Contents · 6 min read
01 / 05 · The list
Why a long list is not a plan
In short
A long list shows what the company could build. It does not show what deserves its money or focus. The board needs a small set of bets and a clear reason for each.
A B2B software company brings its board a long list of feasible AI ideas. One copies features from a new entrant. Another builds a deeper view of customer needs. Both are feasible, but only the second can gain value as customer evidence grows.
One idea copies features from a new entrant. Another builds a deeper view of customer needs. A third changes how the service team delivers its work.
The list also holds an easy workflow task. It could save staff time at once.
The task is useful. The copied feature may close a real gap. Both gains count.
Yet the company cannot treat every sound idea as a top bet. That would turn the roadmap into a backlog.
The leaders also need enough skill to judge the ideas. That skill comes from using AI on their own work. A hands-on strategy workshop can build the same shared view.
Once the team can judge the list, its first cut must start with business return.
02 / 05 · The return
Which ideas change revenue or margin?
In short
Rank each idea by its expected effect on revenue or margin. Do this before you score ease of build. A simple build cannot make a weak return strong.
Use five gates in this order.
Business return
What could this add to revenue? What could it add to margin?
Strategic fit
Does it change what customers buy? Does it aid the firm's core plan?
Feasibility
Can the firm meet the work the bet needs?
Gains over time
Does each use make the next use worth more?
Stop rule
What must stay true? What evidence would end the bet?
For the software company, the deeper customer view has a revenue case. It aims to make the product fit customer needs better.
AI-supported service work has a margin case. It can cut delivery time and improve how the work is done.
The feature copy has no stated return yet. Ease of build does not fill that gap.
The workflow task can aid margin by saving time. Yet that gain has a clear limit. Once the task is automated, the same task cannot yield the saving again.
Keep revenue and margin on separate lines. Write how each gain would occur. Do not invent weights or a standard return hurdle.
An AI readiness assessment should end with three to five opportunities ranked by impact and ease. The high-return list is shorter, but it still needs a test of time.
03 / 05 · The edge
Which high-return bets grow stronger over time?
In short
Among the high-return ideas, favor those that gain value from use. Each new piece of evidence should improve the next choice. A gain that ends at parity should rank lower.
For the software company, deeper customer insight can grow with use. Each approved customer signal adds to what the firm knows. That store can improve the next product choice.
Copying an entrant can still close a useful gap. But the gain ends when the products reach parity.
The workflow task has the same limit. It saves time, then reaches its ceiling.
The service bet can grow in another way. Each project can add know-how to the firm's method. Faster work can lift margin. Better customer fit can make a switch harder.
Does each use add evidence?
If not, the gain may end with the first build.
Does that evidence improve the next choice?
If yes, the bet can gain value over time.
Does the edge end at parity?
If yes, keep it below a bet that builds its own edge.
This is the choice behind creating value that compounds. The team now knows which gains can grow. Next it must state the risk.
04 / 05 · The risk
What must be true for each bet to work?
In short
Write the work each finalist needs. Then name the assumption that would stop it. This makes risk clear without hiding it inside a made-up score.
For the software company, the customer insight bet needs approved customer evidence. It also needs time from people who can judge that evidence.
Its core assumption is simple. Better evidence must lead to better product choices. If that link fails, the team stops or changes the bet.
The service bet has its own test. Faster delivery must improve margin or customer fit. If neither moves, its case is weak.
Each finalist should pass four checks.
Delivery demand
Who must do the work? What current work must stop?
Core assumption
What must stay true for the return to appear?
First break
Which part of the case is most likely to fail?
Stop evidence
What result would cause the team to end the bet?
An AI-augmented management team can test big choices before the meeting. It can expose the weak number and the missing case. The leaders still make the call.
Keep risk beside the return. Do not blend both into one score. With each stop rule written down, the team can choose what to fund.
05 / 05 · The choice
Fund three to five and close the rest
In short
Fund the three to five ideas that pass every gate. Treat them as choices about money and focus. Keep every other feasible idea closed until new evidence changes its case.
For the software company, customer insight makes the cut. AI-supported service work also survives. The remaining places go only to bets that pass all five gates.
The copied feature stays out for now. So does the easy workflow task.
Both may still be sound pieces of work. They have not earned a place among the firm's top bets. New facts can bring them back later.
The roadmap is a record of where the company puts its money. The AI-augmented product organization keeps the proof and stop rules with each bet.
For your own company, four steps:
Name the survivors
The CEO and management team choose three to five. State the revenue or margin case for each.
Record the deal
Each bet owner keeps its evidence current. Add the work it needs and its stop rule.
Set the review
The management team agrees when fresh evidence will be checked. A product portfolio may use a monthly review.
Keep the rest closed
The portfolio owner reopens a rejected idea only when new evidence changes its case.
These choices can now anchor your AI strategy.
Choose what to fund
The management team now has to turn the shortlist into a shared commitment: three to five funded bets, each with a clear return case and stop rule. See The Opportunity gives the team three workshops to build hands-on fluency and decide where AI matters most. The work ends with a ranked roadmap, so the company can fund its chosen bets and keep the rest closed until the evidence changes.
Three to four weeks, three workshops with your management team: hands-on use of AI, custom AI agents for each leader, and a ranked roadmap of opportunities. Fixed fee, no access to your live systems.
Questions CEOs ask about this
How do we choose the three to five AI bets that matter most to our company?
Fund three to five AI bets that can lift revenue or margin, and reject the rest. A feasible idea qualifies only if its return is strong and its edge can grow. Each survivor needs a stop rule. The choice is which ideas deserve your money and focus.
What criteria should we use beyond technical feasibility?
Use business return and fit first. Then test whether the gain grows with use. End with the work required and the assumption that would stop the bet.
How should we compare revenue and margin against risk and strategic-option value?
Keep them separate. Revenue and margin show the expected return. Stop rules show risk, while gains over time show what the bet may make possible next.
Should we prioritize customer propositions over functions or workflows?
Do not rank an idea by its label. Rank what it changes for revenue or margin. A workflow can win if its return and gains over time beat a customer-facing bet.
Which attractive AI initiatives should we deliberately reject?
Reject ideas with weak business returns, even when they are easy to build. Also reject copied features that end at parity and bets with no clear stop rule.
How often should the portfolio be reviewed?
There is no single pace for every company. A product team may review its portfolio each month. Review any bet sooner when its stop assumption breaks.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- AI readiness assessment: the outcome of three to five opportunities ranked by impact and ease. (accessed October 8, 2026)
- AI training for executives: why leaders need hands-on use before they can judge AI bets. (accessed October 8, 2026)
- AI strategy workshop: the path from a shared frame to working tools and a roadmap. (accessed October 8, 2026)
- What does an AI-augmented product organisation look like?: the roadmap as a record of bets and choices. (accessed October 8, 2026)
- What does an AI-augmented management team look like?: how leaders test assumptions before they commit. (accessed October 8, 2026)
- The ROI Paradox: Chasing Returns Destroys Them: why gains that build over time deserve a separate test. (accessed October 8, 2026)
Next scheduled review: February 18, 2027