The task that never makes it into the report

Thanks for coming back for issue seven. This fortnight that gap showed up clearly in a rollout I got pulled into partway through. When leadership first committed to AI, they did what almost every organisation does first: bought the licences, then bought the generic training that came bundled with the vendor relationship. By the time I was brought in, that part was already finished, and the headline number was already sitting at its target. Underneath, the picture told a different story.

The pattern: measuring the training, not the work

When a leadership team commits to AI, the reporting almost always follows the same path. Licences get bought, a training session gets booked through the same provider, attendance gets logged. Someone asks for progress, and the only number on hand is the one that rollout produced: sessions delivered, seats filled, course completions.

That number is real, but it’s a proxy. It tells you the organisation ran a training program. It tells you nothing about whether a single piece of work is now done differently than it was three months ago.

The Sunday-night test

The distinction that keeps proving itself: don’t ask “has this person had the training.” Ask “what’s the task that makes them sigh on a Sunday evening when they think about Monday.” The report that eats a whole morning. The reconciliation nobody wants to start. The inbox that never empties.

That question can only be answered by looking at the work, not the tool. It produces a different kind of record, not “87 per cent completed the course” but “this specific task, done by this specific person, is now handled a different way, every week.” The first number moves the moment a session is scheduled. The second only moves when something in someone’s actual job has changed.

Option one: the training that didn’t stick

The clearest version of this fortnight came from that same organisation. When leadership committed, procurement did what procurement always does: licences bought, then a generic training program delivered by the AI provider itself, run once across the organisation. The headline number, staff onboarded, hit its target early and stayed there.

Underneath that number, the training wasn’t sticking. People sat through it, then went back to working exactly as before. The provider’s job ended when the session did, so there was no structured follow-up, and whatever landed on the day had no second touchpoint to reinforce it. The organisation had a complete record of who’d attended and no record at all of whether anyone’s Monday had actually changed.

Option two: the coaching that wasn’t in anyone’s spreadsheet

That’s the point I was brought in, after the generic training, not to repeat it. What I ran instead was different in kind: small-group and one-to-one sessions, including coaching the AI lead directly, where people brought a real task, not a hypothetical one, and we worked through it live. Someone would arrive with the report that ate their whole morning, and leave with it genuinely done a different way.

That was adoption in the only sense that matters, and it never appeared in the organisation’s own measurement. It didn’t show up as a completion, a seat, or a login, because it wasn’t structured as a session with an attendance sheet. It existed entirely inside a handful of hours each week, uncounted, unscaled, and entirely dependent on there being time to keep running it.

Why the easy number wins by default

None of this happened because anyone was careless. The seat count is the number that exists on day one, because it’s the number the purchase produces. Licences are invoiced per seat, training is booked per head as part of the same vendor deal, and the measurement inherits the shape of the transaction that paid for it.

The task-level number has to be built on purpose, by someone who goes looking for it after the licences and the generic training are already done. Someone has to ask each team what actually changed, name the specific task, and check back later that it’s still true. Nobody sets that up by accident, and nobody arrives at it by scaling up the first number. A 100 per cent onboarding target and a genuinely changed Monday morning are not the same claim, even when the same rollout produces both, and only one of them was ever being measured.

What the gap costs

The cost shows up quietly. Attendance climbs, budget gets justified, and twelve months later the organisation still can’t name more than a handful of tasks that are genuinely different. The generic training didn’t fail, it produced exactly the number it was built to produce. It just wasn’t the number that describes whether the work changed.

The coaching sessions in this fortnight’s example were producing the real thing the whole time. They just weren’t the thing anyone was counting, and they only started once someone went looking past the training the licences had already paid for.