Could AI save your team time? Check the evidence and measure it

01

Where does the 122 hours a year figure come from?

It comes from Google's AI Works report, published in April 2025 with the research firm Public First. Participants in three UK training pilots estimated their own time saved, which the report expressed as the equivalent of 122 hours a year. It is a pilot result, not a UK-wide measurement.

AI Works trained workers in three groups: education staff, small and medium business workers recruited through Enterprise Nation, and members of the Community trade union. Three months later, Public First asked them to estimate their time saved in an average week by choosing a band, such as 1–3 hours or 4–6 hours.

The report says workers in each group estimated savings equivalent to over 122 hours a year, 22% above Public First's modelled 100 hours. It gives group figures for education staff (2.9 hours a week, or 109 hours a year) and union members (2.1 hours a week). We could not find the calculation behind a single 122 figure, so treat it as a headline estimate.

02

What do other UK studies say about time saved by AI?

UK evidence is mixed. A cross-government trial found users estimated they saved 26 minutes a day. A Department for Business and Trade evaluation found small savings on written tasks but no robust evidence that they improved productivity.

The government's Microsoft 365 Copilot trial (30 September to 31 December 2024, 20,000 licences) found users estimated saving 26 minutes a day. This was self-reported: respondents chose time ranges and the report used the midpoints. The report could not identify how saved time was spent. DBT evaluated its own 1,000-licence pilot and found small savings on most tasks, the largest on written tasks, while scheduling and image generation took longer. It found no robust evidence that time savings improved productivity, though this was not a key aim.

03

Which kinds of tasks does AI save time on?

AI saves the most time on text-heavy, repetitive tasks where a first draft helps: replying to messages, summarising documents and meetings, answering routine enquiries and moving information into a standard format. A person still reviews the result.

DBT reported the most pronounced savings for drafting documents, summarising research, searching for information and summarising meetings. In the cross-government trial, users reported savings of 24 minutes on drafting documents and 19 on creating presentations. In a 2023 MIT working paper, 444 professionals doing online writing tasks finished faster, and their work was graded higher, when randomly given ChatGPT. We found no UK study that measures a typical weekly saving per category for small businesses.

04

Where does AI save no time, or cost time?

AI costs time when output needs heavy correction, when the task is already quick, when checking takes as long as doing it yourself, or when it creates extra work. Studies found slowdowns in scheduling, image generation and expert coding.

In the DBT pilot, extra time came mainly from Copilot being unable to produce high-quality output, or from tasks done only because the tool made them possible.

In METR's randomised trial, 16 experienced developers working on large open-source projects they contribute to took 19% longer when allowed to use AI tools. They had expected a 24% speed-up and afterwards still believed they had been 20% faster. The authors say it covers one kind of work with early-2025 tools and do not claim it applies to most software work.

05

How much time could AI save me, and how do I measure it honestly?

Estimate it task by task: hours spent now, the share AI can draft, time to check and correct, and weeks worked. Then time the same task with and without AI, count checking time, and compare quality. The net saving is usually smaller than the gross saving.

Savings belong to tasks, not to a business as a whole, so subtract set-up, training, subscription and checking time. Do not rely on feeling faster: in METR's trial, developers who were slower believed they had been faster. DBT adjusted its diary estimates for outputs never used and for tasks done only because the tool existed.

1. 01
   Pick one task

   Choose a task you do at least weekly, such as replying to quote requests.

2. 02
   Record a baseline

   Time ten jobs the usual way. Note the quality problems you normally see.

3. 03
   Trial it with AIFull step in the download ↓
4. 04
   Count checking and fixingFull step in the download ↓
5. 05
   Compare qualityFull step in the download ↓
6. 06
   Calculate and decideFull step in the download ↓

06

Do saved hours turn into more productivity or profit?

Not automatically. Saved time only adds value when it is used for something worth more, such as selling, serving customers or finishing work sooner. DBT found no robust evidence that time savings improved productivity, and the government trial could not tell how saved time was spent.

Be careful with money figures. A saving in pounds usually multiplies hours by an hourly rate, and both are estimates, so treat it as an upper bound.

Gains also vary by person. In the customer-support study, access to an AI assistant raised issues resolved per hour by 14% on average, with a 34% improvement for novice and low-skilled workers and minimal impact on experienced and highly skilled workers.

07

What data protection checks should come before chasing time savings?

Check what data goes in before you start. The NCSC advises against putting sensitive information into public AI tools, and UK GDPR applies if you use personal data. Choose tasks that need no personal or confidential detail, then widen use once the terms are checked.

The NCSC says queries to public AI tools are visible to the provider, which stores them and is likely to use them to develop its services. The Information Commissioner's Office (ICO) says a business using a generative AI tool to draft client emails needs a lawful basis for the personal data involved.

Google's report recommends giving workers explicit permission to use AI and setting clear policies on AI tools.

08

How do I start reclaiming time this week?

Choose two repetitive text tasks, set the rules for what may be entered, measure a baseline, then trial AI for two weeks. Keep what saves net time and stop what does not. If a task needs your systems or customer data, get advice first.

Good first candidates are routine email replies, meeting summaries and summaries of long documents. They are text-based, frequent and easy to check. Leave customer decisions, sensitive data and professional advice until you have a track record.

Reeve Consult offers AI consultancy that looks at a specific workflow, the information it needs and where a person should check the output. It does not promise a time saving, and the steps above work without us.