Data Synthesis

Data Synthesis -- Research Intelligence

Skill pack for transforming raw research data into structured insights. Covers affinity mapping, thematic analysis, quantitative aggregation, triangulation, insight scoring, gap analysis, and integration with the Memoire research pipeline.

Core Loop

Copy code
COLLECT RAW DATA --> ORGANIZE --> CODE --> THEME --> TRIANGULATE --> SCORE --> OUTPUT INSIGHTS

Every synthesis pass must produce actionable insights, not summaries. An insight is a non-obvious finding that changes a design decision.

Affinity Mapping Methodology

What It Is

A bottom-up clustering technique. Individual observations (sticky notes, quotes, data points) are grouped by natural similarity until themes emerge.

Process

Rules

• One observation per data point -- never combine two findings into one note
• Group by meaning, not by source (do not cluster "all interview 3 data" together)
• If a cluster has more than 15 items, split it
• If a cluster has fewer than 3 items, consider merging or flagging as an outlier
• Relabel clusters after regrouping -- initial labels often drift from final content

In Memoire

bashCopy code
memi research from-stickies       # Pull sticky notes from Figma canvas
memi research synthesize           # Run affinity mapping + thematic analysis
memi research report               # Generate structured insight report

Thematic Analysis

Three Coding Phases

Phase 1: Open Coding

Read through all data and assign descriptive codes to each meaningful segment.

Rules:

• Code everything relevant, skip nothing on first pass
• Use gerunds or descriptive phrases (not single words)
• One segment can have multiple codes
• Target 30-80 unique codes from a typical study

Phase 2: Axial Coding

Group open codes into categories. Identify relationships between categories.

Relationships to identify:

• Causal: A leads to B ("navigation failure causes social workarounds")
• Contextual: A happens during B ("mental model mismatch occurs during onboarding")
• Strategic: A is a response to B ("social workarounds compensate for navigation failure")

Phase 3: Selective Coding

Identify the core theme that connects all categories. This becomes the central narrative.

Copy code
Core Theme: "Users build parallel information architectures through social channels
             because the system's structure does not match their mental models."

Code Quality Checklist

• Every data segment is coded (no orphan data)
• Codes are applied consistently across all sources
• Categories have clear boundaries (no code belongs to 3+ categories)
• Core theme is supported by data from multiple participants/sources

Quantitative Aggregation

Standard Metrics

NPS Calculation

Copy code
Responses on 0-10 scale:
  Detractors: 0-6
  Passives:   7-8
  Promoters:  9-10

NPS = (Promoter Count / Total) * 100 - (Detractor Count / Total) * 100
Range: -100 to +100

SUS Score Interpretation

Statistical Significance

• For A/B comparisons: use chi-squared test (categorical) or t-test (continuous)
• Minimum sample size for usability: 5 per segment (qualitative), 30+ per variant (quantitative)
• Report confidence intervals, not just point estimates
• Flag any metric based on fewer than 5 data points as unreliable

Triangulation Across Sources

Triangulation Matrix

Cross-reference findings across at least 3 data sources to validate strength.

Strength Classification

Insight Quality Scoring

Rate every insight on four dimensions before including it in a deliverable.

Threshold: Only insights scoring 12+ (out of 20) go into the final report. Insights scoring 8-11 go into an appendix. Below 8, discard or flag for follow-up research.

Insight Statement Format

Copy code
[WHO] experiences [WHAT PROBLEM/BEHAVIOR]
because [WHY/ROOT CAUSE],
which means [IMPLICATION FOR DESIGN].

Example:

Copy code
New users abandon the onboarding wizard at step 3
because the required fields feel invasive for a first interaction,
which means we should defer profile completion to post-activation.

Gap Analysis Frameworks

Coverage Gap Matrix

Map research questions against data sources to identify blind spots.

Types of Gaps

Integration with Memoire Research Pipeline

Pipeline Stages

Copy code
RAW DATA (Excel, CSV, stickies)
  |
  v
memi research from-file <path>          # Ingest raw data
  |
  v
memi research from-stickies             # Pull Figma sticky notes
  |
  v
memi research synthesize                # Affinity map + thematic analysis
  |                                      # Outputs coded themes + scored insights
  v
memi research report                    # Structured insight report
  |                                      # Markdown + Figma sticky output
  v
memi spec component <name>              # Insights feed component specs
memi spec page <name>                   # Insights feed page specs
memi compose "dashboard from research"  # Agent builds dashboard from insights

Data Formats

Output Artifacts

Connecting Insights to Specs

When generating component or page specs, reference insight IDs:

jsonCopy code
{
  "name": "OnboardingWizard",
  "level": "organism",
  "researchInsights": ["INS-007", "INS-012"],
  "designRationale": "Deferred profile completion based on day-1 drop-off finding (INS-007)"
}

This creates traceability from research through spec to generated code.