← Writing

        Framework
        
        The AI Project Scorecard

How to tell if an AI project is actually worth building

Most AI projects fail because the job was not worth doing, not because the technology could not do it. This is a simple framework for deciding well before you build. We call it the AI Project Scorecard: five questions that tell you whether a project earns its cost.

Odysi Framework

Method 5-question scorecard

Read 6 min

The useful skill is deciding well before you build. Score each of the five questions from 0 to 2: 0 for no, 1 for partly, 2 for a clear yes. Add them up. The questions matter more than the exact total, but the total is a quick read on whether to proceed.

The test

The AI Project Scorecard

Is the problem real and expensive enough?

It has to genuinely cost you time, money, missed revenue, or errors. A clear yes: you can point at the hours or the cost, and it is significant.

012

Does the work follow a pattern?

AI is strong on pattern-following work, weak where every case needs rare judgment. A clear yes: most of the volume follows a knowable pattern, and only a minority needs a person.

012

Is the data or knowledge actually there?

A tool is only as good as what it draws on. A clear yes: the data or knowledge exists and is accessible, or getting it there is cheap.

012

Does the value clear the cost, including running it?

Set the return against the build plus the ongoing cost. A clear yes: the honest return comfortably exceeds the build and running cost.

012

Can it be owned and run without heroics?

A tool only one expert can keep alive is fragile. A clear yes: once built, your team can run it, and maintenance is manageable.

012

8–10

A strong candidate. Build it, ideally starting with a small proof of concept.

5–7

Promising but not obvious. Fix the weak answers, usually data or scope, before committing.

0–4

Not worth building as framed. Narrow it, solve a different part, or decline. That decision is a win.

The most valuable outcome of the AI Project Scorecard is often a confident no. The cheapest AI project is the one you correctly choose not to build.

Why

Why this beats starting with the technology

Most failed projects skipped this and started from can we build it rather than should we. The technology can build almost anything now, which is exactly why the discipline has moved to deciding what deserves building. The AI Project Scorecard puts that decision first, where it belongs.

Common questions

FAQ: the AI Project Scorecard

How do you decide if an AI project is worth building?

Score it against five questions: is the problem real and expensive enough, does the work follow a pattern, is the data or knowledge there, does the value clear the full cost including running it, and can it be owned and run without heroics. A clear yes on most of them means it is worth building.

What makes an AI project a good candidate?

High-volume, pattern-following work with a real cost behind it, data that already exists, a return that clears the build and running cost, and a tool the team can maintain.

Why do most AI projects fail?

Usually because the project was not worth doing as framed, or the hard parts, data, edge cases, and running cost, were not weighed before building. Starting from can we build it rather than should we is the common root.

Is deciding not to build a failure?

No. A confident no is the cheapest and often the most valuable outcome, because it avoids spending on something that would not have earned it.

What score means I should proceed?

As a guide, 8 to 10 is a strong candidate, 5 to 7 is worth fixing before committing, and 0 to 4 means rethink or decline. The questions matter more than the number.