Most AI rollouts don't fail on the technology. They fail because nobody agreed what it was for, who could use it, or what data was allowed near it.
We're getting a version of the same call every week at the moment. Someone in the business has started using an AI tool, it's clearly useful, and now the question is whether to make it official. It's a good problem to have. It's also the point where it's worth slowing down for an afternoon.
These are the five questions we work through with clients. None of them are technical, which is rather the point.
1. What problem are we actually solving?
“Everyone else is doing it” is not a business case, and it won't survive contact with your finance team. Pick two or three specific, recurring tasks that eat time — drafting the same category of email, summarising long documents, first-pass analysis of a spreadsheet. Measure how long they take now. You need something to compare against later.
2. What data is allowed near it?
This is the question that determines almost everything else. Client information held under contract, health data, anything covered by the Privacy Act 2020 — each of these has an answer, and the answer differs depending on which product and which plan you're on.
Get the classification right first, then choose the tool to match. Doing it the other way around is how organisations end up unpicking a rollout six months in.
3. Who administers it?
If the answer is “whoever set it up”, you have a problem waiting to happen. AI tooling needs to sit inside the same identity and access management you use for everything else — single sign-on, MFA, proper joiner-mover-leaver process. If someone leaves on a Friday, their access should go with everything else, not linger because it was a personal signup on a company card.
4. What does good use look like here?
Write it down. A single page is plenty. Which tasks it's encouraged for, which tasks it isn't, what must always be checked by a human before it goes out the door, and who to ask when someone isn't sure.
Staff are not trying to do the wrong thing. They're trying to get their work done, and in the absence of guidance they'll make a reasonable guess. Guidance is cheaper than remediation.
Every organisation we've helped with this has found the policy conversation more valuable than the technology one — because it forced them to describe how the work actually gets done.
5. How will we know if it worked?
Agree the check-in before you start. Ninety days is about right. Come back to the tasks you measured in question one and look at them honestly. Some will have improved. Some won't. Both are useful findings, and the second kind saves you money.
Where to start
If you can answer all five, you're in good shape to proceed. If two or three of them are uncomfortable, that's not a reason to stop — it's simply where the work is. We can help you get through them, and we've done it on our own business first.
