AI Use Case Discovery
A structured workshop that converts a long list of AI ideas into a ranked shortlist with feasibility and value scored.
Ask ten people in a company where AI would help and you get thirty answers, most of them either already solved by a spreadsheet or impossible with the data available. Discovery is the work of separating those two groups before anyone commits engineering time.
The format
One preparation phase, one workshop day, one analysis phase. The workshop runs best with eight to fourteen participants drawn from the operational teams who do the work, not only from leadership. People who process the invoices know things about the invoices that the process documentation does not record.
Before the workshop
We interview five to eight people individually, review process documentation, and pull a preliminary list of candidates from what we find. Arriving with a starting list changes the workshop from a blank page exercise into a critique exercise, which produces better material in less time.
The workshop
Half the day is expansion: mapping where time goes, where decisions get delayed, where the same information is retyped into a second system. The second half is compression, scoring each candidate live so participants can see the trade offs and argue about them in the room rather than in email afterwards.
After
We verify the top candidates against reality. That means checking whether the data actually exists, whether the volumes justify the build, and whether anything in the process is legally constrained in a way the participants did not mention.
How candidates are scored
| Axis | Question it answers |
|---|---|
| Value | What is the annual value if this works, expressed in hours, error rate or revenue rather than in adjectives? |
| Data feasibility | Does the required data exist, at sufficient volume and quality, and can we legally use it for this purpose? |
| Technical difficulty | Is this a solved pattern or research? Solved patterns ship. Research does not belong in a first project. |
| Integration cost | How many systems must be touched, and how hard is each to reach programmatically? |
| Risk exposure | What happens when the system is wrong, and who is harmed? This drives regulatory classification. |
| Change load | How much does someone's daily work have to change for the value to be realised? |
The last axis is the one most commonly skipped and the one that most commonly kills projects. A model with excellent accuracy that requires forty people to change a habit will underperform a mediocre model that fits into an existing click path.
What you receive
- A full register of every candidate raised, including the ones we recommend against, with the reason recorded.
- A scored ranking of the shortlist across all six axes.
- A one page brief for each shortlisted candidate: problem, proposed approach, data required, rough effort, main risk.
- A short list of quick wins that need automation rather than AI, which we flag separately so you do not pay for a model you do not need.
Some of your best candidates will not need AI
A meaningful share of what surfaces in discovery is better solved with a rule, an integration or a form change. We say so. Recommending a language model for a problem that a validation rule solves is how agencies burn trust in month four.
Common outcomes
Teams usually arrive convinced their priority is a customer facing assistant and leave with an internal document processing candidate at the top of the list. Internal use cases tend to score higher early because the risk exposure is lower, the data is more accessible, and the affected users can be trained directly. Customer facing work is often the right second project rather than the right first one.
Output
Scored use case register, ranked shortlist, one page briefs. Delivered in editable formats. Full IP transfers to you on final payment.
FAQ
Who should attend the workshop?
A mix. Two or three people from leadership for context and decision authority, and six to ten people who do the operational work. If only leadership attends, you get ideas that sound good in a strategy deck and fail on contact with the actual process.
Can this run remotely?
Yes, though in person produces better material. If remote, we split the workshop into two shorter sessions on separate days, because attention drops sharply after about three hours on a call and the second half is where the hard scoring happens.
What if we already have a list of use cases?
Then we spend less time on expansion and more on verification and scoring. Bring the list. We will check whether the data supports each one and rank them, which is usually where existing lists are weakest.
Does discovery commit us to building anything?
No. The output is a decision document. Plenty of clients take the shortlist to their internal team or to another provider. The register is written to be usable by anyone.
Often paired with
AI Readiness Assessment
A two week audit that scores whether your organisation can actually support the AI you want to build, and tells you what to fix first.
Read moreAI ROI Analysis and Business Case
A defensible financial model for an AI initiative, built on your numbers, with the assumptions exposed rather than hidden.
Read moreAI Proof of Concept Development
One narrow use case built against real data in four to six weeks, judged against a threshold set before work begins.
Read moreIs this the right engagement?
Tell us what you are trying to decide. If a different service fits better, or if you do not need us at all, we will say so.