AI Use Case One-Pager
A single page to complete before committing to an AI project — the problem, the deterministic alternative and its cost, where the data is, what failure looks like, and the number that defines success.
Markdown. No sign-up, no email.
One page, completed before anyone estimates the build. Its purpose is to make the cheap alternative visible, because that is the comparison most projects never make.
If a section cannot be filled in, that is the finding.
Use case: _______________ Requested by: _______________ Owner: _______________ Date: _______
1. The problem#
In the user's own words: _______________
Who has this problem, and how often: _______________
What they do today, and how long it takes: _______________
Annual cost of the current approach (hours × rate, errors, delays): _______________
2. The cheap alternative — fill this in first#
| Option | Would it work? | Effort | Coverage |
|---|---|---|---|
| A rule, or a few rules | ___% | ||
| A form, or better data capture | ___% | ||
| A report or query | ___% | ||
| Fixing the upstream process | ___% | ||
| Buying an existing product |
Baseline coverage achievable without AI: ____% Measured, or estimated? measured / estimated — how: _______________
🔴 If this section is blank, the project is not ready to scope. The most expensive AI projects are the ones that should have been a rule, a form or a query — and nobody found out because nobody spent the afternoon.
3. What AI would add, in units of work#
| Baseline | With AI | |
|---|---|---|
| Cases handled automatically | ||
| Additional cases per day | — | |
| Time saved per day | — | |
| Errors avoided per month | — |
The incremental benefit, in one sentence: _______________
Express this as work, not as percentage points. "Eight points more accurate" persuades; "sixteen documents a day" invites the right comparison.
4. Where the errors land#
On which cases will this be least reliable? _______________
What does a wrong answer cost on those cases? _______________
Does the system fail visibly or silently? visibly / silently
A system that is accurate on easy cases and unreliable on consequential ones is worse than its headline number. If the hard cases are also the expensive ones, say so here.
5. Data#
| Is the knowledge written down? | yes / no / partly |
| Where | |
| Who owns it and keeps it current | |
| Volume available | |
| Labelled examples available | |
| Legal basis / permission to use it | |
| Personal data involved | yes / no |
If the knowledge lives in people's heads, this is a documentation project first. Say so explicitly rather than proceeding.
6. Success#
| Metric | Baseline today | Target | How measured |
|---|---|---|---|
What "good enough to ship" means: _______________ Who decides that: _______________
7. What it costs to keep#
- Inference / usage cost per operation: ____
- Who monitors it: ____
- What happens when it degrades: ____
- Who is accountable for a wrong output: ____
- Review cadence: ____
An AI system is not a project that finishes. If nobody owns the monitoring, it will quietly get worse and nobody will connect the complaints to the cause.
8. Decision#
| Build with AI | |
| Build the cheap alternative first | |
| Do both — rules now, AI if the gap justifies it | |
| Do nothing |
Decision: _______________ Reasoning in two sentences: _______________ What would change this decision: _______________
Sign-off#
| Name | Date | |
|---|---|---|
| Prepared by | ||
| Business owner | ||
| Decision taken by |