When to Use This Playbook
What this session is — and is not
The goal is not to impress people with tools. The goal is to help participants understand where AI can add value, where it introduces risk, and how to think about responsible use in real STEM and technical workflows.
— Rankine Innovation Lab · Knowledge Hub
This playbook is for small departments, programme teams, labs, schools, innovation groups, or partner organisations that need a practical and disciplined way to introduce generative AI into existing workflows. It is useful for mixed audiences — educators, researchers, students, technicians — provided the session is built around shared workflows rather than tool jargon.
A successful session is not measured by how impressed participants are. It is measured by whether they leave more precise. A good session produces four specific outcomes: clarity about what AI is and is not good for in this context; at least two workflow points where AI might genuinely help; at least two risks or boundaries requiring governance; and enough host-team feedback to decide what should happen next.
Preparation
What you need before the session begins
Before running the session, the organising team must assemble three kinds of inputs and set one essential boundary. Sessions that skip this preparation stage typically become generic technology talks rather than capability-building exercises.
👥
Defined Audience
Who are the participants? Educators, researchers, students, technical staff, or a mix? The answer should shape every example, the level of depth, and the language used. A mixed audience can work — but only if the session is anchored in shared workflows.
📋
Pre-Session Note
A short document — one page or less — sent to participants before arrival. It explains the aim of the session and sets expectations. Without this, the first 15 minutes are consumed by orientation that should have happened already.
💼
One or Two Workflow Examples
Not generic chatbot demonstrations. Domain-relevant, bounded examples that are easy to discuss critically — comparing a raw model answer with a retrieval-grounded answer, or turning a dense paper into a decision brief outline.
⚠
Boundary Statement
A clear declaration of what the session does not cover: formal institutional policy approval, independent high-stakes use without review, or anything requiring regulated professional judgment. State this upfront, not in the final minutes.
Session Blueprint
The 85-minute structure that works
A strong small-format session runs in 60 to 90 minutes. This structure keeps the session grounded in action and interpretation rather than tool theatre. Each block has a specific purpose — the order matters because trust and precision build sequentially.
Framing
Establish why the topic matters now and why the session is being held — specifically for this group, in this context. Do not begin with model capabilities. Begin with the decisions and workflows participants already care about.
AI Basics — Workflow Logic, Not Model Trivia
Focus on how AI fits into a workflow — not on model architecture, benchmark scores, or version comparisons. The key concepts are: what AI generates from, what it cannot know, and why retrieval and grounding change reliability.
Applied Examples — One Strong Case, One Risky Case
Show one use case where AI genuinely improves a specific task, and one where it fails or misleads. Both must be domain-relevant. The pairing matters: participants who see only successes develop inflated confidence; those who see only risks develop inflated fear.
Hands-On Reflection
Run the workflow-map exercise: participants identify one task they do regularly, break it into steps, and mark where AI could support and where it should not be trusted. This moves the conversation from abstract interest to practical discernment.
Risk and Boundary Discussion
Cover hallucination, source quality, confidentiality, overreliance, and the difference between drafting support and expert judgment. The facilitator should resist two extremes — hype leads to inflated assumptions; fear leads to dismissal before real understanding.
Next Steps — Capture and Commit
Gather the questions, concerns, suggested use cases, and barriers that emerged during the session. Without a named next step — a follow-up pilot, a guidance note, a readiness assessment — the session generates interest but not capability.
Exercise Design
Two exercises that consistently work
Examples make or break a capability session. Generic prompts and chatbot demonstrations leave participants impressed but unequipped. The best exercises are domain-relevant, bounded, and designed to surface critical thinking rather than generate admiration.
Exercise One
The Workflow Map
Participants identify one regular task, break it into steps, and mark where AI could support — and where it should not be trusted without review.
Pick one task you do regularly this month
Break it into 4–6 distinct steps
Mark each: AI-assist, human-primary, or hybrid
Identify which steps carry highest risk if AI is wrong
Exercise Two
Side-by-Side Comparison
Show a generic ungrounded answer next to one that cites approved documents. Ask what changed in trust, usefulness, and reviewability.
Pose a domain-relevant question to both systems
Display answers side by side without context first
Reveal source structures of each answer
Ask: which would you act on, and why?
Measuring Success
What a good session produces
If none of these four outcomes are visible at the end of the session, the event has likely become a generic technology talk rather than a capability-building exercise. Use these as evaluation criteria — not aspirations.
01
Contextual Clarity
Participants can articulate what AI is and is not good for in their specific context — not in general. Generic statements about AI capability do not count.
02
Identified Workflow Fits
Each participant can name at least two specific workflow points where AI might genuinely reduce burden, improve speed, or support quality — with a concrete reason why.
03
Recognised Risk Points
Participants can identify at least two risks or boundaries that require policy, review, or governance in their context — and can explain why those specific risks matter.
04
A Named Next Step
The host team collects enough feedback to decide what happens next — another session, a pilot, a policy note, or a deliberate pause. Without this, capability does not grow.
Critical Awareness
Common failure points — and what to do after
Three failure modes account for most poor capability sessions. Knowing them in advance is the most practical form of facilitation preparation. The post-session sort then converts what emerged into a structured next-action foundation.
⛔ Common Failure Points
Trying to cover every AI concept, governance issue, and application area in one session — scope creep collapses focus
Making the session too product-specific — participants need transferable principles more than they need a temporary tool tour
Failing to define a post-session action — without a next step, sessions generate interest but not institutional capability
Assuming all participants begin from the same level of confidence or concern — some are eager, others cautious; the structure must hold space for both
✓ Facilitation Anchors
Return repeatedly to: what job are we trying to improve? What standard must we protect?
Stay curious but disciplined — resist hype and resist dismissal with equal energy
Ask: what source should ground this answer? What still requires human expertise in this context?
Name what the session does not cover before someone discovers it by accident
The most important post-session step is synthesis. Within 48 hours, sort everything that emerged into four buckets. This structure prevents good feedback from becoming an unorganised pile of post-it notes.
Immediate Use Cases
Ready to pilot or trial now
Low governance complexity
Clear owner and scope
Policy & Governance Questions
Need institutional clarity first
Data privacy or access concerns
Accountability gaps identified
Training Needs
Skill gaps surfaced during session
Prompting or evaluation literacy
Domain-specific knowledge gaps
Do Not Proceed Yet
High-stakes with weak governance
Source base not yet stable
Accountability unclear
References & Source Base
- Rankine Innovation Lab Knowledge Hub research synthesis: AI-for-STEM capability playbook and the broader research-to-action positioning.
- Founder-connected work on generative AI and retrieval-grounded technical support across construction and water-systems domains.
- Rankine positioning of AI as a practical capability layer rather than a novelty overlay — informing session design philosophy.