There's a prediction in Gartner's 2027 outlook that reads like an HR footnote and is actually a strategy warning: by 2027, 50% of enterprises without a people-centric AI strategy will lose their top AI talent.
The supporting numbers make it land harder. Only 27% of executives report having a comprehensive AI strategy. Just 20% believe their workforce is genuinely AI-ready. By 2027 an estimated 75% of hiring processes will test for workplace AI proficiency, and AI-skilled engineers already command a 20–40% salary premium.
Put those together and you get the real 2027 risk: not that you fail to buy the right AI, but that the people who could have made it work leave while you're still evaluating vendors.
Most "AI strategies" are procurement
Here's the diagnosis, and it's uncomfortable because it's so common. Open the document called our AI strategy at most organisations and you'll find: a vendor, a licence count, a shortlist of use cases, and a budget.
That's a tooling plan. It's a necessary artefact and it is not a strategy, because it answers what we're buying rather than what we're becoming.
The four questions a real strategy answers
- What does this change about how work gets done — here? Not "AI will boost productivity." Which processes, whose day, what stops being done at all. Generic ambition produces generic adoption.
- Who decides where AI is used, and who can say no? If nobody owns that, you get shadow AI by default — 98% of organisations already report unsanctioned use. Governance ambiguity isn't neutral; it's a decision to let individuals decide.
- What happens to the person whose role it reshapes? Redeployed, retrained, or quietly managed out? Your people will infer an answer whether or not you give one, and they'll infer the least generous one. This single question does more for adoption than any training budget.
- What are we asking people to become good at — and are we paying for it? If AI proficiency is now a hiring criterion, it's a development obligation for the people already here. Otherwise you're asking for a skill upgrade while signalling that the upgrade is their problem.
Why ambiguity costs you the best people first
The mechanism is worth stating plainly. Your strongest technologists are the ones most able to read a situation — and the most employable. When the AI plan is silent on people, they don't read it as nothing will change. They read it as something is planned that isn't being shared.
The people with the most options act on that first. So you don't lose a random sample of your team; you lose the top of it — precisely the people you needed to lead the work. That's the failure mode behind the Gartner number, and it's self-inflicted.
The asymmetry is stark: clarity costs a conversation. Replacing a senior engineer costs 1.5–2× their salary in recruitment, ramp-up and lost context — the same arithmetic that quietly erodes offshore team economics.
What to do this quarter
You don't need a finished transformation plan. You need to stop being silent:
- Say what AI is for here — augmentation, capacity, quality — and be specific about which processes.
- Name an owner for AI decisions and exceptions.
- Answer the role question out loud, even if the answer is "we don't know yet, and here's how we'll decide." Honest uncertainty beats a vacuum.
- Fund the skill you're now hiring for, for the people already on the payroll.
None of that requires a budget line. It requires a leadership position — the kind of judgement an interim or fractional CTO is often brought in to supply when there's no one senior enough to own it.
The bottom line
The 2027 AI risk that will actually hurt most organisations isn't picking the wrong model. It's having a procurement plan where a people strategy should be, and losing the talent that would have made either one work. Answer the four questions, say it out loud, and you keep the people and get the adoption. Turning technology ambition into an organisational position that people can act on is exactly what our Executive Advisory work is for.
Does your AI plan say anything about people yet? Let's fix that.