The AI Project Risk Scorecard. Ten Signals. One Page. Score It Before You Fund It.

A one-page scoring exercise that surfaces the ten risks deciding whether an AI initiative delivers, weighted so the structural problems can't hide behind good execution, and short enough to fill in with your leadership team in a single sitting.

AI projects that fail usually go wrong in the scoping, weeks before anyone writes a line of code, and nobody names it until the retrospective. By then everyone has settled on the same account of it: the technology was fine, the team was capable, and the project still didn't deliver.

The uncomfortable part is that the reasons were visible the whole time, and they just weren't written down anywhere. This scorecard writes them down. Ten risk signals, drawn from AI initiatives that stalled and the specific decisions that stalled them, scored against the project you're about to fund.

What's Inside

Section 1: The Ten Signals

Five structural signals, covering how the project is defined and where accountability for the output sits. When these are present, execution quality doesn't rescue the outcome.
Three execution signals, covering the cost, timeline, and user-calibration assumptions that hold at prototype scale and break at production scale.
Two capability signals, covering team experience and the exposure that comes with putting AI output in front of the outside world.

Section 2: The Scoring System

Weighted scoring across fifteen points - structural signals count double, so a project can look clean on volume and still score as high risk on substance.
A structural override threshold that supersedes your total. Cross it and the project is high risk regardless of how the rest of the sheet reads.
A four-band interpretation table giving each score a specific instruction: proceed, fix, stop and re-scope, or don't fund at this scope. No score leaves you wondering what to do with it.

One of the Ten

Signal four is adoption.

The business case math is always the same. Hours saved, times users, times loaded hourly cost. The number comes out large and the case gets approved.

New internal tools land somewhere around 40 to 60% adoption in the first ninety days, and AI tools sit at the harder end of that range because people have to trust an output they can't fully predict. So an 80% ROI at full adoption quietly becomes 32% at real adoption. The business case was wrong on the day it was signed off, and nobody names it as the cause at the annual review.

That's one of ten, and several of the others are cheaper to fix than this one.

Who This Is For

Exec sponsors, product leads, and CTOs on an AI initiative with a budget decision coming in the next quarter. There's a real use case, a rough number attached, and a team ready to start.

If you're still in exploration mode with no live use case in sight, this is early. Bookmark it and come back when there's a budget on the table, because the scorecard is built for the moment before you commit and does very little for you before then.

Where This Comes From

The scorecard is the working version of the argument in AI Doesn't Reward Risk. It Punishes Bad Scoping. The article covers why risk and return don't move together in AI the way they do in most investment decisions, and walks through each of the ten signals with the reasoning and the failure cases behind them. Read that for the thinking, use the scorecard to run it against your own project.