# Discovery and brainstorming workshop agenda A workshop is a decision-making instrument, not a room booking. Design the conversation before you invite people: one decision, independent thinking first, criteria agreed before options are seen, and a closed loop at the end. Source: lesson 0040 · Facilitate Discovery and Brainstorming. --- ## Blank ### Prepare (before the invitation exists) | Field | What goes here | | --- | --- | | Decision question | One sentence, visible on the wall the whole time. If you cannot write it, cancel and go find it. | | Decider | The named person who closes the choice after advice. Not "the group". | | Participants | Invite for knowledge, not representation: user/domain knowledge, product, design, engineering, and whoever carries the operational consequence. | | Pre-read | Frame, evidence, constraints, glossary — sent at least a day ahead, with facts, interpretations, and open questions marked separately. | | Criteria | Outcome impact, confidence, effort, reversibility, guardrail risk. Agree these *before* anyone sees a favored option. | | Closing artifact | Name it now: an assumption map with owners, an option table, a decision plus dissent. | ### Run (75 minutes) | Minutes | Mode | Prompt | Output | | --- | --- | --- | --- | | 0–10 | frame | Read outcome, evidence, constraints. Clarifying questions only. | Shared decision frame | | 10–20 | silent diverge | Write facts, assumptions, and questions alone — one per note. | Independent raw material | | 20–35 | share and cluster | Round-robin. Clarify without defending. Name themes. | Named opportunity themes | | 35–50 | silent sketch | At least three mechanisms, including a manual or no-build option. | Three real options | | 50–65 | converge | Score against the agreed criteria. Each person states the strongest case *against* their own preference. | Trade-offs and recorded dissent | | 65–75 | close | Direction, dissent, unknowns, tests, owners, decision date. | Discovery actions | ### Close (the actual output) - **Direction:** `[what the group is now pursuing]` - **Dissent:** `[who disagrees, with what, and what evidence would settle it]` - **Open unknowns:** `[what nobody in the room could answer]` - **Next evidence:** `[test — owner — date]` × 3 at most - **Decision date:** `[when the decider closes it]` ### Facilitator failure modes | Failure | Signal | Move | | --- | --- | --- | | Anchoring | Every option resembles the brief | Restate the outcome; require three different mechanisms | | Authority bias | The room waits for the most senior person | Silent work first; the leader shares last | | False consensus | No disagreement is recorded | Ask "what would make this fail?" and write the answer down | | Workshop theatre | Board full, owner column empty | End with action, owner, evidence, date | --- ## Completed — returns option workshop, 26 August ### Prepare | Field | Entry | | --- | --- | | Decision question | What is the smallest intervention that can cut eligible resolution time from 72 hours to under 4 during the pilot? | | Decider | Priya for direction and cohort; Dana for the technical option. | | Participants | Priya (product), Marco (design), Dana (tech lead), two engineers, Nils (support operations, who works the returns queue), Anya (returns policy owner), and a warehouse supervisor for the first 35 minutes. Seven people. | | Pre-read | Initiative frame v0.1, the 72-hour and 41% baselines with the queries that produced them, the carrier contract summary, and the returns glossary — specifically, what "approved" means to support, to operations, and to finance. | | Criteria | Outcome impact · confidence given today's evidence · effort to first evidence · reversibility · risk to the fraud and manual-queue guardrails. Agreed and posted before anyone sketched. | | Closing artifact | An assumption map with owners and dates, one direction, and recorded dissent. | ### What came out of each block - **Frame (10 min).** The brief said "self-serve returns portal". The wall said "under 4 hours, 25% fewer contacts". Two people had not seen the 41% contact figure before. - **Silent diverge (10 min).** The most useful fact in the room came from Nils's notes: most eligible returns are unopened domestic items that agents approve without real deliberation, and most of the 72 hours is queue time, not decision time. - **Cluster (15 min).** Four themes: policy is ambiguous, customers cannot see status, agents re-key data between three systems, and evidence (photos) is requested inconsistently. - **Sketch (15 min).** Five mechanisms, four of them genuinely different: (1) automated eligibility decision, (2) proactive status notification with no automation at all, (3) agent tooling that removes re-keying, (4) a full self-serve portal, (5) do nothing and add two agents for the season. - **Converge (15 min).** Automated eligibility scored highest on outcome impact and lowest on confidence. Proactive status scored second on impact, highest on confidence, and was reversible in a week. Adding agents was cheapest to try and moved the median very little. - **Close (10 min).** Below. ### Close - **Direction:** test automated eligibility *and* proactive status before committing to a full portal. Do not build the portal shell first. - **Dissent:** Nils believes guest orders will dominate demand and that a signed-in-only pilot will measure the wrong population. Recorded. The evidence that would settle it is the guest share of return-eligible orders. - **Open unknowns:** the codifiable-policy rate; carrier label latency; whether customers understand an automated refusal; whether Risk will accept an automatic decision at all. - **Next evidence:** Priya samples 200 closed agent decisions · Marco runs a five-participant comprehension test · Dana runs a four-hour carrier latency spike · Priya sizes the guest share. All four owned, all four dated inside two weeks. - **Decision date:** the technical direction in two weeks, once the spike lands. **What the workshop did not produce.** An architecture. Choosing a mechanism before the codifiable-policy rate was known would have been an opinion with a diagram attached — and the sample later came back at 63%, not the 90% everyone in the room had assumed. **Note on the dissent.** Nils was right. Guest orders turned out to be 29% of demand. The pilot cohort did not change, but the outcome statement did: it now names the signed-in cohort explicitly, so nobody quotes the headline number as if it covered everyone. **Traces to:** every "next evidence" line is a row in `assumption-map.md`; the dissent reappears in `weekly-review-and-closeout-checklist.md` when the guest share came back at 29%.