| name | Mike Taylor — Synthetic Customer Research OS |
| description | Research-design depth behind synthetic customer panels: the roleplay-then-aggregate mechanism, the grounding ladder (cold-generated, social-grounded via listening gap-fill, transcript-grounded, calibrated), the distribution-vs-individual accuracy split, sycophancy/bias correction, and the research-budget triage that decides which questions synthetic panels answer and which ones still need the real $8-12K focus group. 17 genius patterns, 7 workflows across 3 tiers. Co-author, O'Reilly's Prompt Engineering for Generative AI; co-founder, Ask Rally (synthetic-audience simulator). |
| version | 1.0 |
| format | completion-engine |
| workflows | 7 |
| primary_workflow | workflows/mt-persona-panel-triage.md |
| extracted_from | 1 Marketing Against the Grain video (33:13, 6,916-word transcript, 20 frames, WATCHED) plus secondary corroboration from Vexpower/Ask Rally public material, 2026-07-19 forge |
| tier | system |
Mike Taylor — Synthetic Customer Research OS
A direct question to a chatbot returns "the stock answer, kind of the average of the internet." Mike Taylor's fix is a two-step move: generate a panel of distinct personas, get each to answer independently, then instruct the model to combine their answers "as if these people had collaborated in writing a joint anonymous answer." That mechanism is the surface. This skill is the research-design layer underneath it: how deep to ground a panel before trusting it, when an aggregate verdict is reliable and when an individual prediction is a hallucination wearing a plausible costume, and how to triage the "hundred times more questions than you can afford to answer" between synthetic panels and the real $8-12K focus group.
Where this sits: /buyer-council (skills/geoff-woods-ai-thought-partner/workflows/17-buyer-council.md) is the LIVE operational front door. Its TRIAGE mode already runs this skill's core mechanic (Patterns 1-5) in about 5 minutes with dissent-preserved verdict machinery built in. This skill goes deeper: research design, grounding theory, calibration discipline, and the multi-question research-budget triage buyer-council doesn't attempt. Load /buyer-council for a fast single-artifact gut-check. Load this skill when the call needs grounding-tier discipline, latent-demand discovery, or a real research-budget decision.
Grounding rule: The core mechanism (persona generation, two-step prompt architecture, aggregation phrasing, the $8-12K frame) is VERIFIED, watched directly in the primary source video. The accuracy percentages, the three-tier Grounding Ladder, the self-consistency ceiling, and the sycophancy/vegan-problem findings are LIKELY, corroborated third-party attribution rather than primary-verified. Deploy the method and label the numbers per references/source-quotes.md.
Available Workflows
Tier 1: Foundation (the core mechanism)
| Command | Workflow | Produces | Use When |
|---|
/mt-persona-panel-triage | Persona Panel Triage | Panel verdict: persona list, per-role dissent, joint anonymous answer, directional call | Any decision question needs a fast, cold-generated directional read |
/mt-persona-grounding | Persona Grounding From Transcripts | Transcript-grounded (Tier 2) panel verdict, per-customer responses, real-ask arming — or, when no transcripts exist, a Tier 2.5 social-grounded panel via the built-in gap-fill protocol (harvests a receipted voice-of-customer corpus instead of falling to cold generation) |
