Insights · AI strategy
Where should you invest in AI first?
The average company has more AI ideas than its team can deliver. What separates the companies that capture value from those that only collect pilots is the order in which things get done.
The problem isn't a lack of ideas
In McKinsey's 2026 global survey, nearly nine in ten respondents say their organization regularly uses AI in at least one business function. Only 37% attribute any EBIT impact to AI. About 6% attribute 5% or more of EBIT to it and also report significant value. According to the same survey, that 6% stands out for redesigning workflows around AI and for having visibly committed senior leaders.
Neither has anything to do with having more ideas. Our takeaway: choosing well is worth more than having many options.
In a data and analytics program at a large company, workshops with the business units produced more than 20 initiatives: forecasts, dashboards, models, automations. The question stopped being "what can we do with AI?" and became "what do we do first?" The answer came from a simple matrix, with written criteria and explicit weights, that sorted the initiatives into waves.
What usually goes wrong
- The loudest voice wins. The unit with the most clout gets the team, and the most valuable initiative waits.
- The flashiest idea wins. The project that demos well jumps ahead of the demand forecast that actually moves margin.
- The cost of data shows up late. Much of the effort in an AI initiative usually goes into finding, cleaning, and joining the data, and that only becomes clear after the project starts.
- Nothing but quick wins. A portfolio of quick wins alone doesn't build the capability the team will need for the big project that comes next.
- No one from the business signs off. An initiative without a business owner becomes a pilot that never reaches production.
How to decide: effort × impact, with written criteria
Each initiative gets four scores from 1 to 5, always with the scale written out next to it.
Effort
- Data availability (higher weight): 1 = structured, in a known database; 3 = exists, but scattered or semi-structured; 5 = decentralized, unstructured, or nonexistent.
- Time to value (horizon): 1 = about 3 months; 3 = 6 to 12 months; 5 = more than a year.
Impact
- Business value (higher weight): 1 = marginal gain; 3 = meaningful improvement to a process; 5 = changes how the company operates.
- Team learning: 1 = repeats what the team already knows; 3 = broadens its toolkit; 5 = opens up a new capability.
Effort = 0.7 × data + 0.3 × horizon. Impact = 0.7 × value + 0.3 × learning. The result is a map with four regions:
- High impact, low effort: First wave.
- High impact, high effort: Strategic bet, with a pilot and a decision point.
- Low impact, low effort: Low priority; worth doing only to learn, if it teaches the team something new (a learning score of 4 or 5).
- Low impact, high effort: Drop.
| # |
|---|
Three things we learned to watch for
- The weights change the ranking. Switch 70/30 to 50/50 and the order of the initiatives changes; you can watch it happen in the tool above. That's why the weights have to be agreed up front, together, and not adjusted afterward to confirm someone's preference.
- A blank score is not a score. We've seen a prioritization spreadsheet where a formula turned a blank cell into the top score, and an unscored initiative climbed the ranking without anyone noticing. The tool above leaves any initiative with a missing score off the map.
- It's an ordinal scale. A 4 is not necessarily twice a 2, and the weighted average is an approximation. The matrix structures the discussion; it doesn't measure value in euros. Initiatives close to 3 call for discussion, not a verdict; within each region, the tool ranks by the gap between impact and effort. And first-wave initiatives still need a business case with numbers.
- the list of AI initiatives is longer than your capacity to deliver;
- priorities are set by whoever has the loudest voice, not by written criteria;
- projects stall for lack of data once they are under way;
- you have AI pilots that never made it into production;
- the executive team has asked for "an AI strategy" and there is still no prioritized portfolio.
We run the prioritization with your business units: short workshops, the matrix filled in against agreed criteria, and a portfolio in waves, with the first initiative scoped for a pilot. If the conclusion is that it isn't time for AI yet, we'll say so.