Picture the whiteboard at the end of a two-hour AI workshop. Twenty ideas, colour-coded by department, each with a hand in the air attached to it. Finance wants invoice automation. Marketing wants content generation. Operations wants predictive scheduling. The delegate chairing the session has thirty seconds to decide what happens next, and the two wrong answers are to pick one on the spot or to tell everyone their idea is on the list.
The right answer is to close the workshop, take the full list, and come back with three.
What does picking three AI opportunities actually mean?
Picking three means ranking the full opportunity list against two criteria, business impact if the work succeeds, and genuine feasibility given your current data, infrastructure, and team. Each of the three that survive must carry a specific business outcome, a time-to-value estimate, and a rough investment figure. Everything else goes onto a parking list with a written reason for each exclusion.
The scoring doesn’t need to be sophisticated. A simple two-axis grid, impact on one side and feasibility on the other, moves twenty ideas into four groups inside half an hour. The top-right quadrant, high impact and genuinely achievable with what you have today, produces your shortlist. From that shortlist, you select the three with the clearest path to a measurable result in the first sixty to ninety days.
The discipline is in the limit, not the exact count. Some programmes run four or five concurrent threads without losing focus. The question is whether the active list can be staffed, sequenced, and monitored. If the honest answer is no, the list is too long.
The business outcome attached to each opportunity does two things. It gives you a baseline to measure against once work is under way, and it forces the question of what success looks like before anyone commits time or money. “We will reduce invoice processing time by 40% within sixty days” is a business outcome. “Improve finance operations using AI” is a project description that cannot be evaluated.
Why does a long opportunity list slow the programme down?
A long opportunity list carries fifteen open commitments, each with a champion inside the business who reads its continued presence as a signal that their idea is live. A programme cannot be resourced, staffed, or sequenced when it points in fifteen directions at once. Specificity, rather than ambition or budget, is the primary differentiator between AI programmes that deliver results and those that stall before the first review.
The connection between scope and failure runs deeper than resource allocation. When the programme has no clear boundary, the delegate’s time spreads across early-stage exploration on a dozen fronts, with nothing advanced enough to show results. Boards expect progress in weeks. The gap between a promising proof of concept and a result that registers in the P&L is where many programmes end, and a wide scope guarantees arriving there without the momentum to push through.
Research tracking AI initiatives across organisations puts the pilot failure rate at around 95% before programmes reach measurable P&L impact. The single biggest differentiator for the fraction that succeed is specificity, solving concrete problems in defined operational contexts with clear outcome targets, rather than broad capability roll-outs across the business.
Running three opportunities well, delivering one quick win in the first quarter, and building credibility before expanding is a faster path to a meaningful AI programme than spreading resources across fifteen ideas at once. Scope discipline is the mechanism that makes ambition achievable.
Where does the pressure to expand the scope come from?
The pressure to add more rarely comes from bad intentions. Department heads see AI tools working in adjacent sectors and want the same for their teams. Founders read about a competitor’s deployment and ask why you haven’t moved. Board members bring ideas from their other portfolio companies. Each request is reasonable taken alone. The problem is that none of them were scored against the same criteria as the current three.
This is the moment that breaks programmes. A founder who delegated the AI mandate because it felt technical will re-engage when they hear about something exciting. Their informal question often lands as an instruction. The delegate without a visible, documented process for evaluating new ideas has no clean way to respond without either adding the idea to the active list or appearing to dismiss the founder outright.
A lightweight intake process closes that gap. A single-page form that any department head or senior leader can use to submit an AI idea for formal evaluation gives new requests a place to go that is not the active programme. The idea enters the queue. At the next scheduled review, it gets scored against the same impact-and-feasibility criteria as everything else. If it outranks something on the current list, the swap happens at review, not between sessions in response to an informal conversation.
The intake process also handles the informal AI adoption that is present in many businesses. When teams know there is a legitimate channel for new ideas, they are less likely to procure tools outside the programme and more likely to wait for the next evaluation cycle.
When should you hold the line, and when should you reopen the list?
The default during the first ninety days is to hold the line. No new opportunity should join the active list before at least one quick win has shipped, because reopening scope before any result exists turns a focused programme into a wishlist. The test for adding is straightforward, a new idea must genuinely outrank one of the current three on impact and feasibility, and a delivered result must exist first.
The quarterly review is the natural point to reopen. By then, you have data on what is moving and what is stalling. The ranking criteria can be reapplied to the full pool, including anything that came in through the intake process. If a new opportunity genuinely outranks one of the current three, the swap happens at review, not in response to a phone call between sessions.
The pressure to add before that point usually comes from anxiety rather than strategy. A founder watching competitors deploy AI, a board member with a specific tool in mind, or a department head whose idea did not make the original cut will each generate pressure to move faster or go broader. The parking document is the most effective response. Showing someone that their idea was formally evaluated, scored, and parked with a written reason is a different conversation from telling them it wasn’t selected.
Holding the line is the default for the period when the programme has no track record and the delegate has no delivered results to draw on.
What else belongs alongside the ranked list?
The ranked list of three works best when three companions sit next to it. A parking document that names what was excluded and why, so champions can see their idea was considered rather than dismissed. A lightweight intake form for new requests. And a scoring template that makes the impact-and-feasibility criteria visible and consistent, so every evaluation uses the same basis and the process is defensible when decisions are questioned.
The scoring template also shifts the conversation when a senior stakeholder challenges the list. A visible, documented process moves the question from “why wasn’t my idea chosen” to “how does my idea score on the criteria.” That reframe reduces friction and keeps the programme from becoming a negotiation every time a new suggestion arrives.
These three tools connect to the broader work of the first ninety days. The ranking method belongs in the opportunity identification phase around days thirty-one to sixty. The intake process feeds into the roadmap review in days sixty-one to ninety. The parking document is what keeps the scope anchored when the programme enters its second quarter and the pressure to expand arrives in earnest.
If you are building this process from scratch and want to work through the scoring criteria before your first workshop, book a conversation.
