AI discovery
AI discovery before investing in development
We turn a broad AI idea into a testable use case, data map and informed decision about a pilot.
Discuss your project →The decision to make
Discovery is useful when opportunities are broad and no specification exists. With the process owner, we define the user, repeatable task, current workflow and a verifiable outcome. If rules, search or conventional automation are sufficient, we document that without adding AI complexity.
- Challenge and user definition
- Process and decision-point map
- Available data sources
- Quality criteria and AI boundaries
- Decision: pilot, prepare data or stop the hypothesis
What we examine
We review sample inputs, freshness, permissions, constraints and the source of a correct answer. We identify error consequences, actions requiring human confirmation and dependencies on existing systems. This is not a full security audit or legal opinion, but it surfaces the questions that must be resolved before development.
What your team retains
The result is a documented challenge definition: target use case, users, sources, acceptance criteria, constraints, preliminary architecture and pilot boundaries. It is not tied to a particular model or supplier and can support an internal funding decision.
Fit before scope
Is this the right starting point?
Use these criteria for an initial orientation. The final recommendation follows a review of your context, data and constraints.
A good fit when
- A process or decision that needs improvement is identified
- An accountable sponsor and result user are available
- A measurable outcome or quality criterion can be defined
Resolve this first when
- The goal is simply to buy unspecified AI without a defined challenge
- A guaranteed impact is expected before data are assessed
- No one is accountable for accepting the result
A verifiable first stage
What you can use to make the next decision
The exact scope is agreed before work begins. These are typical decision artefacts, not promised business results.
AI use-case matrix
Challenges compared by value, data availability, risk and ease of verification.
Pilot boundary
One process with defined users, sources, human oversight and stop conditions.
Decision criteria
Quality measures and a process to scale, refine or stop the initiative.
Five questions · no datasets or credentials
Frequently asked questions
Frequently asked questions
Do we need a technical specification?
No. Its absence is a common reason to start with discovery. You need a process owner and the ability to show examples of real work and data.
Does discovery guarantee a pilot?
No. A useful outcome may be to prepare data first, change the use case or avoid AI for this task.
What data should we provide initially?
We begin with structure, descriptions and minimal anonymised samples where sufficient. Access to sensitive data is agreed separately.
Ideas on this topic