Convert Messy Client-Workshop Notes into a Findings-and-Recommendations Skeleton with an Evidence Map

The task

You ran a discovery or design workshop with a client — vendor selection, an SAP migration kickoff, a target operating model session. Now you have eight pages of scrappy notes, a whiteboard photo someone transcribed badly, and a Monday partner review. You need a findings-and-recommendations skeleton that maps every claim back to who said it.

Before AI

The grunt version: read through notes twice, color-code by theme in Word, build a synthesis grid in Excel, then draft a pyramid in PowerPoint. Two to three hours, and the evidence map usually doesn't get built — which is exactly why the partner asks "who said this?" and you can't answer.

The workflow

The structure here borrows from Minto's pyramid principle — group findings under MECE themes, then surface a recommendation per theme. The new bit is forcing an evidence map so every line is traceable to a quote.

Step 1 — Normalize the raw notes into observation units. Paste the workshop notes after this prompt. The model splits them into discrete, attributable observations.

You are a senior consultant cleaning up workshop notes for synthesis. Below are raw notes from a client workshop. Your job is to convert them into a clean list of OBSERVATION UNITS.

Rules:
- One observation per row. Atomic — no compound statements.
- Preserve the speaker/role attribution if present. If unclear, mark as "Unattributed".
- Keep the original phrasing where possible; lightly clean grammar only.
- Tag each observation with a short topic label (2-3 words) you infer from context.
- Output a markdown table with columns: ID | Speaker/Role | Topic | Observation (verbatim or near-verbatim) | Type (Fact / Opinion / Concern / Request / Decision).
- Number IDs sequentially as O1, O2, O3...
- Do NOT synthesize, summarize, or add interpretation yet. Just structure.

Raw notes follow:

NorthStar Logistics — Vendor Strategy Workshop, Tue
Attendees: Priya Chen (CIO), Marco Vellini (Head of Ops), Dana Okafor (Procurement), Sam Liu (IT Architecture), Rita Bose (Finance BP)

— Priya kicked off, frustrated. "We have 14 TMS vendors across regions, nobody can tell me total spend." Wants single TMS by FY28.
— Marco pushed back: APAC ops use Bringg for last-mile, "ripping it out would tank our SLAs in Jakarta and Manila." 3 PL partners integrated there.
— Dana: contracts staggered, 6 expire in next 18mo, 4 auto-renew Jan. "Window is now or 2027."
— Sam flagged tech debt — Oracle TMS on-prem, end of support Dec 2026. "Not a choice, we have to move."
— Rita: finance can't reconcile freight accruals across the 14 systems. Closing takes 9 business days vs. target 5. Audit flagged it last yr.
— Group debated build vs buy. Sam: "build is off the table, we don't have the bench." Priya agreed.
— Marco again — worried about change fatigue, "we just did WMS last year, the warehouse teams will revolt."
— Dana suggested shortlist of 3: Blue Yonder, Manhattan Active, e2open. Priya wants Oracle TM Cloud on the list too "for the renewal leverage."
— Sam: integration with SAP S/4 (going live EMEA Q3) is the gating item. Two vendors have certified connectors.
— Action: Dana to pull spend data by Friday. Sam to draft integration criteria. Reconvene in 2 weeks.
— Off the cuff from Rita at the end: "honestly if we don't fix freight visibility the CFO will mandate something top-down by Q4."

Step 2 — Cluster into MECE themes and draft findings. This runs against the table from Step 1.

Using the observation table above, do the following:

1. Group observations into 3-6 MECE themes (mutually exclusive, collectively exhaustive). Name each theme as a noun phrase a partner would recognize (e.g., "Vendor landscape fragmentation", "Integration constraints").

2. For each theme, write ONE finding statement. A finding is a declarative claim the evidence supports — not a question, not a recommendation. Keep it under 25 words.

3. For each finding, draft ONE recommendation. Action-oriented, starts with a verb, names an owner role if inferable, includes a rough timeframe.

4. Rate confidence per finding: High / Medium / Low — based on how many independent observations support it and whether they're facts vs. opinions.

Output format:

### Theme N: [name]
- **Finding:** ...
- **Recommendation:** ...
- **Confidence:** High/Medium/Low — [one-line justification]
- **Supporting observations:** O3, O7, O12

Flag any theme where you had to stretch to fit observations — don't force MECE if the evidence is thin.

Step 3 — Build the evidence map and surface gaps. The audit layer that catches the "who said this?" problem before the partner does.

Now produce two artifacts:

A) EVIDENCE MAP — a markdown table with columns: Finding # | Observation ID | Speaker/Role | Verbatim snippet (≤15 words) | Type. One row per observation-to-finding link. If an observation maps to multiple findings, repeat it. If any observation from Step 1 went unused, list it in a separate "Unmapped observations" section underneath.

B) GAP & RISK LOG — for each finding, list:
- What we DON'T know yet that would change the recommendation
- Which stakeholder voice was missing or under-represented in the room
- Any observation that contradicts the finding (dissent)

Be specific. "Need more data" is not acceptable — name the data, the source, and why it matters.

Step 4 — Format for the partner review. Optional but worth it if you're presenting Monday.

Produce a one-page executive summary in Minto pyramid order:
1. Governing thought (one sentence — the "so what" across all themes)
2. Three to five key findings as parallel bullets
3. Recommended next actions, sequenced, with owners and timing

Keep it to roughly 250 words. Plain English, no consulting clichés ("synergies", "leverage", "north star"). Write it so a CFO who wasn't in the room understands the call to action.

Gotchas

• Garbage in, garbage out. If the notes don't capture who said what, the evidence map collapses to "Unattributed" everywhere and loses its value. Brief your note-taker to tag speakers.
• MECE is a goal, not a guarantee. The model will sometimes force-fit an observation into a theme to look tidy. Step 2's "flag stretches" instruction helps, but eyeball the groupings yourself.
• Confidence ratings are directional. The model can't actually weigh a CIO's opinion against an architect's fact. Treat High/Medium/Low as a prompt for your own judgment, not a verdict.
• Dissent gets flattened. Workshop dynamics — who interrupted whom, what got dropped — don't survive transcription. If a key disagreement matters, add it back manually before the partner review.
• Don't paste anything you wouldn't email externally. Strip client names, dollar figures, and PII before running this in a non-sanctioned tool. The structured-thinking habit still applies; the data hygiene is on you.

Time saved

Roughly 2-3 hours per workshop. The synthesis grid and evidence map are where the hours usually go — Step 1 and Step 3 collapse that to minutes. You still own the judgment calls in Step 2 and the partner-facing polish in Step 4, which is the right place for your time anyway.

What we got when we ran it

We executed every prompt above, in order, against the sample input (model: claude-sonnet-4-6). Outputs below are real and unedited apart from length truncation.

Step 1 output

| ID | Speaker/Role | Topic | Observation | Type | |----|-------------|-------|-------------|------| | O1 | Priya Chen (CIO) | Vendor Fragmentation | We have 14 TMS vendors across regions and nobody can tell me total spend. | Fact | | O2 | Priya Chen (CIO) | Vendor Consolidation | Wants a single TMS by FY28. | Request | | O3 | Marco Vellini (Head of Ops) | APAC Operations | APAC ops use Bringg for last-mile. | Fact | | O4 | Marco Vellini (Head of Ops) | APAC Operations | Ripping out Bringg would tank our SLAs in Jakarta and Manila. | Concern | | O5 | Marco Vellini (Head of Ops) | Partner Dependencies | Three 3PL partners are integrated with Bringg in APAC. | Fact | | O6 | Dana Okafor (Procurement) | Contract Timeline | Six vendor contracts expire in the next 18 months. | Fact | | O7 | Dana Okafor (Procurement) | Contract Timeline | Four contracts auto-renew in January. | Fact | | O8 | Dana Okafor (Procurement) | Contract Timeline | The decision window is now or 2027. | Opinion | | O9 | Sam Liu (IT Architecture) | Technical Debt | Oracle TMS on-prem reaches end of support in December 2026. | Fact | | O10 | Sam Liu (IT Architecture) | Technical Debt | Migration off Oracle TMS on-prem is not optional — we have to move. | Opinion | | O11 | Rita Bose (Finance BP) | Financial Reporting

…output truncated at 250 words.

Step 2 output

Theme 1: Vendor Landscape Fragmentation - Finding: NorthStar operates 14 TMS vendors with no consolidated spend visibility, creating financial and operational opacity. - Recommendation: Assign Procurement (Dana) to deliver a complete vendor spend inventory by the existing Friday deadline, as the baseline for all consolidation decisions. - Confidence: High — supported by multiple independent facts from Finance, Procurement, and IT, not just opinion. - Supporting observations: O1, O2, O11, O12, O13 --- ### Theme 2: Consolidation Window and Contract Leverage - Finding: Six contracts expiring within 18 months and one end-of-support deadline create a time-bounded, non-recurring opportunity to consolidate. - Recommendation: Procurement (Dana) and IT Architecture (Sam) jointly map contract expiry dates against the SAP S/4 EMEA go-live to define a sequenced consolidation timeline before the next workshop. - Confidence: High — contract expiry dates and Oracle EOS are stated facts; the "now or 2027" framing is opinion but logically consistent with the facts. - Supporting observations: O6, O7, O8, O9, O10 --- ### Theme 3: Integration Constraints and Technical Gating - Finding: SAP S/4 EMEA go-live and Oracle TMS end-of-support are hard technical constraints that narrow the viable vendor shortlist before evaluation begins. - Recommendation: IT Architecture (Sam) publish integration criteria — including SAP S/4 certified connector requirement — within two weeks, prior to any formal vendor scoring. - Confidence: High — EOS date and S/4 go-live are independently stated facts; certified connector count is a specific, checkable claim. - Supporting observations: O9, O10, O20, O21, O23 ---

…output truncated at 250 words.

Step 3 output

A) EVIDENCE MAP | Finding # | Observation ID | Speaker/Role | Verbatim snippet (≤15 words) | Type | |-----------|---------------|--------------|------------------------------|------| | F1 | O1 | Priya Chen (CIO) | "We have 14 TMS vendors across regions, nobody can tell me total spend." | Fact | | F1 | O2 | Priya Chen (CIO) | Wants a single TMS by FY28. | Request | | F1 | O11 | Rita Bose (Finance BP) | Finance cannot reconcile freight accruals across the 14 systems. | Fact | | F1 | O12 | Rita Bose (Finance BP) | Month-end closing currently takes 9 business days against a target of 5. | Fact | | F1 | O13 | Rita Bose (Finance BP) | The freight accrual reconciliation issue was flagged by audit last year. | Fact | | F2 | O6 | Dana Okafor (Procurement) | Six vendor contracts expire in the next 18 months. | Fact | | F2 | O7 | Dana Okafor (Procurement) | Four contracts auto-renew in January. | Fact | | F2 | O8 | Dana Okafor (Procurement) | "The decision window is now or 2027." | Opinion | | F2 | O9 | Sam Liu (IT Architecture) | Oracle TMS on-prem reaches end of support in December 2026. | Fact | | F2 | O10 | Sam Liu (IT Architecture) | Migration off Oracle TMS on-prem is not optional — we have to move. | Opinion | | F3 | O9 | Sam Liu (IT Architecture)

…output truncated at 250 words.

Step 4 output

NorthStar Logistics — TMS Vendor Strategy: Executive Summary **NorthStar must commit to a single TMS platform within the next 90 days or lose the contract and technical windows that make consolidation achievable before 2027.** --- **What the evidence shows:** - The current 14-vendor TMS landscape is costing Finance directly: month-end close runs 9 days against a 5-day target, freight accruals cannot be reconciled across systems, and the issue has already been flagged by audit. - Two hard deadlines are converging: the on-premise Oracle TMS loses vendor support in December 2026, and 6 of 14 contracts expire within 18 months — after which 4 auto-renew and the consolidation window closes until 2027. - SAP S/4 goes live in EMEA in Q3, and only two of the four candidate vendors have certified connectors for it. Integration compatibility must drive shortlisting, not commercial preference. - APAC operations carry real migration risk: three 3PL partners are integrated with the current last-mile platform, and the warehouse organisation completed a separate system change last year. Sequencing matters. --- **What needs to happen next:** 1. **Procurement (Dana)** — deliver full vendor spend inventory by this Friday. No consolidation decision is defensible without it. 2. **IT Architecture (Sam)** — publish integration criteria, including SAP S/4 connector requirements, within two weeks. This gates the shortlist. 3. **CIO (Priya)** — brief the CFO on the programme timeline and early Finance milestones within 30 days, before external pressure forces a less considered decision. 4. **Head of Ops (Marco)** — complete

…output truncated at 250 words.