The Coach's Role in AI Transformation

Organisations spend serious money getting AI adoption wrong in a very specific order: licences first, security review second, infrastructure third, and then, almost as an afterthought, a training day. The figure quoted most often for where this lands is a 15% adoption plateau — I’ve seen it repeated enough to treat it as directionally true, though I’ve never seen the underlying study named, so take it as a shape rather than a settled number. What’s more reliable is the pattern underneath it: the rollout stalls not because people don’t understand the tool, but because nobody on the project owns the part of the job that has nothing to do with the tool.

That gap used to be survivable on a slower clock. It isn’t now, and it’s gotten more complicated, not less. A company that has done the governance work properly no longer hands everyone one tool; it hands them a menu, a consumer chatbot for public information, a business-tier cloud plan for internal work, an EU-hosted option for confidential files, maybe a local model for the narrow, sensitive tasks that shouldn’t leave the building at all. That’s the right policy. It’s also, from the seat of the person actually doing the work, four tools instead of one, each with a rule about which document is allowed to touch it. Nobody adopts that willingly from a slide explaining the sensitivity tiers. They adopt it because someone sits with them while they check, for the tenth time, which of the four boxes a client contract falls into, until checking becomes reflex instead of a stop-and-ask moment.

Knowing how something works and using it as a habit are two different states, and the distance between them is exactly where a coach operates, not an engineer, and not a policy document either. Ten years running a 1:1 coaching practice, guiding people through career changes and the transitions nobody schedules a training day for, taught me something that transfers directly here: people rarely resist change because the logic of it escapes them. They resist because using the new way, in front of colleagues, exposes exactly how much they don’t yet know how to do — and that threatens something more specific than productivity. It threatens the sense of being the person in the room who’s competent. People rarely resist the tool itself; they resist the reason for it that nobody said out loud (Playbook 2, Awareness & Desire), and “compliance told us to” is not a reason, it’s an order.

An engineer solves for the tool working. A coach solves for a specific person being willing to look incompetent in front of their team long enough to stop being incompetent, the knowing-doing gap that gets closed through repetition with feedback rather than a single session (Playbook 4, Building Ability). Most AI rollouts only staff the first job, because it’s the one with a line item and a deliverable. Coaching doesn’t produce a system you can point to when the budget gets reviewed, so it gets treated as optional even in the rollouts that will, on paper, tell you people are the priority.

What that work actually looks like has expanded since I last wrote about this. It’s still working with the resistance instead of arguing it down, since arguing rarely moves someone whose objection was never really about logic. It’s still building enough psychological safety that trying the tool badly, in view of a manager, doesn’t read as an input to a performance review. But now it also includes translating a data sensitivity map into something that doesn’t feel like a compliance cage: not “you can’t use ChatGPT,” which sounds like a punishment for something everyone was already doing, but “here’s exactly which tool covers this file, and here’s why it’s genuinely fine to use it,” mapped to the task this specific person already hates doing (Playbook 3, Role-Based Enablement). Translated badly, governance reads as a wall of new rules layered on top of a tool people already found confusing. Translated well, by someone whose job is the translation, it reads as permission, finally, to stop guessing.

And then staying present through the stretch where the person is visibly worse at their job than they were a month ago, the exact window where adoption quietly regresses without a deliberate hold (Playbook 5, Reinforcement). That window got longer, not shorter, the day “the AI tool” stopped meaning one thing.

The open question isn’t whether this role is useful, it’s who is meant to hold it when most companies don’t have an AI Officer, a DPO, or an ethics committee, let alone a coach. Name the role explicitly even without a department to put it in, rather than letting it default to whoever is already overloaded (Governance Playbook 5, Roles Without a Department). Whether coaching becomes standard practice, or AI transformation keeps quietly failing at the same human step every technology disruption before it also failed at, is the question nobody funding these rollouts seems to be asking yet, and it’s getting more expensive to keep not asking it.