An AI value figure your board can defend.
List each AI use case with its owner and its test-and-control result. The ledger grades how each value was measured and totals only what would survive an audit committee's questions, gross and net.
DBS Bank reports about S$1 billion a year in economic value from AI, and says most of it is measured by comparing a group that gets the AI treatment with one that does not. Most organisations quoting an AI value figure have no control group, no named owner and no written definition. This ledger applies the stricter method to your own use cases. How DBS measures it →
Your ledger
| Use case | Type | Per unit | Annual value (95% interval) | Net of cost | Grade | Actions |
|---|
–
gross, grade A only
–
net of build and run cost
–
of measured value is grade A
| Grade A, reported: revenue uplift | |
|---|---|
| Grade A, reported: cost and productivity | |
| Grade A, reported: losses averted | |
| Disclosed separately (grades B and C) | |
| Excluded estimates (grade D) |
Written definition for the board paper
This ledger applies a grading rule to figures you enter. It does not verify them, and it does not constitute an audit, an assurance opinion or accounting advice. The confidence interval assumes independent units and roughly normal averages; clustered or seasonal data need a statistician's review. The illustrative ledger describes a fictional bank. Your entries are kept in this browser only and are neither sent to Terence Kok nor reviewed by anyone.
Four grades of evidence.
| Grade | What it takes | In the board figure |
|---|---|---|
| A · Measured | Test and control, a named business owner, at least 3 months of measurement, and a 95% confidence interval above zero. | Counted in the board figure. |
| B · Measured, not confirmed | Test and control that misses one of the A conditions: no owner, too short, no spread entered, or an interval that includes zero. | Disclosed separately, with the reason. |
| C · Baseline comparison | Compared with a documented baseline, such as the same period last year. Anything else that changed in the period is counted as AI value too. | Disclosed separately. |
| D · Estimate | An estimate, such as the figure in the original business case, with nothing measured against it. | Excluded. |
How the illustrative ledger sorts into four grades
Board figure 16.8m gross, 15.0m net of cost. That is 64% of the 26.3m measured; the rest is disclosed or left out.
Values computed with this ledger's own rule from its illustrative ledger (a fictional bank), in the ledger's currency and annualised. A use case missing only its owner or a long enough window still drops to B.
Per-unit value is the treatment average minus the control average for revenue, and the control average minus the treatment average for cost and losses, so a positive number is always good. Annual value multiplies it by the units the AI applies to in a year. The 95% interval uses the standard error of the difference between two means, which is why the group sizes and standard deviations are needed.
The Excel template has the same columns, formulas and grading rule, with the illustrative ledger on a second sheet. For the figures before a use case is built, the AI ROI Calculator models the expected return; for the adoption picture across a portfolio, see the AI Value & Adoption Dashboard.
Most AI value figures I am shown in board papers are the business case numbers with the word estimated removed. Nobody set up a control group, nobody owns the number, and the definition changes each time the programme needs a bigger result. This ledger is the discipline I ask finance teams to adopt before any figure goes to the board: a named owner, a comparison group, a written definition, and a smaller number that holds up rather than a large one that does not.
