Watch: the module in 1:08
Read the transcript
Most AI business cases are built to pass approval rather than to be right. This module is about raising the quality of that conversation.
The first question is always: compared with what? A case without a measured baseline is a direction, not a measurement.
Time saved is an input to a management decision. Name the action that captures it, who owns it and when it happens.
Finance then asks three things: what the case is worth today, how fast the money comes back, and how it will show in the accounts. Judge most AI cases over three years at most. A one-year delay can remove a whole year of benefit while every cost still lands.
Finally, stress the two things most likely to move: unit price and demand.
In the exercise, a case that passes at 1.5 falls to 1.1 under stress, and fails. Find the price at which it breaks, and two ways to widen the margin.
By the end of this module you can
- Test a business case for a measured baseline, a conversion mechanism and a full cost denominator.
- Apply NPV, IRR and payback to an AI case and explain why the horizon should be short.
- Classify return dimensions, capture status and portfolio categories without blending them into one number.
- Stress-test a case against price and demand and state where it fails.
Finance lens: this is your core module. Advisor lens: the CFO's ten questions are the client test.
The argument in brief
A business case is the request for money before the return exists. Most AI cases arrive at finance built to pass approval rather than to be right: productivity estimates without baselines, time saved treated as money, and cost models that leave out governance, integration and the people who make the AI useful.
This module is about raising the quality of that conversation. It covers the five things a case needs before finance should accept it, the three finance tests (value today, speed of payback and accounting treatment), how to label returns honestly by dimension and capture status, and how to stress-test a case against the two things most likely to move: unit price and demand.
Key ideas
- The first question is always: compared with what? A case without a measured baseline is a direction, not a measurement.
- Time saved is an input to a management decision. Name the action that captures it, its owner and its timing.
- Ask finance for the discount rate. Price risk through named scenarios, not by stacking premiums on the rate.
- Judge most AI cases over three years at most. A one-year delay inside a three-year life removes a full year of benefit while every cost still lands.
- Accounting treatment does not change the cash or the NPV, but it changes who pays and who approves. State it and have finance confirm it.
Terms to know
- AI ROI
- The return created by AI initiatives relative to the cost of building, operating, and governing them.
- Cost Avoidance vs. Cost Reduction (AI)
- The distinction between prevented future cost increases (cost avoidance) and actual decreases in current expenditure (cost reduction) - a distinction that matters significantly for AI ROI credibility.
- Realised Productivity Gain
- The economic value actually captured from AI-driven time savings, as distinct from the theoretical time freed by an AI capability.
- Value Capture Failure
- The outcome when an AI system performs as designed technically but the organisation fails to convert that technical performance into economic benefit.
- Net AI value
- A management equation: (incremental outcome × monetisation × attribution × persistence × confidence) minus total cost, expected loss and opportunity cost.
Read
In this order. Each opens on the main site.
- 1What CFOs Should Ask of AI ROI ClaimsBaselines, the conversion mechanism and the cost denominator.18 min
- 2The AI Business Case: Finance TreatmentNPV, IRR, payback and capex or opex, with a worked case.9 min
- 3AI ROI ModelsReturn dimensions, proofs, stage gates and capture status.20 min
- 4What to Bring to an AI Investment ReviewThe record a review needs before the next commitment.4 min
Worked example: the same case, as finance sees it
Invented figures. An internally built assistant costs £400,000 to build in year zero and £150,000 a year to run. Booked benefits ramp from £200,000 in year one to £400,000 in years two and three. The horizon is three years.
| Simple view | £1,000,000 of benefit less £850,000 of cost: a £150,000 surplus. |
|---|---|
| NPV at 10% | About £40,000, a quarter of the simple surplus. IRR about 14.6%, so it fails a 15% hurdle. |
| Payback | About 2.4 years, late in a three-year life. |
| Benefits a year late | Costs still land. NPV falls to about −£307,000, with no payback inside three years. |
| Running cost £30,000 higher | NPV about −£35,000, payback 2.7 years. |
The result
A case that looks comfortably positive is thin once discounted, and either a late start or a modest rise in running cost turns it negative. The adoption plan matters more than the model choice.
Discount, keep the horizon short, state the accounting treatment and stress-test before the case goes in.
From The AI Business Case: Finance Treatment, where the figures are explained in full.
Try it: stress-test a case against price and demand
A claims automation service costs £40,000 a month, of which 60% moves with token price and demand. It produces £60,000 a month of booked value. Extra demand produces value at half today's rate. Finance requires £1.20 of value for every £1 of cost.
| Monthly cost and usage share | £40,000 and 60% |
|---|---|
| Monthly value | £60,000 |
| Value from extra demand | 50% of today's rate |
| Hurdle | 1.2 (value per £1 of cost) |
| Stress scenario | price 1.5 times, demand 1.3 times |
- Enter the figures, or open the stress test with them loaded.
- Read the value-to-cost ratio today, and under 1.5 times price with 1.3 times demand.
- Find the highest price multiple, at today's demand, at which the case still meets the hurdle.
- Name one design change and one commercial change that would widen the headroom.
Open the budget stress test with these figures(opens in a new tab)
Show the worked answer
Today: £60,000 ÷ £40,000 = 1.5, which passes the 1.2 hurdle.
Stressed: cost rises to £16,000 + £24,000 × 1.5 × 1.3 = £62,800, and value to £60,000 × (1 + 0.3 × 0.5) = £69,000. The ratio falls to about 1.10, which fails.
Price headroom at today's demand is about 1.42 times: (£60,000 ÷ 1.2 − £16,000) ÷ £24,000.
Design: cache the stable prompt, move work that can wait to batch, or route simple claims to a cheaper model that clears the quality floor. Commercial: commit for the base load at a discount, or negotiate a price cap for the term.
A thin margin is a design question. Know the price multiple at which the case fails before you ask for the money.
Check your understanding
Five practice questions. The final exam uses different questions in the same style.
Practice only. Nothing is recorded.
Progress is saved in this browser only.