How to measure the ROI of AI and automation

Why most AI ROI claims are hard to trust

Vendors often quote hours saved per user per week. Those figures are usually self-reported and rarely tied to money. A partner who saves 40 minutes a day drafting emails may or may not bill more, close more or go home earlier.

In a service business, ROI shows up in a few places: more opportunities handled, more of them won, faster payment, or the same work done with less rework or overtime. A good measurement plan picks the one or two of those that the improvement should move, and ignores the rest.

It also helps to separate leading indicators from results. Usage, turnaround time and correction rate move first and tell you whether the change is working mechanically. Revenue, win rate and cash collection move later and tell you whether it mattered. Track both, but judge the investment on the second set.

Step one: baseline before you build

Pick the workflow and measure it as it runs today. Use five to twenty real recent examples, not a guess from memory. Useful baseline measures include:

Keep the baseline simple enough that someone will actually maintain it. A shared spreadsheet with one row per example, the date, the minutes spent and whether it needed rework is usually enough. Elaborate tracking tends to stop after the second week.

• Staff minutes per instance, for example per proposal, per onboarding packet or per monthly report.
• Elapsed time, such as hours from enquiry to first reply, or days from site visit to quote sent.
• Conversion, such as the share of quotes accepted or the share of enquiries booked.
• Correction rate: how often the output has to be redone or fixed before it goes out.
• Cash measures where relevant: days sales outstanding or the share of invoices paid within 30 days.

Step two: work out break-even in plain numbers

Break-even tells you how much extra business or saved cost the improvement must produce to pay for itself. Here is an illustrative example. It is not a forecast.

Say a 90-day improvement to proposals costs $5,000 in fees and $500 in tools, so $5,500 in total. Your average new client is worth $4,000 and your contribution margin on that work is 40%. Each new client therefore contributes $1,600 toward costs.

To recover $5,500 you need $5,500 divided by 40%, which is $13,750 in additional sales. That is a little under four extra clients at $4,000. If faster, cleaner proposals helped you win five more clients than your baseline, that is $20,000 in added sales, $8,000 in contribution, and $2,500 after the $5,500 cost.

Write that number on the project plan. It turns a vague goal like "use AI for proposals" into a testable one: does the change plausibly produce four more wins, or the equivalent in saved time, within a reasonable period?

If the break-even looks out of reach, that is a finding worth having before you spend anything. Shrink the scope, choose a higher-value workflow, or decide not to proceed.

Step three: count all of the costs

The tool subscription is usually the smallest cost. A fair ROI calculation includes:

• Outside fees for design, setup and project leadership.
• Software: AI seats, automation platform plans such as Zapier, Make or n8n, and any paid integrations. Check current pricing, as it changes often.
• Staff time to document the process, test with real examples and give feedback.
• Training time for everyone who will use it.
• Ongoing care: someone has to fix a broken integration or update a template when prices change.

Turning saved time into money, honestly

Saved time is real only if it goes somewhere. Ask what happens to the hours. Three answers count: the person handles more volume without new hires, the person moves to billable or revenue work, or overtime and contractor spend go down. If none of those happens, record the time saving as a quality-of-life gain and do not put a dollar figure on it.

For example, if an office manager spends six hours a week assembling client reports and an automation cuts that to two, the four hours are worth something only if they go to collections calls, onboarding or another task you were behind on. Note where the hours went in your 90-day review.

Quality gains deserve a line too, even when you cannot price them. Fewer corrections on proposals, fewer missed documents in onboarding, or more consistent client reports reduce the risk of an unhappy client. Record them as counts, such as corrections per 20 proposals, so they can be compared at the review.

An illustrative time-based example

Not every improvement is about winning more work. Suppose an accounting firm spends about 25 minutes per client assembling a monthly management report from QuickBooks Online and a spreadsheet, for 60 clients. That is roughly 25 staff hours a month. The numbers here are for example only.

An automation that pulls the figures and drafts commentary for a person to review might cut that to 10 minutes per client, or about 10 hours a month. The saving is 15 hours a month. If those hours go to billable advisory work at, say, $120 an hour, that is $1,800 a month of capacity. If they simply absorb busy-season overtime, the value is the overtime avoided. If they go nowhere in particular, the honest number is zero dollars and a calmer team.

The same discipline applies: baseline first, agree what the hours will be used for, then check at 90 days whether they were.

Running a 90-day measurement cycle

A simple rhythm works for most firms. In days 1 to 30, agree the workflow, the baseline and the break-even number. In days 31 to 60, build and test with real users and real examples. In days 61 to 90, measure adoption and compare to the baseline.

Adoption is the measure people skip. If only two of eight staff use the new proposal process, the ROI will look poor for reasons that have nothing to do with the technology. Track usage weekly and ask why people avoid it.

At day 90, make one of three decisions: keep and extend, adjust and re-measure, or stop. Stopping is a legitimate result when the numbers do not move.

This is the measurement approach InTheory uses in its 90-day engagements: a written baseline, an agreed break-even, and a review against both.