Why most AI ROI numbers fall apart in the room
The typical AI business case leans on a vendor's headline: 'up to 40% faster.' Finance discounts it to zero on sight, because it is not grounded in your work, your people, or your salaries. A number that cannot be checked is a number that cannot be approved.
A credible case does the opposite. It is conservative, traceable to real workflows, and honest about its own confidence. It is designed to survive scrutiny, not to impress.
The pieces of a defensible case
Without publishing a proprietary formula, the logic of a sound estimate is straightforward. Each input pulls the number down toward what the business will actually keep.
- Productive hours unlocked: the hours AI can realistically save per person per week, grounded in their actual workflows, not a generic benchmark.
- Adoption readiness: not everyone will adopt; readiness rises with fluency, so a less-ready team keeps less of the theoretical gain.
- Realisation factor: verification, rework and approval friction all shrink theoretical time saved into real value.
- Real salary bands: hours are priced against actual cost, not a national average.
Multiply a conservative hours figure by real hourly cost, then discount it for adoption and real-world friction across a working year. The result is deliberately smaller than the vendor number, which is exactly why it holds up.
State the confidence honestly
The most persuasive part of a good case is that it never oversells. Show finance how firm the number is, and it earns trust:
- Directional: a calculator-only estimate with no baseline yet.
- Baseline-backed: employee conversations are done, before follow-up evidence.
- Validated: 30/60/90-day follow-up evidence is in hand.
A conservative number you can defend beats an impressive number you cannot. Confidence stated plainly is what gets a budget approved.
Our AI Litmus module builds this case for you: it grounds the estimate in your teams' real workflows and fluency, and labels every figure with its confidence, so you can take it into the room and stand behind it.
The hours in that calculation have to come from somewhere real. AI Litmus is the AI adoption software that produces them: it reads how effectively each role uses the AI you already pay for, maps where the time actually goes, and separates hours that are recoverable from hours a re-run has already verified. If you want the measurement side first, start with how to measure AI adoption and whether your team is using AI well.
See this on your own teams.
A private walkthrough, calibrated to your roles. About two weeks.
Frequently asked
How do you calculate the ROI of AI training? Estimate the productive hours AI can realistically unlock per person, discount them for adoption readiness and real-world friction like verification and rework, price the remaining hours against actual salary bands, and sum across a working year. Subtract the programme and tool cost. Grounding every input in real workflows is what makes the result defensible.
What is a realistic productivity gain from AI at work? Realistic gains vary widely by role and are much smaller than vendor headlines once you account for adoption and rework. The honest approach is to estimate hours conservatively for each function's real workflows rather than applying a single company-wide percentage.
How do you make an AI business case finance will approve? Keep it conservative and traceable: ground it in real workflows and salaries, discount aggressively for adoption and friction, and state the confidence level explicitly, from directional to validated. A number finance can check is a number finance can approve.
Shobhit Khandelwal
Founder, VMS Culture Labs
Shobhit Khandelwal is the founder of VMS Culture Labs, on a mission to measure what most leaders only guess at: how fluently their teams truly work with AI, and the hidden cost of how people behave at work. He is out to replace workplace guesswork with evidence, and build the kind of workplaces the next generation deserves.
Get the intelligence briefing
Occasional, high-signal notes on measuring AI fluency and culture cost. No spam.
Keep reading