AI how-to guides Operations and processes
How to find AI use cases in your company
Start from business priorities and real pain, not from the technology: run structured workshops with business, data and IT, list where work is slow, repetitive or error-prone, and score each idea on value, feasibility and risk. Give every shortlisted case a baseline metric and a named owner. End with a sequenced backlog that delivery can pick up, not a list of ideas.
What AI does
- Groups and de-duplicates the ideas collected in workshops and interviews.
- Analyses process data, tickets or documents to show where time and errors pile up.
- Drafts a first description of each use case: inputs, outputs, users and systems.
- Helps compare opportunities consistently against the scoring criteria.
What stays with people
- Choosing the business priorities that discovery has to serve.
- Scoring value and feasibility and making the final call on what goes first.
- Owning each use case and its success metric.
- Judging risk, ethics and regulatory fit.
How to put it in place
-
Frame the question and the people
Agree the scope, the business priorities and the question discovery must answer. Invite people from business, data and IT who know the processes and can decide.
-
Run structured discovery workshops
Walk through processes and pain points area by area and capture candidate use cases in a common format. Ask where work is repetitive, where decisions wait for information and where errors are costly.
-
Score value, feasibility and risk
Score each opportunity on impact, data availability, integration effort, time to evidence and risk with the same criteria. Flag cases that may fall into regulated or high-risk categories early.
-
Set a baseline and an owner for each case
Measure how the process performs today, so any improvement can be proven later. Name a business owner and the data and IT counterparts for each shortlisted case.
-
Sequence the backlog and hand it over
Order the cases by value and dependencies, add governance checks for data, risk and model lifecycle, and define the first delivery wave. Hand it to the teams that will build it, with the reasoning behind each choice.
What you need
- Sponsorship from management and clear business priorities.
- Time from people who know the processes in business, data and IT.
- Basic process data to set baselines: volumes, times, errors.
- An agreed scoring model and governance criteria.
- Delivery capacity to act on the first cases.
When it is not worth it
- When there is no capacity or budget to act on the results in the next cycle.
- When the AI use case is already obvious and agreed: go straight to a scoped pilot.
- When leadership has not agreed the business priorities the cases should serve.
Risks and how to control them
- Long lists that never become decisions: limit the shortlist, score everything with the same model and leave with a sequenced backlog.
- Value claims without a baseline: measure today's performance before promising improvements.
- Cases with no owner that stall: name a business owner and the data and IT counterparts for each one.
- Regulatory surprises: check early whether a use case touches high-risk areas under the EU AI Act, such as employment, credit or certain insurance decisions, and involve legal before committing.
How Thinkia does it
We frame scope, business priorities and the question discovery must answer, then run structured cross-functional workshops with a ready-made kit of agendas, canvases and facilitator scripts. Each opportunity is scored on impact, feasibility, time to evidence and risk with a transparent model, and carries a baseline metric and named business, data and IT owners.
Discovery ends with a prioritised backlog with dependencies and a lightweight governance gate for data, risk and model lifecycle, so opportunities enter delivery already aware of the AI Act and the next planning cycle can pick them up without reopening the choice.
The Thinkia solution behind this guide
Frequently asked questions
How do you identify AI use cases?
Start from business priorities and processes, not from tools. Look for work that is repetitive, slow, error-prone or waiting for information, and check whether the data needed exists.
How do you prioritise AI use cases?
Score each one on value, feasibility, time to evidence and risk with the same criteria, then order them by score and dependencies. Quick, measurable cases often go first to build trust.
Who should take part in discovery?
People from the business who own the processes, plus data and IT. Without all three, ideas tend to stall at the first technical or data question.
What should come out of an AI discovery?
A sequenced backlog of use cases, each with a baseline metric, an owner, the data and systems it needs and the governance checks it must pass.
Do we need clean data first?
Not for every case. Discovery should show what data each use case needs and whether it is good enough; some cases can start now while others wait for data work.
Keep exploring
Related guides
- How to move an AI pilot to production
- How to automate processes with multiple AI agents
- How to automate AI governance
Related decisions
- AI pilot vs production: what actually changes when you scale?
- Fix the data first or start with AI now: which comes first?
- Build vs buy AI agents: which agents should you own, and which should you rent?