Free practical AI tool

Give your AI idea a useful job.

Define the work, the people affected, the information involved and the human judgement that must remain before you choose a tool. You will leave with a one-page opportunity and the smallest safe test.

Start the canvas

This canvas will not tell you that AI is the answer. It will help you see whether there is a well-defined piece of work worth testing.

Your entries are processed in this browser. They are not sent to Oddly Capable unless you deliberately copy them into the contact form.

Do not enter passwords, authentication codes, personal data or confidential information that does not belong in this canvas.

Work to improveName one task, decision or flow of work. Avoid “use AI in the business”.0 of 2000 charactersPeople involved or affectedWho does the work, reviews it, receives the outcome or carries the consequence if it is wrong?0 of 2000 charactersCurrent trigger and inputWhat starts the work? What information, request or event arrives?0 of 2000 charactersWhat happens nowList the main steps. Include checking, copying, waiting and rework-not just the official process.0 of 2000 charactersCurrent frictionWhat takes time, creates inconsistency, causes errors or depends on hard-to-find knowledge?0 of 2000 charactersUseful outcomeWhat would be observably better for the people doing or receiving the work?0 of 2000 charactersInformation and sensitivityWhat data or knowledge is used? Is it personal, confidential, commercially sensitive, regulated, copyrighted or unreliable?0 of 2000 charactersHuman judgement that must remainWhich decisions, checks or conversations need a person because of context, consequence, responsibility or care?0 of 2000 charactersPossible role for AI or automationDescribe the job, not a product: classify, find, extract, compare, draft, summarise, route or flag.0 of 2000 charactersFailure and edge casesHow could the output be wrong, biased, incomplete, unsafe or confidently misleading? What happens then?0 of 2000 charactersMeasure of successChoose one or two measures tied to the work: elapsed time, rework, error rate, response quality, findability, confidence or user effort.0 of 2000 charactersSmallest safe testInclude the test, owner, people involved, review date, and stop or change conditions.0 of 2000 characters

Review prompts

The work is specific enough to observe.A person owns the test and its outcome.The information can be used lawfully and appropriately.Human review matches the consequence of an error.Success is about better work, not simply using AI.The test is small enough to stop or change safely.

Useful questions remain

The work is specific enough to observe.
A person owns the test and its outcome.
The information can be used lawfully and appropriately.
Human review matches the consequence of an error.
Success is about better work, not simply using AI.
The test is small enough to stop or change safely.

Talk through this opportunity

A good first AI project is small enough to learn from and important enough to matter.

If the canvas exposed several connected problems, a Working Session can help you choose the first one. If the work is defined but needs a prototype, guardrails or team enablement, a Capability Sprint may be the better fit.

See practical AI support →