Safe AI adoption
Make AI safe to use — on whichever platform you use.
We review what your AI tools can reach, clean up permissions, set acceptable-use rules, train staff, and measure whether it's paying off.
Why this stalls
The blocker is almost never the AI itself.
Every platform hits the same three walls, and none are about the model. Two are already live in most organisations.
Your permissions aren't ready
If files have been overshared for years — and in most organisations they have — an assistant will surface them to the wrong colleague. That clean-up is the real project.
Nobody agreed what "allowed" means
Without a policy, staff make individual judgements about what data goes into which tool. Most will be sensible. You only need one to not be.
Licences bought, nothing measured
Seats get assigned, enthusiasm fades by week six, and at renewal nobody can say whether it was worth it. Adoption is a habit problem, not a purchasing one.
An AI assistant can see whatever the person using it can see.
In most organisations that is more than anyone intended — which is why the first job is finding out what yours can reach, not buying more licences.
Already switched on
Most organisations start here, not at the beginning.
If that's you the sequence is the same, but the urgency changes: find out what it can reach, contain anything alarming, then work back to policy and training.
The programme
Five stages, roughly a quarter.
Exposure review
What's already enabled, what it can reach, and what staff are using unofficially.
Clean-up
Sharing links, permissions and retention put right. The stage that decides the rest.
Policy
A short acceptable use policy people actually read: approved tools, what data goes where, what to do when unsure.
Pilot
A small group, real tasks, agreed measures, long enough to see past the novelty.
Rollout and review
Training, internal champions, then a quarterly look at usage, cost and value.
Tools
We are not tied to any vendor.
We're platform-neutral, and we'll say so when the right answer is a tool we don't supply. The right choice depends on where your work happens, what your clients permit, and what you must evidence about data handling. Often it's more than one tool — and sometimes it's using what you already pay for.
What matters more is the configuration behind it: tenant controls, data residency, retention and audit logging.
Common questions
Is our data used to train these models?
On business and enterprise plans, generally no — but it varies by product and plan, and consumer tiers often differ. We check the terms for your specific plan and write it into your policy.
We have contracts or policies that restrict AI use. Can we still do this?
Often yes, with boundaries. We work out where AI is prohibited, permitted or requires disclosure, and build those rules into the policy and the tooling — so it's enforced by configuration, not memory.
How do we measure whether it's working?
We agree measures before the pilot: time on recurring tasks, licence use, and feedback from the group. Some teams see a lot of value and some little — you want to know which before buying everyone a seat.
We turned something on months ago. Is it too late to do this properly?
No, and it's the most common starting point. We begin with an exposure review so you know what has been reachable and for how long, deal with anything urgent, then work back through permissions, policy and training.
Find out whether you're ready for AI.
The review tells you what your permissions would expose and what staff are quietly using today.