The University of Salford helped law firm Irwin Mitchell build AI and data science skills within its own team. Its case study describes training and shared working practices alongside the technical projects.

Published on 8 September and picked up again in October coverage, the account describes a Knowledge Transfer Partnership involving academic specialists and a dedicated associate.

The work included machine learning for client analysis and resource planning. The team also explored document search using large language models.

Salford reports that the partnership left the firm with skills it could continue to use. That’s the claim I’d want an AI project team to demonstrate at handover.

Ask someone else to repeat the work

Suppose a colleague uses an approved AI tool to help produce accessible versions of course information. They’ve checked the results and the materials are ready to use.

Ask them to record how they did it. Then give those instructions to another colleague and let them try the same process with a suitable document.

You’ll soon find out what the instructions leave out. Perhaps the author knows which sections need extra checking, or remembers a setting they forgot to mention.

Put those details into the guidance. A successful second attempt gives you stronger evidence that the method can survive a change of staff.

Keep the decisions with the prompt

A saved prompt explains what someone asked the tool to do. You also need their reasons for accepting or changing the answer.

Keep an example where the output went wrong. Explain how the colleague spotted the problem and what they did about it.

For course information, that might mean correcting an explanation that became misleading when the language was simplified. Someone reading the finished version would never see the mistake or learn how to catch it.

Record which source material was used and why it was suitable for that tool. The next person needs those boundaries before they start uploading documents.

Give the method an owner

Allow time for a colleague to practise, with someone available to answer questions. Watching a demonstration can leave gaps that only appear when you try the task yourself.

Give the guidance an owner and a review date. If the tool changes, someone needs to check whether the instructions still work.

Salford’s account is a university case study of its own partnership. It gives you an example to examine, with no published financial return you can drop into your own business case.

You can still test the idea on work you already have. Choose one AI-assisted task this month and ask a second colleague to repeat it using the available instructions.

If they have to keep asking the original author what to do, you’ve found the next piece of work: get that knowledge out of one person’s head.

Source