Skip to content

README.md

Latest commit

 

History

History
151 lines (113 loc) · 6.2 KB

README.md

File metadata and controls

151 lines (113 loc) · 6.2 KB
Raw

Running Shoes Concept Test — Demo

Build Panel. Run Concept Test. Review Findings. 3 running shoe concepts. 15 AI personas. Natural language commands.

The Scenario

You are a product manager at a running shoe brand. Engineering has three concepts ready for consumer testing, but a real panel would take weeks and cost real budget. You want directional signal today.

A: Lightweight Comfort — Minimalist, 7.2 oz, responsive foam, breathable mesh, flexible outsole. Built for speed and agility.

B: Soft Cushion Support — Max cushion, 10.8 oz, dual-density foam, structured heel counter, reinforced toe box. Built for protection and comfort.

C: Versatile Style — Classic silhouette, balanced cushioning, 8 colorways, track-to-street hybrid. Built for lifestyle and everyday wear.

3-Step Workflow

Step 1: Build Panel

/persona generate --count 15 Running shoes

15 diverse personas generated from the topic "Running shoes" using topic-only mode. Each persona has a unique archetype, Big Five personality profile, communication style, and lifestyle details that drive differentiated responses.

# Name Age Occupation Archetype
1 Jake Morrison 28 Sales Account Executive Marathon Purist
2 Priya Raghavan 34 Pediatrician Time-Starved Multitasker
3 Howard Ling 67 Retired Civil Engineer Daily Walker Retiree
4 DeAndre Fuller 22 Graphic Design Student / Barista Sneaker Culture Collector
5 Greg Kowalski 42 HVAC Service Technician Reluctant Beginner
6 Elena Vasquez 30 Park Ranger Trail Adventurer
7 Diane Caldwell 45 PE Teacher & Track Coach Veteran Coach
8 Ravi Mehta 33 Senior Data Scientist Data-Driven Optimizer
9 Jasmine Taylor 20 Student / Cross-Country Runner Budget Racer
10 Karen Yoshida 52 Physical Therapist (DPT) Injury-Conscious Expert
11 Zoe Chen 27 Social Media Manager Athleisure Aesthete
12 Frank Novak 55 Auto Body Shop Owner Category Skeptic
13 Matt Sorensen 38 Wildland Firefighter Endurance Extremist
14 Tamara Mitchell 48 Assistant Principal Weekend Warrior Mom
15 Robert Garcia 63 Retired Firefighter Post-Cardiac Walker

Step 2 (Optional): Explore Motivations

/persona ask What frustrates you most about buying running shoes online?

Run this before the concept test to discover the language and pain points that should shape your concept descriptions. Each persona answers independently.

Step 3: Run Concept Test

/persona concept-test Compare 3 running shoe concepts.

A: Lightweight Comfort — Minimalist, 7.2 oz, responsive foam, breathable mesh. Built for speed.
B: Soft Cushion Support — Max cushion, 10.8 oz, dual-density foam, structured heel counter. Built for protection.
C: Versatile Style — Classic silhouette, balanced cushioning, 8 colorways, hybrid design. Built for lifestyle.

Each persona responds independently in its own claude -p subprocess — zero cross-contamination between personas.

Step 4: Review Findings

Results generated on claude-sonnet-4-6 (June 2026; exact model recorded in run_metadata.json). Close-margin preferences are model-sensitive — see the model sensitivity study.

  • Distribution: B 7 (47%), A 5 (33%), C 3 (20%)
  • Purchase likelihood: mean 3.2/5 — no persona scored 5
  • B leads on protection and evaluable features; A holds the performance-first runners
  • The physical therapist (Karen Yoshida) chose B reluctantly — at 52 with an ACL reconstruction, she admits she needs the protection she recommends to her own injury-prone patients, while still flagging her clinical concern that max-cushion stacks can mask proprioceptive feedback
  • Concept C has the smallest volume but highest enthusiasm: all 3 choosers scored 4/5
  • Specs opacity (missing drop, stack height, energy return %) remains the most common barrier — raised by roughly half the panel
  • Wide-width / sizing availability was flagged by 4 of 15 personas as a deal-breaker

Summary | Results JSON | Charts

Sonnet vs Fable

This demo also ships a full-Fable edition in fable/: a different 15-persona panel generated by claude-fable-5 and answered by claude-fable-5 (the results above use an April 2026 panel answered by claude-sonnet-4-6).

Edition Panel Answering model First choice Mean likelihood
This page April 2026 panel claude-sonnet-4-6 B7 / A5 / C3 3.2
fable/ Fable-generated panel claude-fable-5 B9 / C4 / A2 3.53

Same winner with a wider margin: cushioning wins on a completely rebuilt panel too. See the model sensitivity study.

View pre-generated results (no API calls needed)

Results are already committed. Open concept-test/results/report.md to read findings.

What's in the Box

demo/running-shoes/
├── README.md                          <- You are here
├── personas/                          <- 15-persona panel
│   ├── manifest.json
│   └── 15 persona JSON files
├── concept-test/
│   ├── config.json                    <- 3 shoe concepts (A/B/C)
│   └── results/
│       ├── results.json, report.md, summary.json
│       ├── results.csv, persona_comparison.csv
│       ├── run_metadata.json
│       └── chart_overall.png, chart_purchase_likelihood.png
└── fable/                             <- Full-Fable edition (panel + results
    ├── README.md                         both generated by claude-fable-5)
    ├── personas/
    └── concept-test/

Caveats

  • Virtual panel of 15 AI personas; directional only
  • Not statistically representative; use for hypothesis generation
  • AI-generated responses may exhibit positivity bias
  • Panel segments are each represented by a single persona; treat inter-segment splits as directional, not proportional