- Why Most AI Use Case Lists Miss the Work
- Two Ways to Build a List of AI Use Cases
- How to Identify AI Use Cases Across Every Team
- Merging the List Across Teams
- What to Do With the Rest of the List
- One Bus Operator, One Merged List
- Identifying AI Use Cases: What to Settle Before You Ask
- Where GrowthNation Fits
- Build the List From Every Team, Not One Meeting
- Frequently Asked Questions
- How do you identify AI use cases?
- Should AI use cases come from leaders or from staff?
- How do you compare AI use cases from different departments?
- What is the difference between an AI use case and a task?
- Is staff use of unapproved AI tools a source of use cases?
- How often should an AI use case list be refreshed?
Quick Answer: To identify AI use cases across an organisation, ask every team to describe its actual work before anyone runs a workshop or fills in a scoring grid. Then merge the same task found under different names, add up the effort across every team that does it, count the AI people already use, and sort the rest into build, park, fix another way or drop.
Among the 104 English councils that answered a 2025 survey, 43 per cent named a lack of clear use cases as a barrier to deploying AI; only 13 per cent named a lack of senior leadership buy-in. Of those already paying AI suppliers, two thirds said understanding where AI can add value was a barrier. The shortage is rarely the technology; it is a list of work the organisation trusts. Yet most advice on how to identify AI use cases starts in a room: a workshop, a supplier’s catalogue, a grid scoring impact against effort.
That order is backwards. A room can only score the ideas that reach it. Choosing the running order is a real decision, but it is made from a list, and the list has to come from the people who do the work, team by team.
Why Most AI Use Case Lists Miss the Work
In a 2025 interview, a clinical academic at Imperial College London described suppliers arriving with a tool and asking whether a use could be found for it, when the first question should be what problem needs solving. It is one clinician’s view of the NHS, but the pattern shows in how most lists get made:
- The supplier’s catalogue sets the menu. Published examples show what agents can do, but not which of your tasks are worth it.
- A workshop hears the work people will describe in public. Dull or awkward jobs, and work done by people not invited, rarely come up; the parts of a job that are hard to explain even less.
- Ideas come back as headings, not tasks. “AI for complaints” says nothing about who does the work, how often, or what it starts from, so it cannot be counted or compared.
- The AI people already use goes unmentioned. Few staff volunteer, in front of their manager, the tools nobody approved, even when those tools do useful work.
A scoring grid can only rank what reached it. The first step to identify AI use cases worth building is widening who gets asked, not refining the grid.
Two Ways to Build a List of AI Use Cases
One case study compared both routes in a single organisation. Ideas from leaders working top-down promised short-term, incremental improvements. Ideas gathered bottom-up were broader and included genuinely new ones, but needed far more upfront effort. Neither route alone gives a usable list: one misses work, the other needs shaping.
| Built in a room | Built from every team | |
| What gets written down | A heading such as “AI for invoices” | A task described by the person who does it, awkward cases included |
| Whose effort counts | The work of whoever is in the room | The work of everyone who does that task, in any team |
| What a score rests on | A guess made in the meeting | What people said the task takes them each week |
| The same task in two teams | Scored twice, or missed twice | Merged into one candidate with its total effort |
| AI already in use | Rarely mentioned | Named and counted, so the useful parts can be held properly |
| Ideas that do not go first | Forgotten after the workshop | Parked with a reason, and the team is told |
Scoring still has a job, but after the list exists, not in place of it.
How to Identify AI Use Cases Across Every Team
Researchers from Tampere University ran sixteen group interviews across nine functions of a mid-sized European energy company and found 41 candidate use cases. It is a preprint about one company, a senior contact chose who took part, and none had reached daily use when it was written. It shows what asking function by function yields, starting with manual, repetitive work that rarely reaches a strategy document. To identify AI use cases this way across a whole organisation, three habits matter.
Ask About the Work, Not About AI
Ask people for AI ideas and they repeat what they have seen in demonstrations. Ask them to walk through a recent week and they describe where the time goes, what they type twice and who they wait for. The questions resemble those used to automate one person’s tasks; the difference is asking every team.
Count the AI People Already Use
In YouGov’s poll of March 2026, 32 per cent of British workers said they use AI for their job; among them, 71 per cent use tools their employer neither requires nor prohibits. The German Economic Institute found 29.1 per cent of 2,793 working AI users in its 2026 survey used tools their employer had not provided, and read that as a demand signal, not misconduct. Each marks a task someone has already tested. Ask about them in confidence and settle later who keeps them.
Write Every Candidate in the Same Shape
A finance candidate cannot be compared with an engineering one unless both are written the same way: what starts the task, what it produces, who receives it, and the cases that do not fit. That shared shape is what lets you identify AI use cases across teams rather than collect headings. A candidate written this way is ready to compare, and later to pick for a first pilot.
Merging the List Across Teams
For its user research report on generative AI in schools, the Department for Education’s research partners consulted 710 school-based practitioners for a longlist, which a steering group cut to 12 use cases. It is independent research, not policy, but the shape is useful: many voices, one list, a clear point where it is cut. Before that point, four steps turn many answers into one list:
- Put the same task under one name. Engineering’s “parts chasing” and finance’s “supplier queries” may be the same follow-up seen from two ends.
- Add up the effort across every team that does it. A task that looks small in each team can be large across the organisation once the weekly hours are added up.
- Keep the differences, not only the overlap. Each team’s exceptions are the cases an agent will meet, so keep them with the merged candidate.
- Mark candidates that depend on another team’s system. Those need the other team involved before anything is built.
Only after merging does ranking make sense, because each candidate is then judged once, at its full size.
What to Do With the Rest of the List
Staff ideas need more upfront effort, as the same case study found, so most of a good list will wait. How it waits decides whether people answer the next time you identify AI use cases:
- Park it with a reason and a date. “Waiting for the new rostering system” is a reason; “not now” is not.
- Fix the step instead when a form or a system causes the work. Some candidates disappear with a simple fix or a better template.
- Tell each team what happened to what it raised. People who never hear back stop offering ideas, and winning trust back takes longer than a reply.
- Ask again when the work changes. A new system, a new rule or someone leaving can turn a parked candidate into the next one to build.
One Bus Operator, One Merged List
A regional bus operator with around 900 staff across four depots began with a leadership workshop, which produced three headings: AI for complaints, AI for maintenance and AI for rostering. To identify AI use cases from the work instead, it interviewed scheduling, engineering, customer services, driver rostering and finance, and got a far longer list, much of it the same work in different words.
| Candidate | Raised by | How teams described it | AI already in use | Where it sits now |
| Lost property and complaint triage | Customer services; depot managers | “Sorting the inbox”; “chasing depots for items” | An adviser’s personal chatbot | First build, held centrally |
| Defect reports into job cards | Engineering; drivers’ supervisors | “Retyping the defect book”; “chasing missing reports” | None | Second build |
| Driver swap requests | Driver rostering; depot managers | “The swaps spreadsheet”; “texts at 5am” | None | Parked until the new rostering system |
| Parts and supplier follow-ups | Engineering; finance | “Parts chasing”; “supplier queries” | None | Merged into one candidate |
| Timetable change notices | Scheduling; customer services | “Rewriting notices for each route”; “answering the same question” | A scheduler’s personal chatbot | Fixed another way: one template and a rules-based step |
| Concessionary fare claim checks | Finance | “Monthly claim checks” | None | Dropped for now: too few a month, and finance was told why |
None of the three headings survived as written. Complaints became the triage candidate, maintenance the defect and parts work, and rostering waits for a system change. Two candidates already ran on personal chatbots, so an unapproved habit became the first build, now held in one central place.
Identifying AI Use Cases: What to Settle Before You Ask
Agree these before the first conversation:
- Decide who hears the answers. Confidential conversations are what bring unapproved tools into the open.
- Agree one shape for every candidate. Settle the fields before the first interview.
- Say plainly that the list is not a headcount plan. People describe their work honestly only when they believe it.
- Promise no penalty for tools already in use. The aim is to learn what works.
- Give the merged list one named owner. Someone has to merge, park and report back.
- Set the date the list will be asked for again. A list never refreshed drifts from the work.
Where GrowthNation Fits
GrowthNation’s AI agent interviews everyone in every team, by voice or text and in confidence, so the list of candidates comes from the work rather than from a meeting room. Agents built from that list are served centrally and belong to the organisation, not to the account of whoever built them. Hours saved are tracked for each agent in daily use against the starting count.
If your organisation already has a well-evidenced list and needs engineering capacity to build it, a delivery partner may be the better help.
Build the List From Every Team, Not One Meeting
Ask two teams that never share a meeting to walk through last week, then compare their answers with your current list. The gaps show where to identify AI use cases no workshop surfaced.
Book a walkthrough and GrowthNation will interview your teams, then hand you three things:
- One merged list of candidates, in the words of the people who do the work, duplicates combined.
- The AI already in use, which tools and for what, so the useful parts can be held properly.
- A working agent, built from the top candidate and held by your organisation.
Hours are measured against where you started, so every saving can be checked.
Book a walkthrough: every team asked, one merged list, one working agent.
Join the GrowthNation waitlist. Organisations are brought on in batches, so a place now means your teams are interviewed in the next round.
Frequently Asked Questions
How do you identify AI use cases?
To identify AI use cases, ask every team to describe its actual work before running workshops or scoring ideas. Write each candidate in the same shape, merge the same task found in different teams, add up its effort and count the AI tools staff already use. Then rank the merged list and decide what to build, park, fix another way or drop.
Should AI use cases come from leaders or from staff?
Both, but ask staff first. Staff know where the time goes; leaders know the goals and limits. One case study found leaders’ ideas tended to be incremental and quick to deliver, while staff ideas were broader and newer but needed more shaping.
How do you compare AI use cases from different departments?
Describe every candidate in the same shape, merge duplicates and add up the effort across every team that does the task. Then compare them all on one set of measures, agreed before anyone sees which department comes out on top.
What is the difference between an AI use case and a task?
A use case is often a heading, such as AI for complaints. A task is a specific piece of work with a start, an output and someone who receives it. Tasks can be counted, tested and compared; headings cannot.
Is staff use of unapproved AI tools a source of use cases?
Yes, if handled in confidence. In a March 2026 YouGov poll, 71 per cent of British workers who use AI for their job used tools their employer neither required nor prohibited. Each shows a task someone has already tested.
How often should an AI use case list be refreshed?
On a fixed cycle agreed at the start, such as every quarter, and whenever the work changes. Check parked candidates each time, because the reason for parking them may no longer apply.