AI Use-Case Prioritisation Matrix (Free Template)

Every team we talk to has the same problem in a different shape: there are now ten things you could do with AI, and the budget and attention for maybe one. A finance lead wants invoice coding automated. Ops wants the call centre transcripts summarised. Someone read about an agent that drafts quotes. They are all plausible. They are not all worth doing, and they are definitely not all worth doing first.

This matrix is the scoring sheet we use to settle that argument with evidence instead of volume. It ranks each idea against five things that actually predict whether an AI project pays off, then gives you a single number to sort by. Download it, score your shortlist, and the order to build in stops being a matter of opinion.

The five things worth scoring

Score each idea from 1 (poor) to 5 (strong) on each dimension. The weights reflect what tends to kill or carry these projects in practice, not in theory.

Dimension	What you’re actually judging	A “1” looks like	A “5” looks like	Weight
Impact	Money saved, revenue enabled, or risk removed if it works	A nice-to-have nobody asked for	Hours back every day, or a number the board watches	30%
Feasibility	Whether today’s tools can do it reliably enough	Needs research-grade accuracy to be safe	A well-trodden pattern with known tooling	25%
Data readiness	Whether the data exists, is accessible, and is clean enough	Locked in a vendor system or a person’s head	Already in a database or files you can reach	20%
Risk	What happens when it gets something wrong	A wrong answer reaches a customer unchecked	A human reviews output before it counts	15%
Strategic fit	Whether it builds toward where the business is going	A one-off that helps one person	A capability several teams will reuse	10%

Every dimension is scored the same direction: 5 is always better. That includes risk, where 5 means low, contained and reversible (a human checks the output before it counts) and 1 means high (a wrong answer reaches a customer unchecked). Scoring everything the same way keeps the maths simple and stops anyone gaming the result.

The scoring formula

The priority score is a weighted average on a 1–5 scale:

Priority = (Impact × 0.30) + (Feasibility × 0.25) + (Data × 0.20) + (Risk × 0.15) + (Fit × 0.10)

You don’t need to calculate this by hand. The downloaded sheet has the formula built into the Priority score column: type your 1–5 ratings and the ranking updates itself.

A worked example

Three ideas from a regional services business, scored honestly:

Use case	Impact	Feasibility	Data	Risk	Fit	Priority
Auto-summarise inbound enquiry emails into the CRM	4	5	4	4	4	4.25
Chatbot that quotes jobs to customers unsupervised	5	2	3	1	3	3.05
Draft first-pass replies to supplier emails for staff to send	3	5	4	4	3	3.85

The unsupervised quoting bot has the highest impact and is the one everyone gets excited about, and it ranks last, because it is hard to do reliably and expensive to get wrong. The boring email-summary job wins. That pattern is exactly why scoring beats arguing.

How to use it

List every idea on the table, however half-formed. Capture them all before you judge any of them.
Score in a group, not alone. The disagreements are the useful part: if two people score data readiness 2 and 5, you’ve found something to check before you commit a dollar.
Sort by priority score and draw a line. Fund what’s above it.
Re-score quarterly. Tooling improves and data gets cleaner, so last quarter’s “feasibility 2” can become this quarter’s “5”.

Common mistakes it helps you avoid

Funding the loudest idea instead of the best one. Volume in a meeting is not evidence.
Ignoring data readiness until you’ve started. The single most common reason an AI project stalls is that the data turned out to be messier or less accessible than anyone admitted up front.
Treating risk as an afterthought. The cheapest, safest place to notice that an idea is customer-facing and irreversible is on a spreadsheet, before the build.

A high score is permission to start a conversation, not permission to skip discovery. The point of the matrix is to make sure that conversation is about the right project.