The Second Custodian: Making AI Adoption Survive the Champion's Departure

In many UK SMEs, the arrival of an AI tool is not a formal IT project. It begins with an employee who sees a useful application, tests a few prompts, and gradually builds a workflow around the results. By the time the tool is part of daily operations, the same person may be the only member of staff who knows why a particular prompt works, which outputs need checking, and how to reach the vendor when something breaks. That is a key-person risk in plain sight.

The risk is not hypothetical. Staff turnover, parental leave, promotion and restructuring are ordinary features of business life. When the person who introduced and refined an AI tool moves on, the tool may remain switched on while the knowledge that made it useful leaves the building. The organisation is then left with a subscription, a login and a process that nobody fully understands. As UK firms enter the final quarter of 2026 with more AI tools in daily use than formal owners, the question is not whether the champion will leave, but whether the business can continue without them.

Why AI adoption attracts single-person ownership

AI tools are easy to trial at an individual level. A subscription can be bought on a card, a browser extension installed, or a feature switched on inside an existing platform. This low barrier to entry is one of the technology's strengths, but it also means adoption often bypasses the usual controls applied to finance systems, CRM platforms or HR software.

In a lean UK SME, the person who champions the tool is often the person who uses it most. They develop prompt libraries in personal notes, save outputs in local folders and learn the vendor's support route by experience. None of this is careless. It is a natural response to a busy role and a tool that appears to work. The problem is that the operational knowledge remains personal rather than corporate.

What actually leaves when the champion leaves

A leaver's notice period rarely captures the full dependency. Access is the visible part. The harder loss is the tacit knowledge: the prompt that reliably produces a usable first draft, the manual check applied before an output is sent to a client, the workaround for a formatting issue, the name of the account manager who responds quickly.

There is also a governance gap. If the tool processes personal data, the business needs to know what data is entered, where it is stored and which contract terms apply. If the tool supports a regulated activity, the decision log matters as much as the output. When ownership sits with one enthusiastic user, these records may be incomplete or held in an inbox. That is a continuity issue, not merely an IT administration task.

A handover checklist for the second custodian

The practical response is to nominate a second custodian before a departure is announced. The second custodian does not need to be an AI specialist. They need the authority and information to keep the tool running, or to pause it safely, while the business decides what happens next.

1. Name an owner and a deputy. Record the operational owner of the AI tool in the same way as any other business system. The owner is accountable for access, performance and renewal. The deputy is trained before they are needed, not after a resignation.

2. Recover administrative access. Identify the account owner, billing contact and any administrator roles. Confirm that access is held in a corporate account rather than a personal email address. Check multi-factor authentication and recovery routes, and remove the leaver's access promptly.

3. Document the prompt library. Move prompts out of personal notes and into a shared, version-controlled location. Record the purpose of each prompt, the model or tool version it was tested on, and any known limitations. A prompt without context is little more than text.

4. Capture workarounds and checks. List the manual steps that sit around the tool: the fields that must be verified, the outputs that require human review, the exceptions that are handled offline. These are often the most valuable part of the process and the least likely to be written down.

5. Record vendor contacts and contract terms. Keep the account manager, support portal, renewal date, notice period and key contractual obligations in one place. If the tool is embedded in another platform, record that dependency too.

6. Maintain a decision log. Note why the tool was adopted, what alternatives were considered, which use cases were approved and which were rejected. This helps the successor avoid repeating old evaluations and supports consistent decisions about data protection and risk.

7. Test continuity before it is needed. Ask the deputy to run a routine task using only the documented material. If they cannot, the documentation is incomplete. A short simulation is more useful than a folder of screenshots.

Making continuity part of normal operations

A handover checklist is valuable at the point of departure, but it works best as a routine control. UK SMEs can add AI tools to the same quarterly review that covers software access, contract renewals and business continuity. Leavers' processes should include AI accounts alongside email, cloud storage and finance systems. Procurement should ask who will own the tool after go-live and how knowledge will be transferred.

This does not require a large governance function. It requires the recognition that an AI tool used in client work, HR or finance is an operational asset. Assets need owners, records and succession arrangements. The chatbot may stay switched on after the champion leaves, but the business should not depend on the champion's memory to keep it useful.

The most durable AI adoption in a UK SME is not the one with the most ambitious prompt. It is the one that another employee can pick up, understand and run when the original enthusiast moves on. Planning for that second custodian is a modest administrative step with a disproportionate effect on continuity.