Consultants are selling AI strategy like it’s a dashboard with three sliders. Three sliders that you simply move side-to-side to adjust the intensity.
Very neat. Very executive. Very “easy, we can do that”.
After two years doing this in the public sector, I can tell you it’s not easy. It’s much closer to 3-D chess, and the big problem is the Board only ever sees one chess board.
I’ve drawn similar myself, by the way. Before I knew any better. Three tidy axes. The sort of thing that looks authoritative on slide 14… slightly embarrassing by slide 17.
- Speed v readiness
- Centralise v devolve
- Efficiency v trust
Find your balance on each axis, the diagram implies, and congratulations: you’ve got a strategy you can implement.
Lovely diagram. Also a lie.
A useful lie, perhaps. But still a lie.
Because the moment you try to move the dial, the axes turn out not to be independent. They’re wired together. Tug one piece and the others move. Human systems are annoyingly like that. That’s because they have real people in them.
Chess board one: speed v readiness
This is the board your directors can actually see.
“AI by Q3.” “Where are we on the AI thing?” “Can we accelerate this?”
Perfectly reasonable questions… from where they’re sitting.
The problem is that speed is being demanded of a workforce that is, structurally not ready. That doesn’t mean they are lazy, or they aren’t interested in AI. It means the organisation hasn’t prepared them well enough, if at all.
In an NHS trust or a housing association, “move fast” is not a Silicon Valley growth-hack mantra to be implemented. It collides with regulation, safeguarding, caseloads, legacy systems, and people who have excellent reasons to distrust the last thing that was rolled out to “save them time” but didn’t.
Readiness is slow on purpose. It’s a feature, not a bug.
It’s not resistance. It’s memory.
And memory, inconveniently for AI strategy consultants, is an important feature of human beings on the receiving end.
Chess board two: centralise v devolve
Who actually owns AI in your organisation. The centre, or the service areas?
Most commentary treats that as a governance abstraction. A box on an org chart. A nice little arrow diagram.
It isn’t. It’s plumbing.
“Who is allowed to see what?” is the same question whether you’re drawing reporting lines or configuring permission-aware retrieval for RAG AI.
Centralise everything and you get speed. But you also brittleness and resentment. Devolve everything and you get forty pilots, thirty-nine of which die quietly in a cupboard somewhere. But not until they’ve relieved you of large quantities of cash you’d rather have back.
This is where people mistake control for coordination. They are not the same thing.
Build versus buy is often just this same decision wearing a different hat and pretending to be more strategic than it is.
Chess board three: efficiency v trust
This is the consequential game.
And, naturally, it’s the board that nobody upstairs is looking at.
Efficiency is the business case. The cashable saving that got the thing funded. But trust is what decides whether the saving ever shows up.
In children’s social care, for example, an answer that is efficient but can’t show its working is not a productivity improvement. It is a safeguarding problem with a spreadsheet attached.
That’s why grounding, citations and audit logs stop being “features” and become the whole point.
Efficiency without trust is not a smaller win. It’s the pilot that never scales.
Every time.
Here’s the move the dashboard hides
The three chess boards are connected. Pull one piece and the others slide.
Watch how it usually goes.
The board demands speed. The fastest way to hit a date is to centralise. One mandate beats forty conversations. But centralised and imposed is precisely how you destroy trust on the frontline. That’s because people have been ‘done to’ rather than ‘done with’.
So there you are: three apparently separate sliders, and one tug on the first has dragged the other two into the worst possible corner of the inevitable Boston Consulting matrix on slide 28..
That’s the difference between a dashboard and a chessboard.
A dashboard lets you set three dials and walk away.
Chess makes you account for the fact that every move changes the position of pieces you weren’t even looking at… including the ones two boards down.
So what do you do?
First, stop pretending the axes are independent.
Second, treat trust as the constraint that governs the other two, not as the third item on a list and a nice-to-have.
Speed is only worth having at the pace trust can keep up with.
Centralisation is only worth having where it earns trust rather than burns it.
Get the couplings right and “efficient” and “trusted” stop being opposite ends of an axis and start being the same thing.
Which is awkward for the tidy diagram to go ni a slide deck, but rather convenient for reality.
None of this is a recommendation to go slow.
My recommendation is to play all three boards at once. Deliberately. Not just trying to win on the one the directors happen to be watching, and losing the two they aren’t.
We build the boring, safe version of this for public bodies for a living. The permission-aware, audit-logged, shows-its-working kind. We make no apologies for that.
The unglamorous bits we deal with every day aren’t the cost of doing AI properly. They’re the move that keeps your other pieces on the board.
If your AI strategy currently fits on one slide with three sliders, you’re probably only looking at one board.
And in the AI game, that’s how you lose while looking very organised.
