The ROI Portfolio Map: turning AI ideas into investment decisions.

Every company with momentum on AI eventually accumulates the same thing: a backlog. Forty, sixty, a hundred ideas, gathered from workshops, vendor pitches, and the team member who read something over the weekend. The list is genuinely full of good ideas. And it is almost useless, because a list of good ideas is not an investment plan. The question a leadership team actually faces is not “is this a good idea?” — most of them are — but “which three do we fund this quarter, and which fifty do we consciously not fund?”

That second question is an investment question, and it deserves the discipline any other capital allocation gets. The ROI Portfolio Map is the tool that supplies it. It does for an AI backlog what a stage-gate does for a project pipeline: it forces every candidate onto the same axes, scores it honestly, and produces a ranking the leadership team can defend — not because the ranking is perfect, but because the reasoning is explicit and shared.

The goal is not to compute a single magic number per idea. It is to make the trade-offs visible enough that a leadership team can argue about the right things and stop arguing about the wrong ones.

Value against feasibility, risk in the colour.

At its simplest the map is a 2×2: business value on one axis, feasibility on the other, with risk encoded as a third dimension — colour, size, or a flag. Every idea in the backlog lands in one of four quadrants, and the quadrant tells you most of what you need to know about what to do with it.

The 2×2 is the headline. Underneath it sits a scoring table, and the table is where the honesty happens.

Four dimensions, scored the same way for everything.

Value and feasibility are each composites. Scoring them on the same rubric for every candidate is what makes the comparison fair — and what stops the loudest sponsor from winning. A workable rubric scores each initiative, on a simple scale, across four dimensions:

Risk sits alongside, scored separately and never averaged into value — because a high-value, high-risk initiative and a moderate-value, low-risk one are genuinely different bets, and collapsing them into one number hides exactly the distinction the board needs to see. Regulatory exposure, model-failure consequence, and reversibility are the three risk inputs that matter most for AI specifically.

From advocacy to allocation.

The real effect of the portfolio map is not the chart. It is what happens in the room when leadership reviews it. Without a map, AI prioritisation is advocacy: each sponsor argues for their initiative, and the decision tracks organisational power more than business value. With a map, the conversation moves up a level. Nobody argues that their idea should jump the queue; they argue about whether a given score is right — is the data really that ready, is the value really that large, is the risk really that contained?

Those are far better arguments to have. They are specific, they are evidence-based, and resolving them improves the decision rather than just deciding it. The map also makes the “invest to unlock” quadrant legible, which is where the most valuable insight usually lives: the highest-value initiative is frequently not the right first project, because a smaller, unglamorous enabling project — a data cleanup, a labelling effort, a platform foundation — is what makes the headline initiative feasible two quarters later. A list of ideas hides that. A map surfaces it.

And it makes declining defensible. Most of the backlog will not be funded this year. Without a map, those ideas linger as low-grade organisational guilt and resurface in every planning cycle. With a map, declining them is a recorded decision with a reason attached, which is both more honest and far easier to revisit when conditions change.

The map is only as good as its honesty.

One caution, because it is the way this tool fails. A portfolio map is only as useful as the honesty of its inputs. The most common failure is feasibility inflation — every sponsor scores their own data as ready and their own org as prepared, because nobody wants their initiative to land in the low-feasibility row. The fix is to score feasibility with the people who would actually do the work, not the people who want it funded, and to treat data readiness as guilty until proven innocent. A map built on optimistic feasibility scores is worse than no map, because it launders enthusiasm as analysis.

Done with that honesty, the portfolio map converts the single most common AI artifact — a backlog of good ideas — into the thing leadership actually needs: a ranked, scored, defensible set of investment decisions, with the declines as deliberate as the approvals. It is the bridge between “we have a lot of AI ideas” and “here is what we are funding, and here is exactly why.”