Signal Taxonomy Workshop

Purpose

This recipe executes a structured 2-day workshop that transforms raw signal source data into a validated classification taxonomy. Day 1 focuses on knowledge extraction from domain experts. Day 2 focuses on quantification — scoring signal strength, setting false positive thresholds, designing compound signals, and validating against live data. [src1, src3]

Prerequisites

• Signal Source Audit completed — scored inventory of available data sources
• Domain expert confirmed for 2 full days
• Sample raw data from top 5 sources in CSV/JSON
• Workshop space with whiteboard, projector, internet access
• Facilitator assigned with signal methodology knowledge

Constraints

• At least 1 domain expert must participate — taxonomy without domain expertise misses industry nuances. [src2]
• Signal source audit must be completed before the workshop.
• False positive threshold must be explicitly set. Default: 30% max. Regulated industries: 15-20%. [src4]
• Taxonomy validated against minimum 50 real-world examples before sign-off.
• Compound signals must be tested independently before combining.

Tool Selection Decision

Which workshop format?
├── In-person (recommended)
│   └── PATH A: Full 2-day on-site workshop
├── Remote synchronous
│   └── PATH B: 2-day virtual via Miro/FigJam
├── Hybrid (expert remote)
│   └── PATH C: Facilitator on-site, expert via video
└── Async (expert unavailable for 2 days)
    └── PATH D: 4x half-day sessions over 2 weeks

Execution Flow

Step 1: Pre-Workshop Preparation (Day 0)

Duration: 1 day · Tool: Document preparation + data staging

Prepare materials: print audit report, load sample data into shared spreadsheet, create taxonomy template, prepare 20+ company examples for validation, draft and circulate agenda. [src3]

Verify: All materials prepared, sample data accessible, participants confirmed. · If failed: Workshop can proceed without sample data but Day 2 validation will be weaker.

Step 2: Day 1 Morning — Domain Expert Interviews (3 hours)

Duration: 3 hours · Tool: Structured interview + whiteboard

Semi-structured interviews: trigger events, behavioral changes 3-6 months before purchase, relevant public data, false signals. Map onto preliminary signal hierarchy grouped by category. [src1, src5]

Verify: Minimum 20 signal types across 3+ categories. · If failed: Prompt expert with audit report examples.

Step 3: Day 1 Afternoon — Data Source Deep-Dive (3 hours)

Duration: 3 hours · Tool: Audit report walkthrough

Walk each data source with domain expert: extractable signals, individual strength, source-to-hierarchy mapping, gaps, compound opportunities, trigger event sequences.

Verify: Hierarchy mapped to sources. Compound opportunities documented. · If failed: Demo sample data to expert for interpretation.

Step 4: Day 2 Morning — Signal Strength Scoring (3 hours)

Duration: 3 hours · Tool: Scoring framework + spreadsheet

Score each signal: individual strength (1-10), reliability (1-5), timeliness (1-5), compound multiplier (1.0-2.0). Set false positive threshold explicitly. [src2, src4]

Verify: All signals scored, threshold documented. · If failed: Use median scores, document dissent for calibration.

Step 5: Day 2 Afternoon — Live Validation (3 hours)

Duration: 3 hours · Tool: Sample data + taxonomy + scoring model

Validate against 50+ company records. Apply scoring, compare to expert judgment, calculate false positive/negative rates, identify edge cases. [src4]

Verify: False positive rate within threshold. Edge cases documented. · If failed: Tighten weights, add qualifying signals, revalidate.

Step 6: Post-Workshop Documentation (Days 3-5)

Duration: 2-3 days · Tool: Document + JSON schema generation

Produce deliverables: taxonomy document, JSON schema, calibration dataset, false positive analysis, implementation notes.

Verify: Reviewed by domain expert and engagement lead. · If failed: Flag as “provisional,” schedule review call.

Output Schema

{
  "output_type": "signal_taxonomy",
  "format": "JSON schema + document",
  "sections": [
    {"name": "signal_hierarchy", "type": "object", "description": "Category > type > indicator tree with weights"},
    {"name": "scoring_model", "type": "object", "description": "Strength, reliability, timeliness, compound multiplier"},
    {"name": "false_positive_threshold", "type": "number", "description": "Maximum acceptable FP rate"},
    {"name": "compound_signals", "type": "array", "description": "Multi-source combinations with multipliers"},
    {"name": "validation_dataset", "type": "array", "description": "50+ scored examples with outcomes"},
    {"name": "edge_cases", "type": "array", "description": "Exceptions and special handling rules"}
  ]
}

Quality Benchmarks

If below minimum: Extend workshop by half day or schedule follow-up session.

Error Handling

Cost Breakdown

Anti-Patterns

Wrong: Designing taxonomy without domain expertise

Building classification from data patterns alone. Result: classifier flags wrong companies, sales team loses trust. [src2]

Correct: Domain expert drives taxonomy design

Minimum 1 domain expert for full 2 days. Industry knowledge catches false patterns data alone misses.

Wrong: Skipping live data validation

Finalizing taxonomy without testing against real data. Result: elegant on paper, fails on first dataset. [src4]

Correct: Validate against 50+ real examples

Day 2 afternoon dedicated to applying taxonomy to real data. Iterate before leaving the workshop.

Wrong: Setting implicit false positive thresholds

Not discussing acceptable error rates. Result: too many irrelevant dossiers overwhelm sales team. [src4]

Correct: Explicitly set and document the threshold

Agree on maximum false positive rate during Day 2. Write it into the taxonomy as a hard constraint.

When This Matters

Use when designing an intent signal classification system for a specific vertical. This is Phase 2 of the Signal Stack engagement — it transforms raw signal data into a structured taxonomy for automated classification.