“What’s the ROI?” is the question every AI project must answer — and the ones that can’t, don’t get scaled. Here’s how to measure AI ROI properly in 2026. (dgm implements osFoundry as an independent partner.)
The basic equation
AI ROI is simply:
(Benefit − Total cost) ÷ Total cost, measured against a baseline.
Sounds obvious — but most ROI claims fall down on two things: a missing baseline, and an undercounted cost.
Set the baseline first
You cannot measure improvement without a “before”. Before the pilot, capture the current state of the process:
- how long it takes (hours/week);
- what it costs (per transaction, per case); and
- the error or rework rate.
Without this, any ROI number is guesswork — and unconvincing to a board or funder.
Pick concrete metrics
Choose one or two specific metrics per use case, not a vague “productivity improved”:
| Use case | Good metric |
|---|---|
| Document processing | Hours saved/week; cost per document |
| Customer enquiries | Response time; cost per query; deflection rate |
| Quality/inspection | Error/defect rate; rework cost |
| Sales/marketing | Conversion; revenue per rep |
Specific, measurable metrics are what prove value.
Count all the costs
This is where ROI is most often overstated. Include:
- platform usage costs;
- integration work;
- data preparation;
- human oversight (time spent reviewing AI outputs); and
- change management and training.
The licence is often the smallest part — counting only software cost flatters the result.
Then decide
A pilot that demonstrably saved hours or cut a cost against its baseline, net of all costs, is the case for scaling — internally, or for finance and grants. One that didn’t is a cheap lesson, which is exactly why you piloted small.
Where osFoundry and dgm fit
dgm scopes pilots around a single measurable outcome with a baseline captured up front, so ROI is provable, not assumed. We implement on osFoundry, whose usage pricing (cents per query, no per-seat fees) makes the cost side transparent and easy to track — so the ROI maths is honest on both sides. Data control via self-hosting or an EU region (osFoundry publishes US/EU/JP regions, not a UK one).
dgm is an independent integration partner with zero integrations so far. To scope a pilot with measurable ROI, book a consultation with dgm. General information, not financial advice.