AI ROI CALCULATOR
FOR UK TEAMS
How many hours — and how much money — would your team actually get back from AI? Built on published UK data, not vendor hype.
The We Call Shotgun AI ROI Calculator estimates the annual return of AI adoption for UK teams. It combines ONS salary and working-hours data, HMRC employer costs (National Insurance, pension) and measured productivity research — the Federal Reserve Bank of St. Louis, Google's UK AI Works pilots, BCG's AI at Work study — into a conservative-to-optimistic value range, never a single inflated number. For a typical 25-person UK team on median salaries, expected value is roughly £16,000 a year when adoption is left to chance and £38,700 with training-led adoption. Free, instant, no email gate.
YOUR TEAM
Five inputs. Every other assumption is fixed to published research — see exactly how it works.
People whose day-to-day work involves writing, analysis, admin, reporting or communication.
Default is the UK full-time median, £39,039 (ONS, April 2025). Or pick a role:
Your starting point — the share of the team who are regular AI users today, self-taught. Anchored to the St. Louis Fed (37% of workers use AI at work) and BCG (untrained frontline use stalls near 51%). Training doubles wherever you start; it doesn't teleport you to the ceiling.
Advanced: adjust costs
Defaults: £20/seat/month ≈ ChatGPT Team / Microsoft Copilot business tiers; £250/person ≈ a half-day group workshop, pro-rated.
Expected annual value
£16,000
if your team adopts AI without structured training
Hours reclaimed / year
1,010
≈ 0.6 full-time roles
Regular AI users
10 of 25
40% adoption
Expected ROI
2.7x
on £6,000 of year-one costs
Payback
4.5 months
>18 months? Rework the plan.
+£22,700/year with structured training. Training roughly doubles regular users (40% → 79%) and trained users save more hours per week. That is £3.64 back for every £1 of training in year one — and the uplift repeats every year, with the training paid for once.
Estimates use fixed, research-backed coefficients — methodology and sources below. Not a guarantee; see FAQ.
WHAT "WITHOUT TRAINING" ACTUALLY MEANS
It does not mean nobody touches AI. Nobody can stop that now. It means nobody is taught — people teach themselves, in their own time, on their own workflows, with nobody checking the result. That still produces value, and this calculator counts it. It just produces far less than most leaders assume, and it stops climbing early.
How your team learns AI when you don't train them
1. Colleague osmosis
One or two enthusiasts get good, and show the people sitting nearest them. It spreads by proximity and personality, not by role or need — so the functions with the most repetitive work are often the last to hear about it.
2. Trial, error and YouTube
Free tiers, generic prompt lists, a tutorial at the weekend. This reliably produces better email drafts and faster first drafts. It rarely reaches the work that actually costs you money — the reporting pack, the tender response, the case triage — because nobody has mapped AI onto those workflows.
3. Shadow AI
Personal accounts, personal phones, company data pasted into consumer tools. Real hours are saved here — and we count them — but they are invisible to you, unrepeatable across the team, and a live UK GDPR exposure.
What self-teaching gets you
Real, measured value — from a minority of your team, on the easy half of their work. The national evidence for self-directed use is 1.1–2.2 hours per week per regular user (the Danish study of 25,000 workers; the St. Louis Fed), against 2.7 hours in structured programmes. And only 15–55% of the team ever become regular users on their own.
At a standing start that is close to break-even: a 25-person team on our defaults recovers about £6,000 a year against £6,000 of licences — roughly 1.0x. You have bought the tools and received the productivity of a rounding error.
Why it plateaus instead of compounding
- →The ceiling is measured, not theoretical. BCG found untrained frontline use stalls around 51% and never reaches the 79% that trained teams hit.
- →Skills stay generic. Self-taught use lands on drafting and summarising — not on the workflows where your margin actually sits. That is the difference workflow-first training exists to close.
- →Week three is where it dies. The novelty fades before the habit forms — what we call Painful Tuesday. Nobody is there to push through it.
- →Nothing is measured, so nothing is defended. Licences without evidence of return are the first line cut at budget time — activation without adoption.
Your starting point changes everything
The three options in the calculator are where your team is today, self-taught. Training roughly doubles that share — it doesn't teleport a standing start to the ceiling. Figures below: a 25-person team on the UK median salary, expected scenario, tools at £20/seat/month and training at £250/person.
| Where you are today | Regular users (self-taught → trained) | Value self-taught | Value with training | Uplift | Return per £1 of training |
|---|---|---|---|---|---|
| Barely used — no tools, no habit | 15% → 30% | £6,000 | £14,700 | +£8,700 | £1.39 |
| Some people experiment on their own | 40% → 79% | £16,000 | £38,700 | +£22,700 | £3.64 |
| Regular use by part of the team already | 55% → 79% | £21,900 | £38,700 | +£16,700 | £2.68 |
Two things worth reading off that table. The middle row pays back best — a team already curious about AI is the cheapest to convert, because you are adding structure to existing momentum rather than starting the engine. And a team at a standing start needs training the most in absolute terms, but returns least per pound in year one: the honest advice there is to start with a narrow pilot on two workflows rather than a company-wide rollout. In every row the uplift then repeats each year, while the training is paid for once.
One honest caveat about the ROI multiple
If your team is already at regular use, you will notice the headline ROI multiple barely moves when you add training — in the bottom row it reads 3.7x self-taught against 3.2x trained in year one. That is arithmetic, not a verdict: a multiple falls when you invest more, even though you earn more. The trained team in that row takes home £16,700 more every year, and from year two — training already paid for — runs at 6.4x against 3.7x. When two options differ in size, compare the money and the return on the extra spend, not the ratio.
HOW THE CALCULATOR WORKS
Most AI ROI calculators let you type in your own "hours saved", multiply by 52 weeks and assume everyone adopts. Ours doesn't. Every coefficient below is fixed to a published source, and the three scenarios run in parallel — you always see a range, never one flattering number.
1. Fully-loaded hourly cost — derived, not asserted
hourly cost = (salary × 1.25) ÷ 1,700 hours
UK employees cost more than their salary: employer National Insurance is 15% above £5,000 (2025/26) and auto-enrolment pension adds a minimum 3% — roughly 16% of extra mandatory cost at the median wage, before equipment and overheads. We use a conservative 1.25x multiplier from the published 1.25–1.4x "true cost of employment" range. The 1,700 annual hours come from ONS actual working hours (36.7/week) across ~46.4 working weeks after statutory holiday. A £39,000 salary works out at £28.68 per working hour.
2. Adoption — because not everyone will use it
regular users = team × adoption rate (15–55% by current usage; ×2 with training, capped at 79%)
Around 37% of working-age adults use generative AI at work (St. Louis Fed, 2025) and BCG found frontline regular use stalls around 51%. So without training we count 20–55% of your team as regular users, depending on where you are today. Google's UK AI Works pilots found a few hours of training roughly doubled sustained adoption, and BCG measured 79% regular use among employees with 5+ hours of training — that 79% is our hard cap. We never model 100% adoption.
3. Hours saved — matched to the evidence for your scenario
hours reclaimed = regular users × hrs/week (untrained: 1.1 | 2.2 | 2.7 · trained: 2.2 | 2.7 | 5.0) × 46 weeks
This is the number every other calculator lets you guess — and where many quietly cheat, crediting untrained rollouts with savings measured in structured programmes. We match the study to the scenario. Without structured training (self-directed use): conservative 1.1 hrs/week — the ~3% of work hours measured across 25,000 Danish workers using AI in the wild (Humlum & Vestergaard); expected 2.2 hrs/week — the St. Louis Fed's measured survey; optimistic 2.7 — your enthusiasts matching trained-pilot rates on their own. With structured training: conservative 2.2; expected 2.7 hrs/week — Google's UK AI Works pilots, 122 hours a year measured alongside training; optimistic 5.0 hrs/week — what professionals told Thomson Reuters they expect, a prediction, so it's labelled optimistic. And we use 46 working weeks, not 52: UK statutory holiday is 5.6 weeks, and calculators that multiply by 52 inflate results by ~13%.
4. Value conversion — saved time isn't automatically money
gross value = hours reclaimed × (40% | 55% | 70%) × hourly cost
The most honest step, and the one competitors skip. A national study of 25,000 Danish workers (Humlum & Vestergaard) found AI time savings produced almost no measurable wage or output change where organisations didn't deliberately capture the time. Adecco's survey of 35,000 workers found ~79% do reinvest saved time into work. We count 40–70% of reclaimed hours as realised value — never 100%.
5. Costs, ROI and payback — because tools and training aren't free
ROI = gross value ÷ (tools £/seat × 12 × team + training £/person × team)
We subtract what you'd actually spend: business-tier AI seats (default £20/seat/month) and one-off structured training (default £250/person, editable). Payback is the months of value needed to cover year-one costs. Because training is a one-off, its cost lands entirely in year one — from year two only tool seats recur, which is why the trained scenario's steady-state ROI roughly doubles after the first year. Our rule of thumb, borrowed from the more honest end of the industry: if payback exceeds 18 months, rework the plan before rolling out.
What we deliberately don't do
- ❌ No editable "hours saved" field — that's where wishful thinking enters every other calculator.
- ❌ No trained-programme savings in the untrained scenario — each band only uses studies matching that condition.
- ❌ No 100% adoption — the measured ceiling for trained teams is 79% (BCG).
- ❌ No 52-week years — UK statutory holiday exists.
- ❌ No single point estimate — you get conservative, expected and optimistic, always.
- ❌ No email gate — results are on screen, and the shareable link is free.
Known simplifications: the 1.25x on-cost is flat (small employers with Employment Allowance pay slightly less NI); results are annualised steady-state rather than ramped month-by-month; role presets change salary, not hours saved. Role-level research (customer support +15% productivity, writing tasks 40% faster, developers up to 55% faster) is cited in the data table but kept out of the arithmetic to keep the model auditable.
WHAT UK TEAMS CAN EXPECT FROM AI ADOPTION
Pre-computed results at the calculator's defaults — UK median salary (£39,000), team already experimenting with AI, structured training included, business-tier tools at £20/seat/month, training at £250/person. Value is the realised annual £ band after adoption, value-conversion and UK working-week adjustments.
| Team size | Hours reclaimed / year (expected) | Conservative | Expected value | Optimistic | Year-one costs | Expected ROI |
|---|---|---|---|---|---|---|
| 10 people | ~980 hrs | £9,200 | £15,500 | £36,500 | £4,900 | 3.2x |
| 25 people | ~2,450 hrs | £22,900 | £38,700 | £91,200 | £12,250 | 3.2x |
| 50 people | ~4,900 hrs | £45,900 | £77,400 | £182,400 | £24,500 | 3.2x |
| 100 people | ~9,800 hrs | £91,700 | £154,800 | £364,700 | £49,000 | 3.2x |
Without structured training the same teams capture far less — a 25-person team drops from an expected £38,700 to roughly £16,000 a year, because adoption stalls at ~40% instead of 79% and self-directed users save fewer hours (the in-the-wild evidence measures 1.1–2.2 hrs/week, versus 2.7 in trained programmes). Expected ROI falls from 3.2x to 2.7x and payback stretches from 3.8 to 4.5 months — and since training is a one-off cost, the trained scenario runs at ~6.4x from year two. That gap, not the tools, is where most UK AI ROI is won or lost.
THE DATA BEHIND THIS CALCULATOR
Every number in the model traces to one of these published sources. No "our client benchmarks", no vendor decks presented as science.
| Statistic | Figure | Source | Date |
|---|---|---|---|
| UK full-time median gross annual salary | £39,039 | ONS, Annual Survey of Hours and Earnings | Apr 2025 |
| Employer National Insurance rate | 15% above £5,000 | HMRC, rates and thresholds for employers 2025/26 | 2025/26 |
| Minimum employer pension contribution | 3% of qualifying earnings | GOV.UK, workplace pensions | 2025/26 |
| UK statutory paid holiday | 5.6 weeks/year | GOV.UK, holiday entitlement | Current |
| Time saved by generative AI users (measured) | 5.4% of work hours ≈ 2.2 hrs/week | Federal Reserve Bank of St. Louis (Bick, Blandin & Deming) | Feb 2025 |
| UK workers' time saved with AI + training | 122 hours/year ≈ 2.7 hrs/week | Google / Public First, AI Works UK pilots | 2025 |
| Professionals' predicted AI time saving | 5 hrs/week | Thomson Reuters, Future of Professionals | 2025 |
| Regular AI use with 5+ hours of training | 79% of employees | BCG, AI at Work 2025 (10,635 employees) | Jun 2025 |
| Organisations reporting significant AI ROI, with vs without mature upskilling | 42% vs 21% | DataCamp / YouGov, State of Data & AI Literacy | 2026 |
| Workers reinvesting AI-saved time into work | ~79% | The Adecco Group, Global Workforce of the Future (35,000 workers) | Oct 2024 |
| AI time savings without organisational capture | ~3% of hours; near-zero earnings impact | Humlum & Vestergaard, NBER (25,000 Danish workers) | Apr 2025 |
| Enterprise GenAI pilots with no measurable P&L return | 95% | MIT NANDA, The GenAI Divide: State of AI in Business | Jul 2025 |
| Writing tasks completed faster with AI (RCT) | 40% faster, 18% higher quality | Noy & Zhang, Science | Jul 2023 |
| UK businesses using AI | 29% (49% of firms with 250+ staff) | ONS, Business Insights and Conditions Survey | Jun 2026 |
WHY MOST AI ROI NEVER SHOWS UP
The calculator shows what's available. The research is blunt about why most companies don't collect it.
95%
of enterprise GenAI pilots show no measurable P&L impact — the barrier is missing training, workflow integration and measurement, not the technology (MIT, 2025).
~40%
is where team adoption stalls when tools are bought but nobody is trained — licences alone don't change how people work (St. Louis Fed; BCG).
2x
structured training roughly doubles sustained adoption (Google AI Works, UK) — and doubles the share of organisations reporting significant ROI (DataCamp/YouGov, 2026).
That's the entire We Call Shotgun thesis in three numbers: the gap between AI's potential and most companies' results is an execution gap — and it closes with workflow-first training, a right-sized usage policy, and adoption someone actually owns. It's what we do for UK teams from London to Edinburgh, starting at £3,500.
Book a Free 20-Minute CallFREQUENTLY ASKED QUESTIONS
Straight answers about the model, the maths and what the output actually means.
How does this AI ROI calculator estimate hours saved?
We fix hours saved to published research rather than letting you type a guess — and we match the study to the scenario. Without structured training (self-directed use): 1.1 hours per user per week conservative (≈3% of work hours, measured across 25,000 Danish workers using AI in the wild) and 2.2 expected (the Federal Reserve Bank of St. Louis's measured survey). With structured training: 2.2 conservative, 2.7 expected (Google's UK AI Works pilots — 122 hours a year, measured alongside training) and 5.0 optimistic (Thomson Reuters' 2025 Future of Professionals survey — a prediction, so we label it optimistic).
What does "without structured training" mean — will my team learn AI on their own?
Partly, yes — and the calculator counts it. "Without training" does not mean nobody touches AI; it means nobody is taught. In practice AI still arrives through three routes: colleague osmosis (an enthusiast shows the people sitting nearest them), trial and error on free tiers and YouTube tutorials, and shadow AI on personal accounts. That produces genuine savings, which is why the untrained scenario is not zero. But it reaches only 15–55% of the team, lands on generic drafting rather than the workflows where your margin sits, and stalls near BCG's measured 51% ceiling. At a standing start it is close to break-even: about £6,000 of value against £6,000 of licences for a 25-person team.
How much does my starting point change the result?
A great deal, and not in the way most people expect. Training roughly doubles the share of your team who become regular users — it does not teleport a standing start to the ceiling. For a 25-person team: from "barely used" (15% → 30% regular users) training adds about £8,700 a year, or £1.39 per £1 spent; from "some people experiment" (40% → 79%) it adds £22,700, or £3.64 per £1; from "regular use already" (55% → 79%) it adds £16,700, or £2.68 per £1. The team already experimenting converts most cheaply, because you are adding structure to existing momentum. A team at a standing start is usually better served by a narrow pilot on two workflows than a company-wide rollout.
Why does the calculator show a range instead of one number?
Because a single ROI figure is false precision. Measured AI savings vary from roughly 3% of work hours (a Danish national study of 25,000 workers) to 5+ hours a week (professional-services surveys), and outcomes depend heavily on adoption and implementation. We show conservative, expected and optimistic scenarios so you can plan against the low end and aim for the high end.
Why doesn't the calculator assume everyone on my team will use AI?
Because they won't. Around 37% of working-age adults use generative AI at work (St. Louis Fed, 2025), and BCG found frontline regular use stalls around 51%. Even with 5+ hours of structured training, regular use reaches about 79% — which is the cap we apply. Calculators that assume 100% adoption overstate value roughly 2x.
What difference does AI training actually make to ROI?
Three well-evidenced effects. Google's UK AI Works pilots found a few hours of training roughly doubled sustained adoption — we model that as doubling your adoption rate, capped at 79% (BCG's measured ceiling for trained employees). Trained users also save more hours: the in-the-wild evidence for self-directed use measures 1.1–2.2 hours a week, versus 2.7 in structured programmes. And because training is a one-off cost, its ROI improves after year one — at defaults, a trained 25-person team runs at roughly 3.2x in year one and ~6.4x from year two. A 2026 DataCamp/YouGov survey corroborates the pattern: organisations with mature AI upskilling were twice as likely (42% vs 21%) to report significant ROI.
Why is the cost per hour higher than the salary suggests?
UK employees cost more than their salary: employer National Insurance (15% above £5,000 from April 2025) and the 3% minimum auto-enrolment pension add roughly 16% at the median wage, before equipment and overheads. We use a conservative 1.25x fully-loaded multiplier, so a £39,000 salary is about £28.70 per working hour across roughly 1,700 actual annual working hours.
Does time saved by AI automatically become financial value?
No — that is the biggest honest caveat in the research. A study of 25,000 Danish workers found AI time savings produced almost no wage or output change where organisations didn't deliberately capture the time, while Adecco found around 79% of workers do reinvest saved time into work. We therefore count only 40–70% of saved hours as realised value, never 100%.
What is a good ROI for AI adoption in 2026?
For a typical UK team paying for business-tier AI tools and structured training, our model puts expected ROI around 3x in year one — rising to roughly 6x from year two once the one-off training cost is behind you — with payback inside four months. IDC (in a Microsoft-sponsored study) reports organisations averaging $3.7 of return per $1 invested in generative AI. A useful rule of thumb: if your projected payback is longer than 18 months, rework the plan before rolling out.
Is this calculator's estimate guaranteed?
No. MIT research found 95% of enterprise AI pilots showed no measurable P&L impact — mostly due to missing training, workflow integration and measurement, not the technology. Treat the output as the value available if you execute adoption well; the gap between the "without training" and "with training" scenarios is precisely the execution gap.
TURN THE ESTIMATE INTO A NUMBER YOU CAN DEFEND
Book a free 20-minute call. We'll pressure-test your inputs, tell you honestly where AI pays back first in your business — and where it won't.
Book Your Free AI Audit