← All articles

What is the ROI of AI in business in 2026? An honest UK evidence review

AI ROI is real, but it's concentrated. 6% of organisations see ≥5% enterprise EBIT impact; about 60% see no enterprise impact at all. The UK evidence in full.

54%of UK firms now use AI, up from 23% in 2023BCC · March 2026
6%of organisations see ≥5% enterprise EBIT impact from AIMcKinsey · 2025
+71ppproductivity gain expected by current AI users over 12 monthsBCC · March 2026
10–20%function-level cost reductions in eng, IT and manufacturingMcKinsey · 2025

The numbers above tell a hard story: AI is now in most UK businesses, but value at the company level is rare. This piece unpacks where that gap comes from: what AI ROI actually means, what UK firms are seeing on the ground, what it costs to play, how long payback takes, and how to measure any of it honestly.

What AI ROI actually means

AI ROI isn't a single number. The published evidence splits into three categories, and adding them together is where most business cases come unstuck.

Function-level ROI is cost or time saved on a specific workflow: contact-centre handling time, code review throughput, marketing brief turnaround. The easiest to measure, the easiest to win, and the strongest published figures sit here.

Enterprise-level ROI is earnings before interest and tax (EBIT) impact at the company level, traced back to AI use. Much harder. McKinsey's State of AI 2025 found only the AI high-performers (that 6%) report 5% or more EBIT impact attributable to AI.

Hidden ROI is the third category: avoided costs, retention improvements, decisions made better. Real but speculative. Useful as a tie-breaker between options, not as the load-bearing justification for a business case.

What UK firms are seeing on the ground

The most recent BCC and University of Essex study of 668 UK firms shows adoption running well ahead of measurement. Up from 23% in 2023, 25% in 2024 and 35% in 2025: near-tripled in three years.

The interesting cut is by familiarity. Firms already using AI expect a +71pp productivity gain over the next twelve months. Firms still planning it: +46pp. The fence-sitters: +26pp. The "never touching it" crowd: −3pp. The people closest to the technology are the most optimistic about what it can do.

How much does AI cost a UK SME?

The sterling reality, based on published vendor pricing in April 2026.

  • Per-seat productivity AI: £18 to £50 per user per month for ChatGPT Pro, Claude Pro and Microsoft 365 Copilot. The published list price for Microsoft 365 Copilot is £24.70 per user per month.

  • Customer-service or knowledge-base bots on off-the-shelf platforms: £200–£2,000 per month plus implementation. Build cost £5,000–£40,000 depending on how many systems they need to talk to.

  • Bespoke retrieval-augmented generation (RAG) on private data: £25,000–£200,000 to design and deploy, plus £500–£5,000 a month to keep it running.

  • Custom AI agent or fine-tuned model: £50,000–£500,000 to build, with running costs that scale with usage.

  • The people line: McKinsey's guidance is roughly £5 of people investment for every £1 of AI technology spend. Most UK SMEs are spending pennies on the pound.

How long is the payback period?

The published payback ranges, by category:

  1. Per-seat productivity AI: 3 to 9 months in well-adopted teams. Payback hinges almost entirely on whether usage is real. UK firms tracking active-user rates rather than licence counts tend to land at the lower end.

  2. Off-the-shelf bots and assistants: 6 to 18 months. The wide range is the price of integration: clean data and a clear workflow → fast; legacy data with conflicting sources → slow.

  3. Bespoke deployments and custom agents: 12 to 36 months. The function-level win usually shows up in year one. Enterprise EBIT impact, where it appears at all, lands in years two and three.

How should a UK leader measure AI ROI honestly?

Five gates to put in front of any AI investment before it's signed off, and to keep checking after it ships.

  1. Measure outcomes, not adoption. What got done, not what got licensed.

  2. Separate function-level wins from enterprise claims. They aren't the same investment and they don't add together.

  3. Demand objective measurement. If the only evidence is staff saying they feel faster, you don't have evidence.

  4. Budget the people line. Training, role redesign and change management are where function-level wins turn into enterprise impact.

  5. Set a 12-month review gate. Decide upfront what continue, expand or stop looks like, then hold the line.

AI ROI is available in 2026, but it isn't free and it isn't even. It rewards leaders who adopt deliberately, measure objectively, and accept that enterprise-level impact is earned rather than bought.

Fancy seeing where AI could take you?

Let's Talk