Use the register to answer practical questions

Start with a shared document or a short table, not new software. Its job is to answer a few questions that otherwise become difficult six months later: which tool was used, for what task, what information went in, what came out and who owns the workflow.

This is not a list of every time somebody asks an approved tool to improve a sentence. Record the recurring uses that affect customer communication, operational decisions, business records or how the team spends time. If a colleague could reasonably need to repeat, review or stop the use, it deserves a line in the register.

The UK Government’s AI Management Essentials guidance was designed as an accessible starting point for organisations of different sizes, particularly SMEs. The useful lesson for a lean team is proportion: build enough visibility to make sound decisions, then add detail only when the use requires it.

Start with five fields that change a decision

Keep the first version short enough that it is actually used. Each entry should describe the workflow in ordinary business language, rather than trying to document the technology in depth. A new starter should be able to understand what the team uses the tool for and where a person takes responsibility.

  • Purpose: the specific job it helps with, such as drafting a first response or summarising internal meeting notes.
  • Tool and owner: the approved service and the person accountable for keeping the use appropriate.
  • Information boundary: what may be entered, what must not be entered and where the source material comes from.
  • Human check: who reviews the output, what they check and when they can override or stop it.
  • Review date: the point at which the team will decide whether to keep, change, expand or retire the use.

Describe the work, not the marketing promise

An entry such as ‘using AI for sales’ is too broad to govern. ‘The sales coordinator uses an approved tool to turn their own notes into a first draft of a follow-up email; the account lead reviews names, commitments and pricing before sending’ is specific enough to review.

That description makes the real trade-offs visible. It shows that the tool is assisting a draft, not deciding what the business promises. It identifies the source of the information and the person who checks the result. It also makes it easier to spot when a seemingly low-risk use has drifted into something more consequential.

Avoid claiming a tool is accurate, compliant or safe in the abstract. Those are not properties a small business can assume from a supplier’s marketing. The relevant question is whether this particular use, with this information and this review step, is suitable for the work in front of you.

Good AI governance starts with a workflow that a colleague can explain without using the word ‘AI’.

Give higher-risk uses a bigger pause

Some entries need more than five fields. Pause for a fuller assessment if the use handles sensitive or substantial volumes of personal information, produces a recommendation about a person, affects employment or credit decisions, or could cause real harm if it is wrong. In those cases, bring in appropriate data protection, legal or specialist advice before treating the workflow as business as usual.

The ICO’s guidance on AI accountability says organisations should identify and record the degree of human involvement in a decision process, and make sure any review can be meaningful. A person who simply clicks approve without the context, authority or time to challenge an output is not a useful safeguard.

Where personal data is involved, check the use at the design stage rather than after it is embedded. The ICO’s privacy-by-design guidance emphasises building data protection into processing and business practices throughout their lifecycle. In practical terms, decide what data is needed before connecting a new tool to a customer record or shared drive.

Make review part of ordinary management

Bring the register into a meeting that already happens: a monthly operations check, an owner’s review or a team improvement session. Read only the entries that have changed, produced an issue or reached their review date. That keeps the conversation attached to real work rather than turning it into a separate compliance ritual.

Ask a small set of questions: Is the workflow still useful? Has the information boundary changed? Are people performing the stated check in practice? Have customer, staff or supplier concerns emerged? Is the owner still the right person? A clear ‘no longer used’ decision is valuable too; retire the entry and remove access where appropriate.

If several teams start using the same tool in different ways, do not force them into one vague policy line. Give each meaningful workflow its own entry, then look for a shared rule that will actually help: approved accounts, a common data boundary, a review expectation or an escalation route.

Build the first version in an hour

Invite the people closest to the current experiments and list the two or three uses that are already recurring. Write one entry for each, agree the owner and put a review date in the diary. Do not wait for perfect knowledge. A small, honest register is more useful than a grand policy that cannot describe what is happening now.

Then make one practical improvement from what you find. You might replace a shared account, remove customer information from an early test, make a reviewer’s checklist explicit or stop a workflow that has created more editing than value. The register earns its place when it leads to one clearer decision.

As AI use grows, the document may become a more formal governance record. For now, its value is simpler: it helps a small business stay deliberate about a fast-moving set of tools and keeps responsibility with the people who understand the work.

Sources and further guidance

This article is practical guidance, not legal advice. The resources below provide current UK guidance on foundational AI management and data protection. Seek advice suited to your sector and use case before expanding a higher-risk use.

Make the next step useful.

Bring us the workaround, the repeated report or the AI idea that has not found an owner yet.

Discuss a challenge