Three months ago, you gave someone the AI mandate. There was an announcement, perhaps a bit of internal excitement, possibly a budget line. You moved on to everything else demanding your attention.
Then the board asked how it was going.
That pause before you find an answer you can give confidently is what this is about. Handing off the AI programme to a capable operator is the right call for a founder running a business at pace. Assuming it is therefore done is where people tend to get caught out.
What does delegation actually mean for an AI programme?
Delegation means someone else owns the execution. They choose the tools, manage the vendors, run the pilots, and account for results. In those terms, handing the AI mandate to a capable operator makes sense. A founder who tries to personally manage every AI project slows the programme down. The handoff itself is the right call.
The issue is what founders assume once the handoff is complete. The moment someone else is accountable, there is a natural pull to treat the thing as done. You delegated; therefore it is in hand. With many operational decisions, that logic holds. The sales process is the sales director’s problem. The hiring pipeline is HR’s problem. Those are genuine hand-offs. With AI, something different is happening, because the work involves decisions and judgements that have historically sat with the founder, and passing those down is not as clean as it appears.
Why does this mandate resist complete hand-off?
The reason AI resists complete delegation is that a meaningful part of the work involves codifying how decisions get made. Which bids to pursue, how capacity gets allocated, which clients get priority attention. Those judgements have historically sat with the founder. A delegate can manage the programme, but they cannot document your reasoning without you in the room at the key moments.
Spencer Stuart’s research on CEO involvement in AI frames this directly. Treating AI as technology to be installed rather than something to be shaped at leadership level is a category error. The work requires active engagement from above the implementation team, not just a capable person running the programme underneath. BCG’s 2025 research on AI adoption found that roughly half of companies remain stuck in early stages, unable to move past proof-of-concept. Sustained leadership engagement above the implementation layer is a consistent factor in those that have progressed.
There is also a paradox worth naming. AI implementation done well reduces the degree to which the business relies on founder judgement. That is good for operations and for any exit conversation. But if the programme is built around a delegate who replicates the founder’s instincts without documenting the underlying process, the business can become more dependent on the founder, not less. The programme appears to be working. The actual risk profile has barely shifted.
The fix is to use the AI programme as the forcing function to document founder decision-making, not replicate it. That is a different brief for the delegate, and it requires the founder to show up for the conversations that make it possible.
Where does ‘handed off’ tip into ‘abandoned’?
The clearest signal is usually a board question. Someone asks how the AI programme is going and the founder has to reconstruct an answer from memory. The programme has been running but the founder cannot speak to it with authority. That gap between ‘delegated’ and ‘in hand’ is information about how the programme has actually been led.
MIT NANDA’s 2025 research on generative AI in business found that roughly 95% of pilots stall or show no measurable improvement to the bottom line. The gap sits in how the work is led and integrated, not in the quality of the technology. Programmes without active senior sponsorship tend to generate activity and reports, but they do not produce the kinds of workflow change that add up to real outcomes.
Change management research has found this consistently. Technology programmes fail not because the tools do not work, but because the leadership work around them is underestimated. NACD’s research on AI governance notes that board-level expectations for executive accountability cannot be fully met by a delegate alone. The accountability attaches to the role, not the mandate.
There is also the reverse failure. A founder who delegates verbally but keeps pulling decisions back teaches the organisation something clear. AI is still the founder’s call. The delegate loses authority, the team waits for sign-off, and the programme slows for the wrong reason.
When does the founder need to be in the room?
Founders rarely need to attend every project meeting. When they reclaim every decision after handing off the mandate, the delegate loses authority and the programme stalls. The right level of involvement is presence at the moments that shape direction rather than manage delivery. Setting the mandate clearly, agreeing decision rights early, and appearing periodically enough that the team knows this has real backing from above.
BrainStorm’s research on technology adoption found that organisations reaching meaningful activation within 90 days consistently showed active C-suite sponsorship, ongoing leadership communication, and role-specific guidance. The sponsorship that matters is presence at the moments that set direction, not involvement in the operational layer.
Decision rights are the practical shape of this. The founder and delegate should agree early on which calls belong to the delegate, which belong to the founder, and which require the founder to weigh in before anything proceeds. Without that conversation, the delegate hedges and the founder interferes. Neither serves the programme.
Two or three times a year, the founder should review outcomes against the original mandate. Are we building the decision-making capability we described, or just adding tools? That review is the oversight the founder role requires. Everything else can stay with the delegate.
How does this connect to the exit-readiness work you are already doing?
For founder-led businesses with an exit in the next few years, this sits alongside the owner-dependency question. M&A advisors consistently flag founder-centric operations as a discount to exit multiple. AI could be the mechanism that reduces that dependency by codifying the decisions and knowledge that currently sit with the founder. That only works if the founder is present when the codification happens.
M&A advisory research puts the founder-dependency discount in a wide range, but consistently notes it as one of the largest single variables in exit valuation for owner-managed businesses. AI done well can shift that, by turning founder knowledge into documented process. But AI built around a delegate who mimics the founder’s instincts without documenting the underlying logic can make the dependency worse, not better. The programme can look healthy from the outside while the actual risk profile has barely shifted.
The question worth sitting with is whether the current programme is moving the business toward a version that could run on explicit criteria rather than founder presence. If the founder has been largely absent from the programme, that is hard to assess honestly. Delegated delivery is the right call. Owned outcome is the part that stays on the founder’s desk regardless of who is running the work.
