Three percent. That is the share of organisations whose leaders are highly prepared to manage AI-enabled ways of working, according to a July survey of C-suite and HR executives by ManpowerGroup Talent Solutions and Everest Group. The sample is modest, 80 senior leaders across the US and UK, so treat it as directional rather than definitive. But the direction is the point, and almost everyone is reading it wrong.
The instinctive interpretation is that leaders need more AI training. Book the workshops, run the literacy programmes, certify the executive team, and the number climbs. That interpretation is comfortable because it is purchasable. It is also the wrong diagnosis, and if you run a founder-led business, it is the wrong diagnosis in a way that will cost you the next two years.
The wrong diagnosis everyone is reaching for
When an AI initiative stalls, the post-mortem almost always lands on one of three culprits: the technology was not ready, the vendor oversold, or the organisation needed better change management. Occasionally all three. The prescriptions follow accordingly: switch tools, switch partners, or commission another round of enablement sessions.
Notice what all three explanations have in common. They locate the problem outside the leadership structure of the business. The tools failed us. The vendor failed us. The staff would not adopt. Nobody asks the more uncomfortable question: when the AI surfaced something that required a decision, who in the business was actually allowed to make it?
In most founder-led companies between 15 and 150 people, the honest answer is: the founder. Every meaningful call routes through one person, formally or informally. That structure survived every previous initiative because previous initiatives moved at the speed of meetings. AI does not.
What AI actually changed
Here is the reframe. AI does not create founder dependency. It makes founder dependency impossible to hide.
A working AI deployment is, at its core, a machine for surfacing decisions. It flags the churn risk before the renewal call. It drafts the proposal that needs approving. It identifies the pricing anomaly, the stalled deal, the campaign that should be killed this week rather than reviewed this quarter. Each of those outputs is only worth something if someone acts on it, and acts on it quickly.
Now run that stream of decisions through a business where the founder is the de facto approver of anything that matters. The maths does not work. One person cannot personally absorb decision volume that has been deliberately accelerated by software. So the outputs queue. Insights expire while they wait for the Monday meeting. The team learns that acting on the AI’s output without sign-off is risky, and acting with sign-off is slow, so they quietly stop acting at all. Six months later the board asks why the AI investment has not produced ROI, and the post-mortem blames the technology.
The constraint was never the model. It was decision rights. AI just surfaced decisions faster than a founder-centric structure could process them, and in doing so put a stopwatch on a problem that used to be invisible.
This is a pattern, not one survey
If this were only one modest survey, you could dismiss it. It is not.
The Adecco Group reached a structurally similar conclusion from a completely different angle in its July whitepaper on workforce orchestration: organisations are failing to redesign work fast enough to capture AI’s productivity gains, and they are eroding employee trust in the process. Read that carefully. The gap Adecco describes is not a skills gap. It is a work-design gap: who does what, who decides what, and how fast the organisation can reconfigure itself around a new capability.
RSM’s 2026 Middle Market AI Survey of 1,030 senior leaders in the US and Canada completes the picture. Adoption itself is no longer the story: 86% of middle-market firms have partially or fully integrated AI, and 97% report satisfaction with the results so far. RSM’s own framing of what comes next is telling: now comes the hard part. The tools are in. The satisfaction scores are high. What remains is the organisational work of turning capability into outcomes, and that work is executive work, not technical work.
Three different research houses, three different methods, one converging finding: the businesses struggling with AI are not struggling with software. They are struggling with the operating model wrapped around it.
A diagnostic you can run this afternoon
Strip away the survey data and you can test this in your own business with one exercise.
Name the last three AI-related decisions made in your company. Not the outputs the tools produced, the decisions: adopting a recommendation, changing a price, reallocating budget, acting on a flagged account, killing a workflow. Then answer honestly: who made each call?
If the answer to all three is you, the founder, then you do not have an AI adoption problem. You have a delegation problem with an AI accelerant. And it is worth asking the follow-up: how many recommendations never became decisions at all, because they arrived faster than your approval loop could turn?
This is the moment to be sceptical of the most seductive counterargument: that this is simply messy year-one adoption, and time will smooth it out. Time will absolutely improve tool fluency. Your team will prompt better, trust the outputs more, and build cleverer workflows. But time does nothing to decision rights. Nothing about month eighteen redistributes authority that month six left concentrated. That is precisely what the ManpowerGroup finding is measuring: not whether leaders understand AI, but whether the leadership structure can manage AI-enabled ways of working. Fluency ages into competence. Bottlenecks age into culture.
The fix is structural, not educational
None of this means training is worthless. It means training is insufficient, because no amount of AI literacy fixes a business where meaningful decisions still route through a single person.
The real work looks less like a workshop and more like an operating-model redesign, and it starts with three questions. First, decision rights: for each category of decision your AI systems now surface, who is empowered to act without escalation, and up to what threshold? Second, executive capability: do you have leaders in place who can own those decision categories end to end, or has every senior hire been shaped around executing the founder’s calls rather than making their own? Third, management cadence: does the rhythm of your business, its reviews, its reporting, its escalation paths, run at the speed your tools now surface decisions, or at the speed of the calendar you set five years ago?
Founders who have been through this recognise the pattern, because AI is not the first thing to expose it. A fast-growing sales team exposes it. A second product line exposes it. An acquisition exposes it. AI is simply the first initiative that exposes it in every function at once, on a timescale measured in weeks, with the evidence sitting in a dashboard the whole leadership team can see.
That is uncomfortable, but it is also clarifying. A vague frustration (“our AI projects keep stalling”) becomes a concrete, addressable piece of work: build the executive layer, distribute the decision rights, reset the cadence. Businesses that had already begun the founder-led to executive-led transition are finding AI adoption suspiciously easy. Businesses that had deferred it are finding AI adoption suspiciously hard. Neither is a coincidence.
AI did not create a new problem for founder-led businesses. It just put a deadline on an old one.
Blaque Software works with founders on exactly this transition: building the executive capability, decision structures, and management cadence that let a business scale beyond its founder. If your AI initiative keeps stalling in your own inbox, the tooling is probably fine. Let’s talk about the operating model.