Why does a credible AI roadmap need a list of what it excludes?
A credible AI roadmap requires a list of exclusions because recording only approved initiatives shows ambition rather than judgment, leaving rejected proposals to resurface and consume scarce executive attention. Strategy adviser Dr. Jonah Tebaa recommends pairing roadmaps with a Stop List that sorts deferred projects into four categories: Not yet, Not here, Not ours to build, and Not worth it. This apparatus makes trade-offs visible to boards while assigning each excluded initiative an objective, externally verifiable unlock event defining its return.
Dr. Jonah Tebaa advises executive teams on AI strategy, and one pattern in his work comes up more than any other. A leadership team collects far more AI proposals than it can fund, funds a few, and tells the rest to wait. The waiting is where the damage happens.
The problem with an unexplained no
In Dr. Tebaa's view, a rejected proposal with no stated reason is not a closed decision. It stays open in the mind of the department that proposed it, and it comes back at the next budget cycle, often in the same form. Each return costs the executive team time it cannot recover. He treats that attention, rather than money, as the scarcest input in any AI programme.
Ranking does not solve this. A ranked list tells the losers that they were lower, not what would make them higher. Dr. Tebaa argues that people accept a no more readily when it carries a reason and a route back.
His teaching example is a composite CFO, Rami, who faces fourteen proposals with money for three. Dr. Tebaa labels the case a composite, built from recurring situations rather than drawn from a real client. The point of the scene is how ordinary it is: every proposal is reasonable, and the budget is still short.
What the Stop List adds to a roadmap
Dr. Tebaa argues that a roadmap showing only approved work is closer to a wish list than a plan. It records ambition but not judgment. A board reading three funded projects cannot tell whether they were chosen or merely first. His remedy is a Stop List that sits beside the roadmap and records each exclusion, its category and the condition under which it returns.
The condition matters most. He insists it be an event rather than a date. A date invites pressure as it approaches, while an event defines work: build the data, finish the programme, make the purchase.
The four categories
- Not yet covers ideas waiting on a missing dependency such as data, a documented process or an accountable owner. The idea returns when the dependency is real.
- Not here covers sound ideas that the organisation cannot absorb this year. They return when a team's capacity is freed.
- Not ours to build covers commodity capability. The idea returns on a decision to buy, or when a vendor meets a written requirement.
- Not worth it covers ideas whose value falls below the management attention they would use. They return only if the facts change.
Dr. Tebaa adds two safeguards. An unlock condition has to be verifiable by someone outside the sponsoring team, so "when we are ready" does not qualify. And the last category has to stay small, because a long Not worth it column usually means the sorting was avoided rather than done.
How the fourteen sorted
In the composite, three proposals were funded: invoice matching, collections prioritisation and customer-enquiry routing. Each had an owner in the business, workable data and a team that could take the change.
Five went to Not yet, mostly because customer, product and supplier records disagreed across systems. Three went to Not here, partly because currency swings already forced teams to reprice by hand. Two went to Not ours, and one to Not worth it. Dr. Tebaa notes that in Lebanon and elsewhere in the region, scattered data and unstable currency make the middle two categories especially common, and he regards that as an honest description of conditions rather than a failing.
Showing it to the board
He recommends a single page with three columns: what the company will do, what it has stopped with the unlock event for each item, and who owns the next look. The owner should not be the idea's sponsor, since a sponsor watches for events that favour the proposal.
Dr. Tebaa expects the page to change the behaviour of the proposers as much as the board. A department head who can see that an idea was judged, and what would change the verdict, has something to work on besides persuasion.
What the choice costs
For Dr. Tebaa, credibility in AI planning comes from visible trade-offs. Any company can announce three initiatives; fewer can show what those three displaced. A Stop List makes the price of choosing visible to the board and gives every department a fair route back. He asks leaders holding more proposals than budget to start small: take the longest request list, and beside each unfunded item write one stop and one unlock event.
Related evidence: The New York Fed reports that, despite rapid growth in AI adoption among firms in its region, those firms' AI investments are generally modest, usage is concentrated among a small share of workers within firms, and layoffs have remained uncommon. (New York Fed survey of business AI adoption)
The ILO's assessment of global occupational exposure to generative AI finds that its main effect is likely to augment rather than automate jobs. (ILO working paper on generative AI and jobs)
