Step five
One initiative, so nothing gets counted twice. Capacity created, savings actually realized, the cost of getting there and the timing of each are kept apart, because collapsing them is how AI business cases go wrong.
Subtracted once, undiscounted, at time zero. It is not repeated in year one; year-one costs go in the table below.
Currency in USD millions; fractions in percent, constrained to 0–100.
| Input | Year 1 | Year 2 | Year 3 | Year 4 | Year 5 |
|---|---|---|---|---|---|
| Eligible labour cost base Caps how much labour saving is even possible. | |||||
| Adoption | |||||
| Gross effort reduction | |||||
| Review and rework cost | |||||
| Realizable fraction of net capacity | |||||
| Incremental revenue | |||||
| Contribution margin on that revenue | |||||
| Cannibalized revenue | |||||
| Cannibalized contribution margin | |||||
| AI operating expense | |||||
| Implementation operating expense | |||||
| AI capex | |||||
| Related D&A | |||||
| Incremental operating NWC (level) End-of-year level. The model uses the year-on-year change. |
The persistent policy grows year-five incremental NOPAT at 2.0% and reinvests at a 12.0% terminal ROIC. It is a disclosed policy choice, not a company disclosure.
| Year | Gross capacity | Net capacity | Realized benefit | Incremental EBIT | Baseline cash tax | Scenario cash tax | Incremental tax | Incremental FCFF | Discounted |
|---|---|---|---|---|---|---|---|---|---|
| Time zero | — | — | — | — | — | — | — | -$10.0m | -$10.0m |
| 1 | $5.0m | $2.2m | $1.3m | -$6.9m | $41.7m | $40.0m | -$1.7m | -$5.9m | -$5.5m |
| 2 | $11.9m | $9.0m | $5.4m | -$1.1m | $45.3m | $45.0m | -$0.3m | -$0.8m | -$0.7m |
| 3 | $18.0m | $14.9m | $9.0m | $5.7m | $49.1m | $50.5m | $1.4m | $5.2m | $4.0m |
| 4 | $21.9m | $18.6m | $11.2m | $10.7m | $53.1m | $55.8m | $2.7m | $9.2m | $6.6m |
| 5 | $24.5m | $21.1m | $12.6m | $12.5m | $57.5m | $60.6m | $3.1m | $10.7m | $7.1m |
Finite-horizon incremental NPV
$1.6m
Five years, discounted at 8.5%
Incremental enterprise value
$1.6m
Under the none persistence policy
Share of the value gap it closes
0.6%
Gap to close $247.9m
Solve
Holds every other input fixed and solves for one declared unknown: the same realized labour benefit in each forecast year. The answer is bounded by the eligible labour cost base.
P1, not built
Structured extraction of AI claims from reports and earnings-call transcripts is a later release. It is not implemented here, and no claim objects are generated. Statements can be consistent with a claim without showing that AI caused the change, so claims, observed results and modelled effects would stay in separate columns when that arrives.