Almost every company now has an AI initiative, and many have dozens of pilots. Far fewer can point to a measurable change in cost, revenue or risk. The technology is rarely the bottleneck. What is missing is the discipline to choose: which use cases deserve money, which need groundwork first and which should never start.
Two questions for every use case
Score each idea on two axes. Value: how much cost, revenue or risk it moves, at full scale, in money. Feasibility: whether the data exists and is good enough, whether the process can absorb the change and whether the risk is acceptable. Plot them together and the portfolio organises itself.
Quick wins: fund the journey
High value and high feasibility: document processing, customer service assistance, demand forecasting. Data is available, the process owner is clear and results show up in months. These pay for the rest of the programme and build credibility.
Big bets: prepare the ground
High value but low feasibility: autonomous planning, dynamic pricing, predictive maintenance without sensor history. They can transform the business, but only after the data foundations, the process redesign and the governance are in place. Fund the groundwork, not the model.
Low value: say no
Easy but low value ideas are attractive because they demo well. A handful are fine as learning projects; dozens drain attention from the work that matters.
Governance that speeds things up
- One portfolio, one owner: a small committee that reviews the map every quarter, moves money between use cases and stops what does not deliver.
- Clear risk tiers: internal productivity tools need light rules; decisions about customers, credit or people need human review and audit trails.
- Shared foundations: data platforms, security and evaluation methods built once and reused by every use case.
Treating AI as a portfolio does not make it less ambitious. It makes ambition affordable: early wins create the budget and the trust that the transformational bets will need.
Bartor S.
Founder
