Define useful work
AI use case template: connect the first project to the bigger goal
An AI use case describes a piece of work you want to change. It should also explain why that change matters to the business. Use this template to connect a wider goal, such as serving more clients, to a first project your team can build and judge.
Describe the whole job, how it helps the business and what the next project could reuse.
Define your AI use case
Keep the first version short. Someone should be able to read it and explain the idea back to you without seeing a software demo.
Choose a first project that changes something meaningful
For a service firm, the goal might be to offer ongoing advice to clients between major projects. A useful first workflow could turn a client request into a plan that an expert approves.
For a large business, regional teams might wait days for help from a central specialist team. The first workflow could prepare an answer from approved methods and send unusual cases to a specialist.
Both examples show how a first build helps more people use what the business knows. A meeting summary might help someone today, but that alone does not explain a larger investment.
Watch the whole job before choosing the part AI will do. If a request waits three days for approval, faster drafting may barely change the wait.
Completed example: ongoing planning support
This is a made-up proposal for a service firm. The same structure could support an internal expert service, with a department as the user and budget holder.
Worked example
An ongoing client-planning service
Goal
Help more clients apply our methods between major projects.
Demand: not yet proved. Check what clients need and would buy.
Owner: the person responsible for the service.
First workflow
A client submits a request to update one part of its plan.
The system finds the current agreement, goals and approved methods.
It checks for missing details and prepares a draft plan with source links.
An expert checks the advice and approves what the client receives.
The job ends with an approved plan and a record of the decision.
Outside this build
No changes to client systems, automatic spending or new service promises.
Other types of advice and a full client portal need separate agreement.
Required information
Current agreements, client goals, methods and previous approved decisions.
Keep each client's records separate. Name who maintains each source.
If something is unclear
Ask for the missing information or send the question to an expert.
Do not guess which agreement or rule applies.
What could follow
Progress reviews could use the approved plan and the same client records.
They would need fresh data, new checks and a separate delivery decision.
First test
Can the system apply the right method and spot when expert help is needed?
Use ordinary requests, missing agreements and conflicting instructions.
Count expert review and correction time.
Also test client demand and the cost of supporting the service.
Next decision
Agree the test and its spending limit before building the full service.
Name the decisions people keep
Write down who checks the advice, who can approve a change and who acts on it. Those can be different people.
In a large business, a central expert may approve the method while a regional manager approves a local action. The system needs to know both rules.
Our journalist-request build sorts requests and drafts replies. At launch, a person approved every outgoing message. That is a delivered example of dividing the work between AI and people.
Compare ideas by the change they support
Ask which idea matters most to the business and what facts support it. Then ask what it builds that a later project could use. Name that later use so people can check the link.
An idea should not win just because it is easy to show in a demo. Avoid a project so broad that it needs every system to change first.
Use normal software for clear rules and calculations. AI may help with varied requests and written knowledge. Both can be part of the same workflow.
Our workshop helps your team compare these choices and choose the first project. Detailed mapping and design then form part of the agreed build project.
Further reading
Written by Said Lopez, founder of LeverageAI and author of Lean AI. Our guides draw on the method in the book and our work with clients.