Skip to content

AI Use Case Workshops: Formats, Agenda and Outcomes

Most companies that decide to "do something with AI" quickly discover that the hard part is not the technology but choosing where to start. A workshop is the usual answer, yet the word covers anything from a two-hour talk for the leadership team to a week-long sprint with a working prototype. This guide compares the common formats by who attends, how long they take and what they actually produce.

What is an AI use case workshop?

An AI use case workshop is a facilitated working session in which a team identifies, describes and ranks specific tasks in its own organisation where AI could help. Unlike a training course, it ends with a work product: a short, prioritised list of documented candidates and an agreed next step.

The workshop is one of several ways to find use cases. Process walk-throughs, interviews with the people doing the work and analysis of system logs often supply the raw material that a workshop then condenses. For an overview of those methods and where each one falls short, see Finding AI Use Cases: Methods That Work in Mid-Sized Companies. This guide focuses on a narrower question: which workshop format fits your situation.

Three things separate a useful workshop from a brainstorm:

  • Real material. Participants work on actual artefacts such as a quotation, a service ticket or an inspection report, not on buzzwords.
  • One template. Every candidate is written up the same way, so ideas can be compared fairly.
  • A decision at the end. Someone with authority picks the first candidate to test and names who owns it.

What workshop formats are there?

Five basic formats cover most situations. They differ in length, in who attends and in how deep the result goes. They also combine well: a leadership briefing often leads to a one-day use case workshop, which in turn leads to a deeper session on the top candidate.

FormatWho attendsLengthWhat you get
Leadership briefingexecutive teama few hoursshared expectations, broad themes
Use case workshopbusiness units, IT, decision-makerone dayshortlist of comparable profiles
Department deep diveone team and its process ownerhalf a day per teamdetailed candidates for one area
Sprintsmall fixed team incl. decision-makerseveral daysprototype tested with users
Build daydomain experts plus developersone to two daysworking feasibility check

The lengths are typical ranges, not rules. What matters is that the format matches the result you expect. A two-hour briefing will not produce a ranked roadmap, however good the facilitator.

Leadership briefing

A leadership briefing aligns expectations at the top. It is useful when views range from "AI will replace whole departments" to "it is all hype". The outcome is a set of themes and a decision about where to look more closely. It rarely produces use cases you can rank, because the people who know the processes in detail are not in the room.

One-day use case workshop

The one-day workshop is the default choice when you want to leave with a shortlist you can defend. Experts from two or three teams, someone from IT and a decision-maker spend a day on real workflows. The day follows a fixed arc: shared basics, collecting pain points, condensing, ranking, committing. A separate guide on a one-day workshop agenda goes into the timetable in detail.

Department deep dive

A deep dive concentrates on one area, such as field service or sales quotations. With the process owner and a few experienced staff, the team walks through the workflow step by step. The candidates are more precise than in a broad workshop, but they cover only part of the business. Deep dives work well once you know where to look, or as a series after a leadership briefing. The overview of AI use cases by department is good preparation material; the deep dive then tests which of those ideas hold up in your own workflow.

Sprint

The sprint format comes from the Design Sprint, which GV describes as a five-day process: set a goal and map the problem on Monday, sketch on Tuesday, decide and storyboard on Wednesday, prototype on Thursday and test with customers on Friday. For AI projects it suits a single candidate where the open question is whether people would actually use the result. It is too heavy a tool for finding many use cases.

Build day

On a build day, domain experts sit with developers to answer a technical question quickly, for example whether a language model can reliably extract the right fields from fifty real delivery notes. The answer is technically meaningful but says little about business value or adoption. It also needs approved test data and someone to provide the environment.

How does a good workshop work? The building blocks

Whatever the format, effective AI workshops share the same building blocks: a shared starting point, a structured way to collect input, one profile template, a transparent ranking and a binding close. Drop one of them and you get either an unsorted idea list or a decision without a basis.

Shared starting point. Participants do not need prior knowledge, but they need a common baseline. A few live demonstrations on familiar material work better than slides about model architecture, for example a model summarising an incoming enquiry and routing it to the right product team, including a case where it gets it wrong.

Structured collection. Open discussion favours the loudest voices. Liberating Structures offers simple counter-measures. In 1-2-4-All, everyone first reflects silently, then shares in pairs, then in groups of four, and only then does the whole group hear the strongest points. The site lists it at roughly 15 minutes and says it works with four or more people.

One profile template. Every candidate is described the same way: trigger, inputs, desired output, current cost of errors, frequency, owner and affected systems. A candidate without a complete profile does not get ranked.

A small set of criteria. Value, data availability, technical feasibility, risk and organisational readiness are the usual ones. A consistent set of criteria with written reasons matters more than an elaborate scoring formula.

A binding close. The session ends with a decision: which candidate is tested first, by whom, by when, and what result would make you stop.

How do you choose the right format?

Three questions usually narrow the choice to one or two formats. How clear is it already where to start? Who is available, and for how long? What result do you need to make the next decision? Answer those honestly before you book a room.

Who attends. A workshop is only as good as the knowledge in the room. To produce rankable use cases you need three perspectives: someone who runs the process every day, someone who understands the systems and data, and someone who controls budget and priorities. Without the operational view, results stay superficial. Without the decision-maker, the shortlist goes nowhere. The HPI School of Design Thinking makes a similar point about its own method, stressing that different perspectives, experience and skills are essential (page in German).

Length. Short formats are easier to schedule but go less deep. In our own operations, filling in and ranking one profile properly as a group takes far longer than coming up with the idea. Plan for a few candidates done thoroughly rather than many done superficially.

Depth of outcome. It helps to distinguish three levels. Themes are broad and have no data or owner attached. Profiles are comparable and ranked. Validated candidates have been tested on real data or with real users. Each level costs more time than the one before and answers a different question.

Starting pointFormat that fits
leadership disagrees on what AI can doleadership briefing
goal is clear, entry points are notone-day use case workshop
area is chosen, details are missingdepartment deep dive
one candidate, adoption is uncertainsprint
one candidate, feasibility is uncertainbuild day

What do you need in place before the workshop?

The key prerequisites are organisational rather than technical: a clear question, the right people, real examples from daily work and a commitment that time for a test will be available afterwards. Technical infrastructure only matters for build-day formats.

A sharp question. Write down what the workshop must answer, for instance: "Which two service tasks do we test next quarter?" A goal such as "explore the potential of AI" almost always produces a long, non-committal list.

Material from daily work. Ask teams beforehand for real, anonymised examples: a typical quotation, an export of service tickets, an inspection report. With that material in the room, every idea can be checked against reality straight away.

Baseline AI literacy. In the European Union, Article 4 of the AI Act has applied since 2 February 2025. It requires providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff. A workshop is not a training programme, but it can be one documented element of one. This is general orientation, not legal advice.

Employee representatives. Where a works council or a similar body exists, it may have information and consultation rights when new technology changes how people work. In Germany, Section 90 of the Works Constitution Act explicitly names the use of artificial intelligence in work processes and requires the employer to inform the works council in good time and consult it early enough for its suggestions to be taken into account. Inviting representatives to the workshop, or briefing them right after it, avoids reopening chosen candidates later.

How does a use case workshop run, step by step?

A one-day workshop runs in five steps: preparation, a shared introduction, collecting pain points along real processes, condensing and ranking, and committing to a next step. The follow-up in the days afterwards is part of the job, because that is where a shortlist either turns into a test or quietly fades.

  1. Prepare (one to two weeks before): set the question, invite participants, collect process examples, prepare the profile template and criteria.
  2. Introduce: a short, concrete demonstration of what AI does well and badly on familiar documents. Clarify expectations.
  3. Collect: gather friction points along the processes people brought, using a structure such as 1-2-4-All. Collect problems, not solutions.
  4. Condense and rank: cluster the findings, complete profiles for the most promising candidates and rank them on the agreed criteria.
  5. Commit: the decision-maker names the first test candidate, an owner and a date for the first check-in.

The People + AI Guidebook from Google PAIR is a useful companion for the question of whether AI is the right tool at all. It warns that a rule- or heuristic-based solution often works just as well as an AI version, if not better, and lists situations where AI is probably not the better choice, including when predictability matters or when the cost of errors is very high. That question belongs on every profile.

The same guidebook distinguishes automation from augmentation. Automation suits tasks that are difficult, unpleasant or need scale; augmentation suits tasks people enjoy or where personal responsibility for the outcome matters. Recording that choice for each candidate during the workshop pays off, because it shapes both effort and acceptance.

What should a workshop produce?

A use case workshop should end with three things: a short, ranked list of complete profiles, written reasons for the ranking, and an agreed first test with an owner, a date and a stop criterion. Photos of sticky-note walls are working material, not an outcome.

A complete profile covers at least:

  • Task: what happens today, what triggers it and what it produces.
  • Data: which documents or systems supply the input, in which format and where they live.
  • Current effort and failure modes: how often the task occurs and what an error costs, as estimated by the people who do it.
  • Role of AI: automate or assist, and what human check remains.
  • Risks: data protection, employee consultation, regulatory classification, dependence on individual people.
  • Ownership: who is accountable for the use case on the business side.

From these profiles you can tell what kind of solution is worth exploring: a specialised AI agent for a business process, for example in technical service or quotation, or a simpler workplace tool first. Where sensitive data is involved, where the system runs becomes a question of its own, covered under sovereign AI.

What are the limits and risks of AI workshops?

A workshop can only be as good as the knowledge and data in the room. The usual weaknesses are a goal that is too broad, missing operational staff, ranking by gut feeling and no follow-up date. Knowing these in advance lets you design the format around them.

Solutions instead of problems. Many participants arrive with a tool in mind ("we need a chatbot"). If you collect tools rather than friction points, the list fills up with answers to questions nobody asked.

Optimism without a data check. In a workshop, almost every idea looks feasible. Whether the data exists in usable form often only becomes clear afterwards. Make each profile state what is known about the data and what is assumed.

Hierarchy in the room. When the most senior person names a favourite early, other ideas go quiet. Silent individual work at the start and a facilitator with no stake in the result both help.

Facilitators selling a product. If the facilitator sells a particular tool, it is no surprise when that tool comes out on top. Ask beforehand how the ranking stays neutral.

A workshop instead of a decision. A workshop prepares a decision but cannot replace it. If it ends without a binding next step, it was mainly an information session.

Candidates involving employee data. Ideas that would evaluate staff performance or behaviour need separate legal review and early involvement of employee representatives. Flag them in the profile rather than quietly ranking them high.

How do you get started?

Start with the question the workshop has to answer, then pick the format. For most organisations without previous AI projects, a one-day use case workshop across two or three teams is the most practical entry point, followed by a test of the top candidate within a few weeks.

A short checklist:

  • Phrase the goal as a question with a time frame.
  • Choose two or three teams and name one person from each who does the work daily.
  • Decide who makes the final call and make sure they attend the close.
  • Collect real, anonymised examples beforehand.
  • Brief employee representatives on the goal and scope where they exist.
  • Reserve time for the first test before the workshop takes place.

If you would rather not facilitate it yourself, the AI workshop page outlines how a facilitated session is structured. For what comes after the workshop, from the first test to running systems, the page on a fractional head of AI is the natural next read.

Frequently asked questions

How long should an AI use case workshop be?

A full day is a realistic frame for a shortlist of comparable use case profiles. A two- or three-hour session builds shared understanding but rarely produces candidates you can rank.

Who should be in the room?

At least one person who does the work every day, one who understands the systems and data, and one who can commit budget and people. Without the decision-maker, the shortlist tends to stay on paper.

Is an ideation workshop the same as AI training?

No. Training builds knowledge; a use case workshop produces a work result. A short, concrete introduction is part of a good workshop, but the bulk of the time goes into real processes.

Should employee representatives be involved?

Where a works council or similar body exists, early involvement avoids renegotiating selected use cases later. In Germany, for example, the Works Constitution Act requires timely information about plans that involve AI in work processes.

What should happen in the weeks after the workshop?

The top candidate gets a small, time-boxed test on real data, with a named owner and a clear stop criterion. Booking that test before the workshop ends is the simplest way to keep momentum.

Sources

  1. GV: The Design Sprint gv.com
  2. Liberating Structures: 1-2-4-All liberatingstructures.com
  3. Google PAIR: People + AI Guidebook – User Needs + Defining Success pair.withgoogle.com
  4. HPI School of Design Thinking: The Design Thinking Process (German) hpi.de
  5. EU AI Act, Article 4: AI literacy artificialintelligenceact.eu
  6. German Works Constitution Act (BetrVG), Section 90 gesetze-im-internet.de