An AI business case is only useful if it survives contact with a finance director. That means an explicit baseline, a range rather than a number, and a cost model that includes the parts vendors leave out.
An AI ROI assessment quantifies what a candidate initiative is worth against today's measured baseline, subtracts full implementation and running costs including review time and ownership, and reports a value range, a payback period and a confidence level. It also states what would have to be true for the case to fail — which is the part that makes it credible.
Why AI ROI figures lose credibility
Three habits do most of the damage: presenting a single number instead of a range, counting hours released as if they were cash, and omitting the running costs of review, monitoring and ownership.
Finance teams discount all three instinctively. Separating hours released, cost avoided and cash effect — and saying which will appear in a budget line — restores the conversation.
The cost side, in full
Discovery and specification time from people with day jobs
Build or configuration, plus integration with systems not designed for it
Testing against real historical cases, including the awkward ones
Training and the temporary productivity dip during adoption
Running cost: model usage, licences, monitoring, and reviewer time
Ownership: someone must notice when it stops working
The value side, expressed honestly
Three value types that finance treats differently
Value type
Example
How finance reads it
Hours released
Fewer minutes per case handled
Real, but not cash until capacity is redeployed or not hired
Cost avoided
Fewer errors, fewer credit notes, less rework
Credible when the historic cost is documented
Cash effect
Working capital released by faster collection
Recognised immediately, and the strongest argument
Payback rather than a headline return
Payback period is more useful than a percentage return for a decision of this size and uncertainty. It answers the question an owner actually asks — when do we get the money back — and it degrades gracefully when the value lands at the low end of the range.
Every case is presented with a low, expected and high scenario, and with the specific assumption that most influences the outcome identified so it can be tested early.
What you receive
Documented baseline for the process in scope
Value model with low, expected and high scenarios
Full cost model, including running and ownership costs
Payback period and sensitivity to the dominant assumption
A business case document structured for finance approval
Business impact
Investment decisions made on evidence rather than vendor projections
Realized value measurable after go-live, including the misses
Faster internal approval because the objections are pre-answered
Weak cases identified before budget is committed
Where this is the wrong fit
Justifying a decision already taken
Processes where no baseline data can be obtained or estimated
Buyers seeking an industry benchmark instead of their own numbers
Frequently asked questions
How do you calculate AI ROI?
Measure the baseline — volume, handling time, loaded cost, error cost — model the change as a range, subtract full implementation and running costs including review time, and report payback with a confidence level.
What is a realistic payback period?
It varies by process and volume. Rather than quoting a benchmark, we model your case and state which assumption the result is most sensitive to.
Do hours saved count as savings?
Only if capacity is redeployed or a hire is avoided. Otherwise they are real operational relief but not a budget line, and the business case should say so.
Can we try the numbers ourselves first?
Yes — the AI ROI calculator applies the same deterministic model to your own inputs before any conversation.