An AI workshop is typically a half-day or full-day session with a mixed group: operational staff who know what problems exist, technical staff who know what AI can do, and leadership who need to make investment decisions. Making that combination productive is a facilitation challenge as much as a technical one.

Preparation

The workshop itself is not where the work starts. Preparation separates useful workshops from ones that produce post-it notes nobody acts on.

Pre-workshop interviews - Speak to 3-5 participants before the day. Ask: what are your most repetitive processes? Where does information get lost or delayed? What would you do if you had twice as much staff? These answers seed the brainstorm and tell you where resistance or excitement is likely to come from.

Context document - A one-page summary of what AI can realistically do today, with specific examples from the relevant industry, reduces the time spent on capability education during the session. Distribute it in advance; do not read it aloud.

Stakeholder mapping - Know who has decision-making authority, who has institutional knowledge, and who might be concerned about automation and their role. Seat accordingly. Design exercises so senior voices do not automatically dominate the first hour.

Agenda Design

A productive full-day workshop agenda:

  • 0:00-0:30 - Context setting: what AI can and cannot do, relevant examples, framing the day
  • 0:30-1:30 - Process mapping: facilitated walk-through of 2-3 key operational processes, identifying pain points
  • 1:30-2:30 - Use case brainstorm: structured ideation, participants generate candidates individually before group discussion
  • 2:30-3:30 - Prioritization: impact/feasibility scoring, voting, and discussion
  • 3:30-4:00 - Next steps: who does what, what needs to be assessed further, what is the path to a PoC

Resist the urge to fill all the time with presentation. The ratio of facilitated discussion to presentation should be at least 2:1.

Brainstorming Techniques

Generic brainstorming produces generic ideas. Structured techniques produce more actionable outputs:

Trigger-based brainstorming - Give each table a category trigger: “Where do you copy-paste data between systems?”, “Where does something sit in someone’s inbox waiting to be processed?”, “What report takes the most time to produce?” These triggers surface concrete process problems, not abstract wish lists.

Role reversal - Ask participants to describe the process from the perspective of the data: “You are the invoice. Where do you get born, where do you sit, where do you get stuck?” Produces operational insight that direct questions often miss.

Prioritization

Use a 2x2 grid: impact (high/low) vs. effort to implement (high/low). Start by populating it collectively, then discuss disagreements. Disagreements on placement are often the most informative conversations - they reveal differing assumptions about process complexity or data availability.

Short-list 2-3 candidates for further assessment. More than 3 is usually too many to pursue seriously with limited resources.

Gap Management Between Sessions

In multi-session engagements, maintain momentum between workshops:

  • Send a short summary within 24 hours while the discussion is fresh
  • Assign specific people to specific questions that need answering before next session
  • Follow up directly with the stakeholders who need to provide information - do not assume they will volunteer it
  • Use a shared document (Notion, Confluence, or even a simple email thread) to track open items visibly

The biggest risk in multi-session work is that participants experience each session as isolated rather than progressive. Your job is to create continuity.