All articles

How to know if your AI change paid off

Most firms can't tell if their AI changes actually paid off. Here is a simple, one-page measurement plan to set up before the work starts.

Key figures

Businesses using AI reported higher productivity, with mostly modest gains
50%
Businesses using AI saw no immediate effect on their costs
64%
Organisations tracking changes in employee productivity
38%

Sources: UK Government, AI adoption survey, 2026 · Deloitte, State of Generative AI in the Enterprise, 2024

Measure the task before you change it. Pick one or two things that matter, like time, errors, cost, or response speed. Decide who will check them. Then, compare the numbers after a set period.

The UK Government's 2026 survey showed half of businesses using AI reported higher productivity, but the gains were modest. Alongside this, 64% saw no immediate effect on costs. Their earlier 2025 survey found only around one in eight AI users had seen higher revenue.

A 2024 Deloitte survey of 2,770 business leaders in 14 countries found 54% were seeking efficiency and productivity gains from generative AI, yet only 38% were tracking changes in employee productivity. You cannot see a gain you never measured.

Seeking productivity gains versus measuring them
  • Seeking efficiency and productivity gains54%
  • Tracking changes in employee productivity38%

Sources: Deloitte, State of Generative AI in the Enterprise, 2024

Measure before you change

A baseline comes first because once the new way is running, your old numbers are gone. For two to four weeks, measure the task exactly as it works today. Record the time taken, how often things go wrong, and how many people touch it.

A 2025 government survey found small businesses tend to decide about AI by informal trial and error, while larger firms more often run formal return on investment analysis. A simple baseline is enough to decide better.

Pick one or two clear measures

Match the measure to the reason for the change. That could be hours saved per week, errors caught, time to reply to a customer, or cost per order. Tracking one or two is enough.

An IBM study of 500 UK leaders in 2025 found the biggest challenges in measuring return were high upfront costs that distort short-term return (37%), difficulty attributing outcomes to AI alone (35%), and a lack of skills (31%). Measuring one task before and after keeps the link clear.

Name an owner. Set dates

A measure with no owner goes unchecked. Name one person to record the numbers, and set review dates at 30 and 90 days. Count the time people spend checking AI output as part of the cost. A saving that needs constant correction is smaller than it looks.

If the numbers do not move by the review date, change your approach or stop.

Your one-page measurement plan

  1. 01The task you are changing.
  2. 02The reason for making the change.
  3. 03One or two measures showing the goal was met.
  4. 04The baseline, measured for two to four weeks first.
  5. 05The person who records the numbers.
  6. 06Set review dates, such as at 30 and 90 days.
  7. 07What you will do if the numbers do not move.

Every recommendation in our AI Readiness Audit comes with a clear way to tell if it worked. We measure every build against it.

Sources

Clinton Onyekwere

Clinton Onyekwere founded Clinton AI in Leeds. We help organisations adopt AI and build lasting internal skills.

Get our newest articles

We only email you when a new piece goes live.

See where AI fits your business.