← Back to Blog

Blog post

What a credit actually meters

By Hannah Whitcombe · Commercial lead · 29 July 2026 · 7 min read

There are exactly two ways to pay us, and keeping them apart is the first thing to understand about the commercial model.

The platform is a fixed monthly fee per organisation with a credit allowance inside it, plus credits beyond that allowance if you use more than it covers. Services — converting a workbook or a process into an application — are quoted per engagement and invoiced separately. Services are never paid in credits, and credits never buy consulting time. A pricing page that blurs the two produces a customer who thinks the subscription includes an implementation, which is the single most expensive misunderstanding available in this category.

What a credit is spent on

Three things, and this list is exhaustive rather than illustrative.

  • Storage — what the results you keep occupy in your organisation's own store.
  • Compute — what a job costs to finish, which is a function of how much work the operation does rather than how long you were logged in.
  • Model usage — what an application spends when it asks a language model something. Not every application does; the ones that summarise, classify or narrate are the ones that will.

The allowance is sized so that a typical month fits inside it, which means the bill is flat in practice. The months that are not typical are the interesting ones: a year-end, or the afternoon somebody re-ran the whole prior year to answer an auditor's question. Both of those show up as usage before they show up as an invoice.

Credits are prepaid, in dollars, and they expire

Credits are bought in advance, in US dollars, and purchased credits expire twelve months after they are bought. That is stated plainly rather than buried, because an expiry discovered at renewal is a bad conversation and an expiry known at purchase is just a fact about the product.

Promotional credits — the ones that arrive with an offer rather than an invoice — expire faster, in thirty to ninety days. The urgency is the point of a promotion; a promotional credit that never expired would be a list-price discount pretending to be a campaign.

The spend cap is the part to set on day one

Every organisation can set a hard cap: a "never bill past this" switch, with alerts before it. It is the feature most likely to be ignored during a trial and most likely to be the reason somebody trusts the platform in month four.

Set it. Set it deliberately high if you want, but set it. The failure it prevents is not a fraud scenario — it is the ordinary one where a job is left running in a loop by a caller nobody remembers writing, and the first anyone hears of it is an invoice.

There is no per-user price, and there is not going to be one

This is the question we get most often, and the answer is a product fact rather than a positioning choice.

Applications here run programmatically. The valuable thing is a run — an operation submitted, a job finished, a result written — and a run has no seat attached to it. An application called nightly by a scheduler at three in the morning is doing the most valuable work in the account and there is no human logged in for any of it.

So charging per user would meter the wrong thing twice over: it would bill you for people who watch and not for the work that happens, and it would give you a reason to share logins, which is a governance problem we would then have to solve. Billing is at the level tenancy already works at, which is the organisation.

What that means practically: adding the whole finance team to an application costs nothing. Grant access to whoever needs it. What costs is what the applications do, which is the thing you are actually buying.

The figures in this piece — including the ones in the panels — are illustrative. They show what a result of this shape looks like, not a measurement of anything that happened.

Read next

Stories for the same roles