Skip to content
Guillermo AyalaAI consulting & development

Blog · AI for small and midsize businesses

A 90-minute workshop to find AI opportunities with your team

A productive AI workshop ends with a few business problems worth testing and named owners for the missing evidence. This agenda keeps the discussion grounded in actual work, gives frontline participants a voice and turns interesting ideas into a short, reviewable backlog.

Cross-functional team arranging business-problem cards around a round table.
AI-generated editorial illustration.
In this article
  1. Prepare before people enter the room
  2. The ninety-minute agenda
  3. The opportunity card
  4. Prevent three predictable problems
  5. Send a usable result the next day
  6. Related reading

An AI workshop can easily become a contest for the most impressive idea. A more useful outcome is three candidate tasks with evidence, owners and an agreed next check. Ninety minutes is enough to produce that shortlist when participants arrive with examples; it is not enough to validate a business case or promise delivery.

I adapted this agenda for a small team using the discovery and prioritization principles in OpenAI's use-case guide. Keep the session focused on one department or workflow. Mixing every business function usually makes comparison less meaningful.

Prepare before people enter the room

Invite a process owner, two people who perform the work, a subject expert and someone who understands the systems. One person can cover several roles in a small company. Name a facilitator who can stop tool discussions and a recorder who captures assumptions without presenting them as facts.

Each participant brings two examples of recurring work: a permitted sample input, the expected output and a rough account of frequency and difficulty. Remove unnecessary personal information. The process owner brings the team's main operational objective, such as reducing delayed quotations, and explains what decisions the group can make.

Prepare a board with four areas: observed problems, candidate tasks, unanswered questions and shortlisted opportunities. Keep a separate parking list for product names and unrelated ideas. A sales demonstration should not occupy the workshop's evidence-gathering time.

The ninety-minute agenda

  1. Minutes 0-10: agree on the problem boundary. State the workflow, users and desired improvement. Explain that suggestions will be evaluated as tasks, with no commitment to a particular technology.
  2. Minutes 10-25: silent problem collection. Participants write individual observations before discussing them. One card contains one task, a concrete example and its effect. This reduces the influence of senior voices.
  3. Minutes 25-40: group by underlying cause. Merge duplicate tasks and distinguish missing information, repetitive drafting, rules, waiting for permission and specialist decisions. Preserve disagreements as questions.
  4. Minutes 40-55: complete opportunity cards. Pairs describe the input, proposed output, accountable owner and first test. Return vague cards for clarification.
  5. Minutes 55-70: compare impact and effort. Use consistent low, medium and high definitions. Flag data access and consequential errors separately so enthusiasm cannot hide them.
  6. Minutes 70-82: challenge the leading candidates. Ask what would make each fail, whether a simpler process change would help and which difficult example should be tested first.
  7. Minutes 82-90: assign evidence and next decisions. Select up to three candidates. Name who will obtain each missing fact and when the owner will review the result.

The opportunity card

Use the following prompts on a single page: who experiences the problem; what event starts the task; which inputs are required; what an acceptable output contains; observed frequency and handling time; where judgment or permission is necessary; proposed first test; and the owner authorized to accept the result. Add an evidence label to each assertion: observed, estimated or unknown.

Consider a hypothetical maintenance company where participants propose an autonomous scheduling assistant. During the card exercise, staff reveal that equipment availability is updated manually and emergency jobs follow unwritten rules. The group narrows the idea to drafting a morning handover summary from approved notes. Operations owns review, while the scheduling proposal remains dependent on better availability data.

The workshop has exposed a dependency and identified a testable task. It has not proved that summarization will save time. The team still needs representative samples and a comparison against its current handover method.

Prevent three predictable problems

If people vote for their own department's idea, return to the shared operational objective. If every task receives high impact, require an observable effect such as missed deadlines or avoidable corrections. If the group confuses clean demonstrations with feasibility, ask for the least convenient real input.

Do not average away an unresolved information-access issue. The NIST generative AI profile provides a risk-management reference; the workshop's practical responsibility is to record who must resolve each issue before testing.

Send a usable result the next day

The recorder sends the shortlist, completed cards, open questions and named follow-up owners. The process owner corrects factual misunderstandings and confirms the next review. Keep rejected ideas with a brief reason so the same debate does not restart every month.

Judge the workshop by the quality of its decisions: are the tasks specific, are assumptions visible, and can someone run the next check? If you need facilitation that connects business questions with technical constraints, start with a focused AI strategy workshop.

Back to the blog · Practical AI guides