How to Measure the ROI of an AI Consultant Engagement

Bottom line: the ROI of an AI consultant engagement is measured by comparing hard, tracked cost or time savings (plus new revenue directly linked to the work) against the total fee paid, over a defined period, and most businesses get this wrong because they measure activity instead of money. Do it by setting a baseline before you start, tracking one or two numbers weekly, and checking at 90 days whether the saving has landed in your accounts or just quietly vanished into everyone's diary.

Start with the number nobody sets: the baseline

If you don't know what a process cost before the AI consultant touched it, you cannot measure what changed. This sounds obvious. It is also the single thing I see skipped most often.

Before any engagement I run, I ask the client to give me three numbers from the four weeks before we start:

• Hours spent on the task, by whom, at what internal cost per hour
• Volume: how many quotes, calls, invoices, leads, or replies went through the process
• Error or drop-off rate, if there is one (missed follow-ups, late invoices, unanswered enquiries)

I worked with a recruitment agency in Manchester in 2025 where the office manager was spending roughly 14 hours a week screening CVs and drafting first-response emails to candidates. Her time cost the business about £22 an hour loaded (salary, NI, pension, overhead). That's £308 a week, or roughly £16,000 a year, on one task nobody had ever priced out loud before I asked.

Without that baseline, any claim about the engagement "saving time" is a feeling, not a figure.

The formula, and why the simple version lies to you

The textbook ROI formula is:

(Gain from investment minus cost of investment) divided by cost of investment, expressed as a percentage.

For the recruitment agency, the engagement was a 90 day build: screening automation, an AI-drafted first response, and a handover workflow into their applicant tracking system. The fee was £6,500. After the build, the office manager's time on that task dropped to 4 hours a week, saving 10 hours, or £220 a week.

Over the first 90 days (roughly 13 weeks), that's £2,860 in labour saved against a £6,500 fee. On paper that's a negative return in quarter one. Anyone reporting ROI only at the 90 day mark and stopping there is going to undersell or oversell the work, depending on which quarter they picked.

Run it to 12 months instead: £220 a week times 52 weeks is £11,440, against the same £6,500 one-off fee. That's a return of roughly 76% in year one, and the saving repeats every year after with no further fee attached. That's the number that matters, and it's also the number that gets left out of most sales pitches because it takes patience to calculate.

The uncomfortable bit: saved hours don't automatically become saved money

Here's what most write-ups on this topic won't tell you. Freeing up 10 hours a week doesn't put 10 hours of value in the bank unless someone decides what happens to those 10 hours.

In the Manchester case, the office manager didn't get 10 hours back and go home early, and the business didn't cut her role. She absorbed the freed time into other admin that had been piling up: onboarding paperwork, a supplier chase list, client check-ins that hadn't happened in months. All useful. None of it was tracked, priced, or reported as a return.

This is the trap I'd call reabsorption, and it's why so many businesses feel like an AI project "didn't really pay for itself" even when the automation worked perfectly. The time was saved. Nobody decided what the saved time was worth doing next, so it disappeared into the general noise of the working day and never got counted.

If you want a real ROI figure, you have to name, in advance, what the freed hours will be redirected to, and put a value on that too. Redirected to sales calls? Track pipeline added. Redirected to reducing a temp staff bill? Track the invoice that shrinks. Redirected to "catching up"? Be honest that this is a soft benefit worth writing down but not one to put in a hard ROI column.

Three categories of return, and only one of them is easy to measure

1. Hard cost savings

Labour hours reclaimed and priced at loaded cost, software subscriptions cancelled because the AI consultant replaced three disconnected tools with one workflow, reduced overtime, reduced agency or freelance spend on tasks now automated. These are countable in pounds within a month or two.

2. Revenue directly attributable to the work

This is where most people get sloppy and claim credit for growth that would have happened anyway. Only count revenue you can trace: a faster lead response time that measurably lifted conversion rate, a follow-up sequence that recovered quotes that would otherwise have gone cold, a booking system that reduced no-shows. If you can't draw a line from the specific change the consultant made to the specific pound that landed, don't put it in this column. This is also the category where the biggest, most quotable ROI stories get made up, so be the person in the room who checks.

3. Soft, real, but not countable in pounds

Staff stress reduced, fewer errors reaching customers, better data for decisions, capacity to take on more clients without hiring. All valuable. None of it belongs in your ROI percentage. Write it down separately so it doesn't get lost, but don't let it inflate the headline number.

A step-by-step you can run

1. Before the engagement starts, record baseline hours, cost per hour, volume, and error rate for the process being changed.
2. Agree with the consultant, in writing, what "done" looks like and by when. If you're at the stage of choosing someone, this is exactly the kind of thing to pin down before signing, and it's covered in detail in how to hire an AI consultant.
3. Set a 30/60/90 day check-in, not just a final one. Early numbers will look worse than they end up being, and that's normal, not a red flag.
4. At each check-in, re-measure the same baseline numbers using the same method. Don't switch how you calculate loaded hourly cost halfway through, it wrecks the comparison.
5. Decide in advance what freed-up time will be redirected to, and track the outcome of that redirection separately.
6. Calculate ROI at 90 days and again at 12 months. Report both. The 90 day figure tells you whether the build worked. The 12 month figure tells you whether it paid for itself.
7. Subtract ongoing costs (API fees, subscription costs, any maintenance retainer) from year-two-onward calculations, because the fee for the original engagement won't repeat but the running costs will.

What a realistic ROI range looks like

Fees for a focused, single-process AI consulting engagement in the UK typically run £3,000 to £15,000 depending on scope and how much custom build is involved, a range I break down on what an AI consultant costs. Against that, a well-scoped engagement that automates one repetitive, high-volume task tends to pay back within 4 to 9 months and return 60% to 150% in year one once you count both hard savings and traceable revenue.

Engagements that promise ROI inside 30 days are usually measuring the wrong thing, most likely hours logged in a demo rather than money that moved. Engagements that never show a number by month six probably weren't scoped tightly enough in the first place, which is a strategy problem more than a measurement problem, and it's the exact gap a proper AI strategy consultant is meant to close before any build starts.

Where this gets industry-specific

The baseline and the formula stay the same everywhere, but what counts as a "hard" saving changes by sector. A firm doing AI automation consulting on invoice processing measures days-to-payment and late fees avoided. A hospitality business measures no-show rate and review response time, which is exactly the territory covered in work on AI for restaurants and hospitality bookings and reviews. A legal practice measures billable hours reclaimed from document review, always within the strict compliance boundaries laid out for what's safe to automate in a law firm. If your consultant can't tell you, in your sector's language, what the equivalent metric is for your business, ask them to before you sign anything, not after.

The report that tells you the truth

By the end of a run engagement, you should have a one-page comparison: baseline hours and cost, current hours and cost, fee paid, hard saving to date, traceable revenue to date, and a plain sentence on where the freed-up time went. If your consultant can't produce that page, or won't, treat it as information, because a good one builds this measurement in from day one rather than backfilling a nice-sounding number at the end. That's part of what separates a solid AI consultant delivering results in 90 days from someone who delivered a demo and a good story.

The Manchester agency now reviews that one-page report every quarter. The number moves. Some quarters the office manager's redirected hours produce a traceable win (three extra client check-in calls turned into a contract extension worth £4,200). Some quarters they don't. That honesty is worth more than any single glossy ROI slide, because it's the version you can run your business on.

For the closest example to your business, start with AI consultant by industry.

I built a free engagement rate calculator for exactly this.

Related: the AI time and savings calculator.

Frequently asked questions

What's a reasonable timeframe to expect ROI from an AI consultant?

Most focused engagements start showing a positive return between month four and month nine, once you account for build time, staff adjustment, and the fee itself. Anyone promising full ROI inside 30 days is almost certainly counting activity, not money.

What's the simplest ROI formula to use?

(Gain from investment minus cost of investment) divided by cost of investment, run as a percentage. Calculate it at 90 days and again at 12 months, because the two numbers usually tell different, both useful, stories.

Should I count saved staff hours as pure ROI?

Only if you decide, and track, what those hours get redirected to. Saved time that just gets absorbed into general admin with no defined next use is real, but it isn't ROI you can put a pound figure on with a straight face.

What if the consultant can't show me a clear number after the engagement?

Ask for the baseline they used and how they measured it. If there wasn't a baseline set before work started, any ROI figure they give you now is a guess dressed up as a report.

Further reading

• Anthropic's own documentation
• the NIST AI Risk Management Framework