What Does a Workflow Actually Cost? Credits, Metering and Honest ROI

Automation only pays off if you can prove it pays off. The question sounds simple — what does one workflow cost to run, and what is it saving? — and most platforms answer neither half honestly. They either bill you for capacity and call it cost, or they compute a return from numbers nobody entered.

Here is exactly what we measure, how the mechanism works, and where it stops.

Credits are a reservation system, not a meter reading

The obvious way to charge for agentic work is to add it up at the end. We do not, because by then the money is spent.

Instead, when a workflow is submitted, its cost is reserved against the group's credit account before any agent starts, and the balance is checked at that moment. If the group cannot cover the run, the run does not start — so an expensive mistake is prevented rather than explained afterwards.

What happens next is the part worth understanding:

• If the run completes, the reservation settles into a deduction.
• If the run is rejected, cancelled, or never proceeds, the reservation is released and the credits go back.

Reservation, release and deduction are recorded as separate, audited transaction types against the group's account. That is deliberate. A single "amount spent" number cannot tell you the difference between work you paid for and work you held budget for and then abandoned — and two organisations with the same deduction total and very different release rates are running very differently. Only one of them has a problem worth looking at.

Charges a person decided are kept apart from charges the platform metered

There is a fourth transaction type: a manual deduction an administrator applies for value delivered outside the platform — human effort on an engagement, say.

It could have reused the ordinary workflow deduction. It deliberately does not, because the moment those two mix, the spend breakdown stops being able to answer "what did the automation actually cost?" A number that silently blends metered consumption with human judgement is worse than no number, because it looks precise.

Cost, rolled up where you can act on it

For IntegrationHub tenants, cost is tracked as entitlement cost plus transaction cost — what the tenant costs to have, and what it costs to use — per tenant and per day, rolled up by product with drill-down beneath it.

Separating those two is the useful part. A platform owner can see not just what an integration costs but which half of it is the standing charge, which is usually the half you can actually do something about.

ROI needs one number, and it has to come from you

Cost is half an argument. The other half is what the automation replaced — and that is a number the platform genuinely cannot discover. So it asks.

Each workflow carries an estimated manual baseline: how long the process takes a person. Set it, and the ROI view puts credits consumed against hours given back.

We would rather be blunt about the shape of that. The baseline is entered by an administrator, not measured. The ROI figure is exactly as honest as the estimate behind it, and we would rather hand you a number you can interrogate than one that arrived from nowhere. A platform that claimed to derive your manual effort automatically would be guessing, and presenting the guess as a finding.

If you set one thing on the workflows that matter, set this. Cost is measured for you; value needs the one number only you know.

Where the measurement stops

Some things we do not measure, and we would rather say so than let a dashboard imply otherwise.

We do not currently discover idle integrations for you, or produce utilization and zero-activity reporting that would tell you what to retire. Those are real gaps, not features we are being coy about — and until they exist, capacity decisions still need a human looking at the estate.

That is the same rule we apply to every claim on this site: something we cannot demonstrate does not get stated.

Why this shape

Reserve-then-settle costs more to build than a meter. What it buys is the ability to say no before the spend, and an audit trail that distinguishes intent from outcome.

For automation that runs unattended and on a schedule, those are the two properties that decide whether finance ever trusts it.