An owner or CFO does not want a productivity story. They want four numbers. Here is how to calculate AI ROI from your own costs and present a one-page case that survives scrutiny.
A general manager walked into a budget meeting with one slide. It said: the assistant saves the team six hours a week. The owner asked a single question — "six hours of whose time, and what does that cost me?" — and the GM did not have an answer. The budget was declined in four minutes, and the product was fine. The case was the problem.
Why "it saves time" gets rejected
Owners and CFOs do not fund time. They fund cost avoided or revenue that would otherwise walk out the door. "Saves time" is a feeling. "Recovers 26 hours of front-desk labor a month at a stated $22 an hour, which is $572, against a tool cost of $180" is a claim they can check.
The moment you attach a rate and a count, the conversation stops being about belief and becomes arithmetic. That is what you want, because arithmetic can be argued and belief cannot.
There is a second reason the vague version fails. The people approving this budget have been burned before. They bought a tool that promised time savings, and it became another system to maintain, another login, another monthly line item. So your case must not sound like a promise. It must sound like a conservative estimate with its assumptions written down where the sceptic can attack them. Let them attack the assumptions — that means they are taking it seriously.
The four metrics that survive scrutiny
Four numbers, and four only. Everything else is noise that weakens the ones that matter.
| Metric | How to compute it | What you need first | How a sceptic reacts |
|---|---|---|---|
| Hours recovered | weekly hours saved × 4.33 weeks/month | a one-week time log, not a guess | "Prove the log." |
| Labor value | hours recovered × fully loaded hourly rate | the real rate: wage + tax + benefits | "Use the real rate or none." |
| Response-time change | median first-reply time before, minus after | two samples of the same queue | "Is that median or best case?" |
| Cost per resolved conversation | monthly tool cost ÷ resolved conversations | a count of conversations from the tool | "What counts as resolved?" |
Two supporting figures appear on the page but never in the headline: leads captured that would otherwise be lost, and payback period. They support the four; they do not replace them.
A worked example you can swap your numbers into
Read this as a template, not a result. Every figure below is a placeholder, and the output changes the instant you change an input. The tool does not promise any outcome — the arithmetic does whatever your inputs tell it to.
| Metric | Formula | Example input | Result |
|---|---|---|---|
| Hours recovered (monthly) | weekly hours × 4.33 | 6 h/wk × 4.33 | 26 h |
| Labor value recovered (monthly) | hours × fully loaded rate | 26 h × $22 | $572 |
| Response time | median before → median after | 3 h 10 m → 40 s | −189 min |
| Leads captured (monthly) | unanswered contacts now answered | 8 contacts | 8 |
| Leads captured (revenue) | contacts × your close rate × your average value | 8 × [your rate] × [your value] | your number |
| Cost per resolved conversation | tool cost ÷ resolved conversations | $180 ÷ 900 | $0.20 |
The row you cannot fill in from a template is the revenue line, and that is deliberate. Close rate and average job value are yours, not the tool's. If you want to include revenue at all, use your existing historical close rate on comparable enquiries and label it as an assumption. Do not invent a better one because the tool is new.
Hard savings and soft savings
This distinction is where most budget cases die, so be ruthless about it.
- Hard savings are money that leaves or does not leave an account you can name. Overtime you stop paying. A contractor shift you do not renew. A part-time hire you defer. Hours on a payroll line that visibly changes.
- Soft savings are real but unbanked. "The team could focus on higher-value work." "People are less stressed." "We could take on more volume." None of these show up as a lower number anywhere, and the first sceptic in the room will use them to discredit the parts of your case that are solid.
Keep soft savings in a clearly labelled footnote. Never let them into the headline figure.
[!WARNING]
If you cannot point to a payroll line, an invoice, or a budget row that changes because of the automation, the saving is soft. Label it soft and keep it out of the total. A smaller honest number gets approved far more often than a large one that falls apart under one question.
The one-page report format
Hand over one page. Not a deck, not a spreadsheet dump. This table, filled in with your figures and your assumptions stated in the same box as the number they affect.
| Section | One line |
|---|---|
| Problem | What specifically is failing or costing us today |
| What changes | The single workflow the automation takes over |
| Cost | Tool cost per month + one-time setup, both stated |
| Hard savings (annual) | Hours × rate, extended to 12 months |
| Payback period | Setup cost ÷ monthly net benefit, in months |
| Break-even volume | Conversations/month at which cost = benefit |
| Assumptions | Rate, hours, close rate — each labelled as an estimate |
| What we are not claiming | Soft savings, and any revenue we have not modelled |
The last row is the one most people leave out and the one that earns trust. Telling the approver what you are not counting signals that the rest is solid.
Paste this into a document or the body of the email before you build anything else. Fill every bracket or delete the line — a blank bracket is a question the approver will ask out loud.
BUDGET CASE — [workflow name] [date]
Goal: [the single workflow this replaces]
Cost: $[tool]/mo + $[setup] one-time
Hours: [x] h/week, logged [dates] × $[loaded rate] = $[monthly]
Response: median first reply [before] → [after]
Convos: $[tool]/mo ÷ [n] resolved = $[x] per conversation
Hard saving: $[annual]
Payback: [setup] ÷ ([monthly saving] − [monthly cost]) = [n] months
Break-even: [n] conversations/month
Assumptions: rate and hours are estimates; close rate not assumed
Not claimed: soft savings; revenue not modelled above
Ask: $[total] to run this for [n] months
What to do today
- [ ] Run a one-week time log on the workflow before you claim any hours. Measure, then calculate.
- [ ] Pull the fully loaded hourly rate (wage + tax + benefits), not the wage.
- [ ] Fill the worked-example table above with your numbers and see whether the payback still holds when you halve the hours saved.
- [ ] Write the one-page report, soft savings in the footnote, and name your assumptions out loud.
The approval does not come from a bigger number. It comes from a smaller number you can defend line by line.
This guide lives at AgentFast, where you'll also find a free audit tool that checks your own agent's instructions against these failure modes. Posted here with a canonical URL pointing at the original.
Top comments (0)