ROI and the business case
Executives fund outcomes, not architectures. A clear, honest business case is often what turns a promising pilot into a production budget — and FDEs who can build one are rare. You don't need a finance degree; you need a simple model, defensible assumptions and the discipline to show ranges.
The basic model
Annual value = volume × time saved per unit × fully-loaded cost per hour
+ (error reduction × cost per error)
+ (revenue impact, if any)
Annual cost = build cost (amortised) + infrastructure + model/API usage
+ support and maintenance
ROI = (annual value − annual cost) ÷ annual cost
Payback = upfront cost ÷ monthly net value
Start with the component you can measure best — usually time saved — and add others only if you have evidence.
A worked example
The numbers below are illustrative, to show the mechanics.
A support team handles 2,000 tickets a day. Discovery found agents spend about 6 minutes per ticket searching knowledge articles. In the pilot, an answer assistant cut search time to about 2 minutes on the tickets where it was used, and it was used on 60% of tickets.
| Input | Value |
|---|---|
| Tickets per year | 2,000 × 250 working days = 500,000 |
| Tickets using the assistant | 60% → 300,000 |
| Time saved per assisted ticket | 4 minutes |
| Hours saved per year | 300,000 × 4 ÷ 60 = 20,000 hours |
| Fully-loaded cost per agent hour | $25 |
| Annual value | $500,000 |
Against that, estimate costs: model usage per ticket × volume, hosting, and the engineering time to maintain it. If annual cost were $150,000, net value is $350,000 and ROI is roughly 2.3×.
Show ranges, not a single number
Every input is an estimate. Present three scenarios so executives see you have thought about uncertainty:
| Scenario | Adoption | Minutes saved | Annual value |
|---|---|---|---|
| Conservative | 40% | 3 | $250,000 |
| Expected | 60% | 4 | $500,000 |
| Optimistic | 75% | 5 | $781,000 |
If the conservative case still pays back, the decision is easy. If only the optimistic case works, say so — that honesty is worth more than the deal.
Time saved is not money saved unless the time is redeployed. Say what the freed capacity becomes: handling growth without new hires, faster response times, or higher-value work.
Where the inputs come from
- Volumes — the customer's systems or reports (ask for them in discovery).
- Baseline times — shadowing, system timestamps, or a short time study.
- Improvement — your POC or pilot measurements, not vendor benchmarks.
- Costs per hour — the customer's finance team; never guess salaries out loud.
- Your costs — measured token usage from the pilot, scaled to production volume.
Presenting to executives
Lead with the outcome and the number, then the evidence, then the ask:
- Headline — "The assistant saves an estimated 20,000 agent-hours a year."
- Evidence — pilot results against the baseline, with the quality metric.
- Range — conservative to optimistic.
- Costs and risks — what it takes to run, and what could reduce the value.
- The ask — the decision you need, with a date.
Keep the whole business case to one page, with the model in an appendix. If the executive wants the details, they will ask — and you will be ready.
Interviewers love "How would you justify this project to the CFO?" Walk through the model out loud, name your assumptions, and show a conservative case. The structure matters more than precise numbers.
Senior Full-Stack Engineer · AI & AWS. I turn ambiguous customer problems into production systems — cloud, data and AI.