AI ROI calculator for consultancies and professional services firms
What would AI actually return?
The return on an AI project in a professional services firm is the value of the hours it frees, less what it costs. Hours saved per person a week × the people who actually use it × working weeks × the value of an hour × the share of freed time put to valuable work, against the fixed fee and a year of running costs. Put in your own figures below; the method and every assumption are explained underneath.
Hours, people and price, in one number.
Change any figure and the results update as you type. Start with one task you can measure, not the whole firm.
- Net return, year one
- £5,400
- Payback on the fee
- 6.3 months
- Return on cost, year one
- 64%
- Hours freed a year
- 690
- Value of that time a year
- £13,800
- Cost in year one
- £8,400
- Break-even, hours per person a week
- 1.2
- Net over three years
- £28,200
10 people × 2 hrs × 46 weeks × 75% using it = 690 hours. 690 hours × £40 × 50% put to work = £13,800 a year.
The starting inputs are an example to change, not a result we have achieved. Nothing you type leaves your browser.
How the calculator works it out.
The calculator does seven sums, in order. None of them is clever, which is the point: a finance director should be able to check every step with a pen. Here they are with the example inputs the calculator starts from: ten people, two hours saved each a week, 46 working weeks, three quarters of the team using the tool, an hour worth £40, half of the freed time put to valuable work, a £6,000 fixed fee and £200 a month to run it.
| Step | Example |
|---|---|
| Hours freed a year10 × 2 × 46 × 75% | 690 hours |
| Value of that time690 × £40 × 50% | £13,800 |
| Cost in year one£6,000 + 12 × £200 | £8,400 |
| Net return, year one£13,800 − £8,400 | £5,400 |
| Return on cost£5,400 ÷ £8,400 | 64% |
| Payback on the fee£6,000 ÷ (£13,800 ÷ 12 − £200) | 6.3 months |
| Break-even hours£8,400 ÷ (10 × 46 × 75% × £40 × 50%) | 1.2 hrs per person a week |
On those figures the project returns £5,400 in its first year and pays back its fee in 6.3 months. That is an illustration of the arithmetic, not a forecast and not a result from one of our projects. The break-even line is the one to look at first: each person needs to save 1.2 hours a week for the project to recover its fee and a year of running costs within twelve months. If the task you have measured clearly beats that, the project deserves a closer look. If it only works with heroic assumptions, it does not.
Six inputs decide the answer.
Hours saved per person a week
This is the number most often made up. Measure it on one task: how long the task takes today, how often each person does it, and how long it takes with the tool. A proposal that took two to three hours and now takes five to ten minutes, written three times a week, is around seven hours. A general assistant spread thinly across the day saves much less per person, and the saving is harder to see.
Share who actually use it
Buying licences is not the same as adoption. Some people will use a new tool every day, some once, and some never. Training, and a tool built around the way your team already works, are what move this number.
Working weeks a year
Most full-time workers in the UK are entitled to 28 days of paid leave a year, which is 5.6 weeks and can include bank holidays. That leaves about 46 weeks before sickness, training and quiet periods, so 46 is a reasonable ceiling, not a floor.
Value of an hour
Use the fully loaded cost of the people doing the work: salary, employer National Insurance, pension and a share of overheads. Use the charge-out rate only for hours that will genuinely be billed. Valuing every saved hour at the charge-out rate is the quickest way to an impressive number that nobody will see in the accounts.
Freed time put to valuable work
Saved time is only worth money if it goes somewhere: more client work delivered by the same team, a hire you no longer need to make, overtime or contractors you stop paying, or faster turnaround that wins work. Time that dissolves into a slightly shorter week has a value to your people, but it will not pay the fee. Half is a cautious starting point; move it once you know where the hours will go.
Fee and running costs
Put the one-off fee in the fee box and everything that recurs in the monthly box: licences, hosting, model usage and support. If your own people will spend real time scoping, testing and learning the tool, add the cost of that time to the fee. A fair ROI counts every pound on both sides.
What an hour of your team really costs.
Salary is only the start. Since April 2025 employers have paid Class 1 National Insurance at 15% on earnings above £5,000 a year; the rate and threshold are unchanged for 2026 to 2027. Under automatic enrolment an employer must pay at least 3% of an employee's qualifying earnings, the band between £6,240 and £50,270, into a workplace pension.
For someone on £45,000, that is £6,000 of employer National Insurance and £1,163 of minimum pension, a statutory cost of £52,163. Spread over 37.5 hours a week for 46 weeks, 1,725 hours, that is about £30.24 an hour before a penny of overhead. Office space, software, management time, benefits and non-billable time all push the real figure higher, which is why the calculator starts at £40. The helper inside the calculator does this sum for any salary; add your own overheads on top.
How many hours does AI really save?
The largest public UK study is the government's own. In the Microsoft 365 Copilot experiment run by the Government Digital Service from September to December 2024, about 20,000 government employees in twelve organisations reported saving an average of 26 minutes a day, roughly two hours a week. The report is clear that the savings were self-reported and that they varied widely between people and professions, so treat two hours as a sense check for a general assistant, not a promise.
Tools built for one workflow behave differently: they save more on less. The proposal generator we built for Clear AIMS takes the average proposal from two to three hours of work to five to ten minutes. At The Litmus Partnership, a benchmarking platform handles 2,000 lines in around 45 minutes where 200 to 300 lines took eight hours, and across an eight-month programme the firm now saves 268 hours a month, every figure collected from the people who use the tools.
Keep two kinds of number apart. Hours identified are the opportunity a review finds: our two-week review for Litmus identified 9,672 hours a year across twenty-six initiatives, and the review for Plan Ahead PM identified 6,722. Hours saved are what the tools deliver once they are built and used. An honest ROI case is built on the second, and uses the first only to decide what to build next.
- Self-reported saving, UK government Copilot trial
- 26 min a day
- Clear AIMS proposal, before and after
- 2 to 3 hrs → 5 to 10 min
- Saved each month at The Litmus Partnership
- 268 hours
- Identified in one two-week review
- 9,672 hours a year
Five ways AI ROI gets inflated.
- Counting licences as adoption. A hundred seats is not a hundred users. Use the share who will use it in a normal week.
- Valuing every hour at the charge-out rate. Only billed hours earn fees. Everything else is worth its cost, at most.
- Assuming all freed time becomes revenue. Some of it becomes a calmer week. Say where the hours will go, or discount them.
- Forgetting the running costs. Licences, model usage and support recur every month. Over three years they can outweigh the fee.
- Adding up hours across tasks that overlap. Two tools cannot both save the same hour. Build the case one workflow at a time.
From a calculation to a business case.
A calculator is only as good as the hours you put into it. If you already know the one bottleneck you want gone, a 30-day implementation sprint, from £6,000, spends its first week agreeing exactly what to build and puts it live within the month; if it is not live and delivering value in 30 days, we keep working at no extra cost until it is. If you are not sure where AI would pay back, the AI Strategy Review, from £12,000, finds, costs and prioritises every opportunity in your firm through surveys, stakeholder interviews and a review of your tech stack, so that you see the number and the return it is based on before you commit to any build. Every service and what it includes is on our pricing page, and our guide to AI implementation cost sets out what a first year of AI work costs a professional services firm.
We will not quote a build before we know where your hours go, and neither should anyone else. If you would like a quicker first read on where your firm stands, the free AI maturity assessment takes about two minutes. For the wider picture, read our guides to AI for consultancies and AI for professional services firms.
Questions
How do you calculate the ROI of AI in a professional services firm?
Multiply the hours a tool saves each person a week by the number of people who actually use it and the weeks they work in a year. Multiply those hours by what an hour is worth to the firm, then by the share of the freed time that goes into work that earns or saves money. Take away the fee and a year of running costs, and divide what is left by those costs. The payback period is the fee divided by the net value each month.
How many hours a week does AI realistically save per person?
It depends on the work, and averages hide a wide spread. In the UK government's Microsoft 365 Copilot experiment, about 20,000 government employees reported saving an average of 26 minutes a day, roughly two hours a week, but the figures were self-reported and varied a great deal between people and professions. A tool built for one specific workflow can save far more on that task: the proposal generator we built for Clear AIMS takes a proposal from two to three hours to five to ten minutes. Measure your own tasks before you trust any average.
Should saved hours be valued at cost or at the charge-out rate?
Use the employment cost per hour unless the freed time will genuinely be billed. Most saved time is absorbed into work the firm already does, so valuing every hour at the charge-out rate overstates the return. If capacity is your constraint and you have the work to fill it, the charge-out rate is fair for the share of hours that become billable, which is what the calculator's 'put to valuable work' setting is for.
What should count as the cost of an AI project?
The fee, plus every monthly cost that comes with it: licences, hosting, model usage and any support. Count the time your own people spend agreeing what to build, testing it and learning it too, which is easy to forget. The calculator takes the fee and a monthly running cost; add internal time to the fee if it is material.
What is a good payback period for an AI project?
There is no universal answer, and you should be wary of anyone who gives one before they know where your hours go. The useful test is the break-even figure: how many hours each person must save a week to recover the fee and a year of running costs within twelve months. If that number looks easy against the tasks you have measured, the project is worth a closer look; if it looks heroic, it is not.
Can AssumeAI build an ROI model for our firm?
Yes. The AI Strategy Review, from £12,000 over about six weeks, uses surveys, stakeholder interviews and a review of your tech stack to find, cost and prioritise the AI opportunities in your firm. If the review does not surface opportunities worth many times the fee, we keep working until it does. For The Litmus Partnership a two-week review identified 9,672 hours a year across twenty-six initiatives.
Sources
- Government Digital Service, Microsoft 365 Copilot Experiment: Cross-Government Findings Report (2 June 2025)
- HMRC, Rates and thresholds for employers 2026 to 2027 (employer Class 1 National Insurance)
- GOV.UK, Workplace pensions: what you, your employer and the government pay
- GOV.UK, Holiday entitlement
Rates and thresholds are for the 2026 to 2027 tax year. Method and sources last checked on 21 October 2026. Client figures are from our published case studies.