Model the case with assumptions you can argue with.
Every vendor ROI calculator is built to produce a large number. This one shows its arithmetic. The automation share is capped well below 100% because review and judgment do not disappear, hourly cost uses 1,880 productive hours rather than 2,080 raw ones, and the platform run rate is subtracted before payback is calculated.
If you disagree with a default, that is the point — bring the corrected figure to the briefing.
Show your work
Every default, stated.
A model whose assumptions are hidden cannot be challenged, which means it also cannot be trusted. These are the figures behind the arithmetic above, and they are deliberately conservative.
| Scenario | Hours per unit | Automation share | Run rate | Build investment |
|---|---|---|---|---|
| Test Engineering | 3.2 | 62% | 8% | $285,000 |
| Business Intelligence | 21 | 68% | 7% | $240,000 |
| Autonomous Workflows | 1.4 | 55% | 10% | $320,000 |
Why 1,880 hours, not 2,080
2,080 is fifty-two weeks at forty hours with nobody ever taking a day off. Once paid time off, holidays, training, and non-project time are removed, 1,880 is the conventional planning figure. Using the larger number would understate hourly cost and flatter the payback.
Why the share is capped
Review, judgment, and exception handling remain human work — that is the design, not a limitation. The mechanical share is what gets absorbed. Any calculator claiming 90% or more is modelling a system with no human in the loop, which is a system your security team will not approve.
What is not counted
Defect escape reduction, audit preparation time, faster release cadence, and the cost of decisions made on stale data are all real and all excluded, because they are hard to defend in a finance review. The figures above are the floor, not the ceiling.
Disagree with a number? Good.
Bring your corrected assumptions to a technical briefing and we will rebuild the model against your actual cycle times and volumes. A business case you argued your way into is one that survives the committee.
Both phases are scoped in published ranges, so you can set this model against a real number.