of how many people are on the account.

Deloitte's 2026 technology predictions describe why this matters now. AI agents could conceivably give one user the power of many users and reduce the number of seats a customer needs, which could affect a vendor's revenue. Usage pricing, Deloitte notes, is often attractive because it is quantifiable and therefore auditable.

For the buyer, the trade is cost control against cost certainty. Usage pricing can be cheaper in a slow month. It can also produce a larger bill than expected in a busy one.

What Snowflake's annual report shows about consumption revenue

Snowflake, a data platform sold on consumption, is one example. Its fiscal 2026 annual report (year ended Jan. 31, 2026) says customers use compute, storage and data transfer, and that the company generally recognises revenue as that consumption happens, not evenly across the contract term.

Customers buy in one of two ways. Under capacity arrangements they commit to a set amount of consumption at specified prices. Under on-demand arrangements they are billed monthly in arrears. Capacity contracts typically run one to four years, and the weighted-average term for contracts signed in fiscal 2026 was about 3.1 years.

On-demand is a small and shrinking part of the business. It was about 1% of revenue in fiscal 2026, against 2% in fiscal 2025 and 3% in fiscal 2024. Product revenue was $4,472.3 million in fiscal 2026, up 29% from $3,462.4 million a year earlier, and the net revenue retention rate was 125% as of Jan. 31, 2026.

The filing says product revenue rose $1.0 billion primarily because existing customers consumed more of the platform, as evidenced by that rate.

Why a contract is not a forecast

Snowflake reported remaining performance obligations of about $9.8 billion ($9,771.5 million) at fiscal year-end. Roughly 46% of that was expected to be recognised as revenue in the 12 months ending Jan. 31, 2027. That works out to about $4.5 billion.

The figure measures non-cancelable contracted amounts that will be invoiced and recognised as revenue in future periods. Because revenue follows consumption, the timing of any customer's spend depends on when that customer actually uses the platform. The filing includes a risk factor titled "We may not have visibility into our future financial position and results of operations".

It says consumption can fall short of forecasts, including in response to new software releases or hardware improvements that may make the platform more efficient, adverse macroeconomic conditions, or holidays. It also says customers have shortened committed contract durations as they rationalise budgets and prioritise cash flow management, and that sequential growth in remaining performance obligations has historically been highest in the fourth fiscal quarter.

The arithmetic shows the gap. A subscription priced at $100,000 a year books $25,000 a quarter. A consumption commitment of the same size can book far more in one quarter and far less in the next, while the contract value stays the same.

A subscription priced at $100,000 a year books $25,000 a quarter, while a consumption commitment of the same size can book far more in one quarter and far less in the next.

How vendors cushion the swings: commitments, floors and seats

Snowflake's customers typically consume the platform under capacity arrangements, in which they commit to a certain amount of consumption at specified prices.

Salesforce's Agentforce pricing runs a meter beside user licences. Its May 15, 2025 announcement priced Flex Credits at $500 per 100,000 credits. Each agent action uses 20 credits, or $0.10.

At those rates, 10,000 actions in a month cost $1,000, and a customer that triples volume triples the bill. The same announcement described per-user Agentforce licences with unmetered employee-facing use, priced per user per month. Pricing for Agentforce for Sales, Service, and Industries was to be announced at general availability.

A third piece, the Flex Agreement, lets a customer convert user licences into Flex Credits or the reverse as priorities change. Salesforce's May 15, 2025 announcement therefore describes a seat model and a usage model, with a way to move budget between them.

How far the shift is expected to go, and where the evidence is thin

Deloitte cites a Gartner forecast, dated July 16, 2025, that by 2030 at least 40% of enterprise SaaS spend will shift toward usage-, agent- or outcome-based pricing. That leaves up to 60% on other models.

Apart from the Gartner figure it cites, Deloitte gives no date or percentage for the shift. It expects mixes of licences with usage, value or outcome pricing to become standard, but says standard practices could take years to emerge, if they emerge at all. Outcome-based pricing, it says, has "a long way to go" before widespread use, and says agreed definitions of terms such as agent, task and outcome should come first.

For an operator buying software, the practical questions are what the unit is, what the commitment floor is, and what happens above it. Snowflake's filing says unused capacity can often be rolled into a later order, generally when more capacity is bought. For an operator selling software, the filing is a reminder that remaining performance obligations and recognised revenue move on different clocks.