Three weeks with no update from your delegate. You chase, get a response about “continued progress” and a timeline that has shifted by a month. Your delegate is still in the role, still attending the right meetings, but you cannot tell whether the initiative is doing what early-stage AI work is supposed to do, absorbing time before anything shows, or whether it has effectively stopped moving and nobody wants to say so.
That gap is where this post sits. You handed someone a mandate, you have given it time, and now you are sitting with a decision many founders avoid naming directly: give it more time, or step in.
The answer depends less on how long the initiative has been running than on which signals you are actually reading.
What does a genuine stall actually look like?
Meaningful ROI from an AI initiative typically takes 12 to 24 months, so absent financial results in month four tell you nothing useful. What matters at this stage is whether the activity that will eventually produce results is actually moving. Are users adopting the tool? Has at least one process been formally changed? Are any quick wins scoped and in progress? A genuine stall shows in those leading indicators, well before any P&L line moves.
BCG’s 2025 research on AI adoption found a consistent pattern: usage rates climb while commercial impact lags. Adoption precedes measurable output by design, and the measure you apply at the 90-day mark should reflect that. The right question is whether the conditions for results are in place, not whether results have appeared.
The leading indicators worth watching are weekly active users on the deployed tool, the number of workflows that have changed as a result of the initiative, and whether the team responsible can name at least one problem that now takes noticeably less time. Lagging indicators, things like cost reduction or revenue improvement, tell you whether it worked. They cannot tell you whether it is currently working.
Where this distinction matters most is in the conversations you are having with your delegate. If every update is framed in terms of what is planned or in progress, with nothing described as actually different from last month, that pattern is worth paying attention to.
Why does stepping in too early make it worse?
When a founder steps in on a delegated initiative before the work has actually stalled, the damage is rarely visible straight away. The organisation reads the intervention as a signal that the mandate was never fully genuine, which makes every future handoff harder. The delegate shifts to managing up rather than managing the initiative, exactly the dynamic you handed it off to avoid.
Spencer Stuart’s work on AI leadership describes how founders who delegate because the work feels too technical tend to disengage until results fall short, then re-engage with course-correction demands that are difficult to deliver without proper authority. The pattern is recognisable: a formal delegation followed by increasing informal involvement, more questions in meetings, requests to be kept in the loop on decisions. The delegate reads all of it and adjusts their behaviour accordingly.
Korn Ferry names a related version of this dynamic the AI readiness paradox. Organisations assign AI leadership to strong operators who lack the specific competencies the task needs, then hold them to standards they were never equipped to meet. When the founder steps back in, it confirms the framing that the delegate was always a placeholder.
Before you act, ask whether the slow patch you are watching traces back to your own involvement. Have you moved the goalposts since the brief was set? Have you withheld a decision the delegate needed? Is the quiet you are reading as stalling actually the delegate waiting on you?
Where do the warning signals show up?
The signals of a genuine stall appear in the texture of conversations, not in financial reports. If your delegate is reporting activity but struggling to name a specific decision made, a process changed, or a team member whose work now looks different, the initiative has likely become management theatre. Named specifics after 90 days tell you more than any timeline does.
One practical test: ask whether anyone outside the immediate project team knows the initiative exists and can describe what it does. If the answer is no after three months, the initiative is contained rather than integrated, and containment is a stall in a different frame.
Other signals to look for include updates that describe inputs rather than outcomes, budget conversations being avoided or deferred, and the team’s vocabulary shifting from what has changed to what is being explored. Poor data readiness is a commonly cited underlying cause at this stage; Schellman’s research found 77% of firms identify data quality as the biggest barrier to AI delivering value. If that is the real blocker, it is worth surfacing in your reset conversation rather than treating as an excuse.
When should you act, and when should you wait?
Deciding whether to act comes down to the quality of evidence in front of you. Two questions help. The sunk-cost check asks whether, knowing what you know now, you would start this initiative today. The scope check asks whether the person leading it knows exactly what authority they still hold. Both cut through timeline anxiety and focus you on the actual decision.
The sunk-cost check is borrowed from capital allocation. If a well-informed decision-maker, starting fresh, would fund the initiative on current evidence, the patience is probably worth it. If the honest answer is a long pause followed by “I think so”, the initiative may be running on inertia rather than genuine conviction.
The scope check matters as much. Spencer Stuart’s research notes that founders frequently withhold the decision authority their delegate actually needs, then interpret slow progress as a performance problem rather than a setup problem. If your delegate regularly comes back to you for decisions that should sit within their own brief, the mandate was never as clear as you thought it was.
Propeller’s dual-ROI framework offers a useful distinction between trending ROI, the early-stage leading indicators showing directional movement, and realised ROI, the financial outcomes you can actually record. At the 90-day mark you should be seeing trending ROI signals. Realised ROI comes later, and measuring too early tells you nothing useful about whether the initiative is working.
What does a clean intervention look like?
A clean intervention resets scope and decision rights rather than seizing the work. The goal is to give the initiative better conditions, not to signal lost confidence in the person leading it. That distinction separates a productive reset from a reverse delegation, where the founder reclaims the mandate in everything but name and the original handoff is abandoned without anyone saying so.
In practice, a clean reset is a structured conversation with your delegate. Confirm the mandate is still theirs. Clarify which decisions they can make without coming back to you, and name any that genuinely need your sign-off. Agree on two or three specific indicators you will both track from here forward. Set the conditions under which you would mutually declare the initiative paused, rather than letting it drag on without a resolution.
A useful scope-check question before that conversation is whether this initiative would still be worth pursuing if you stripped out the AI component. Would the underlying process change be valuable on its own? If yes, the initiative has real foundations and the stall is recoverable. If no, it was always about the technology rather than the problem, and a renewed mandate will not fix that.
The founder who builds this capability, knowing when intervention helps versus when it pulls the mandate back into their own orbit, ends up with a business that can actually run AI initiatives without constant oversight. The initiative in front of you is the test. What you do with the signals is the answer.
