How to calculate AI ROI from your own numbers — and get the budget approved

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.

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.

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.

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.

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.