How do you measure the ROI of AI training in a company?

Task by task. Name the processes the training targets, measure how long they take before it, measure the same processes 60 to 90 days after, and price the difference at the loaded hourly cost of the people doing the work. Then subtract the cost of the training, net of the FUNDAE credit. That is a number a CFO can audit.

What makes the number credible

  • A baseline measured before the training, not reconstructed from memory afterwards.
  • Hours attributed to a named process with an owner, never to a department.
  • The same measurement, taken the same way, by the same people, before and after.
  • The loaded hourly cost — salary plus employer contributions and overhead — not the gross salary.
  • The FUNDAE credit counted once, as a reduction of the cost. It is not a saving.

The alternatives, and what each one buys you

Four ways companies answer this in practice, from the cheapest to the only one that survives a finance review:

  • Satisfaction scores and completion rates. Free, immediate, and a measure of attendance rather than of work.
  • Self-reported time saved. Fast to collect and impossible to audit; useful as a signal, not as a number.
  • Tool telemetry. Real usage data, but usage is not value — a licence that gets opened every day can still change nothing.
  • Task-level before and after. Slower to set up, and the one a CFO will accept, because every figure in it traces back to a process.

Where the number breaks

  • Hours freed that nobody reassigns. The time exists; the saving does not.
  • A pilot team that volunteered is not the company. Enthusiasts move faster than the average of a department.
  • Quality that drops while time falls, paying for the gain somewhere the measurement cannot see.
  • The same hour counted twice, once in each of two processes that share it.
  • Annualising one month. Adoption curves are not straight, and seasonality is not a return.

The evidence to keep

Dated, in units, and per process. That is the difference between an answer that can be quoted and one that merely sounds right.

  • The list of processes, with an owner for each.
  • The before and after measurements, with the date each was taken and by whom.
  • Everything else that changed in the same window — a new tool, a reorganisation, a hiring freeze.
  • The FUNDAE documentation for the credit actually claimed.

Next step

Put your own headcount, hours and cost into the FUNDAE calculator on the home page to get an order of magnitude. If you would rather start from the processes, send us the three you would measure and we will tell you what a defensible baseline for them looks like.

Related

Success cases

What the measurement looked like in companies that ran it — scope, format and what each one took away.

See the cases

Talk to us

Bring the three processes you would measure first. We will tell you whether the baseline is realistic before anyone signs anything.

Talk to us