[nevrai]
· 9 min read

26 Data Points × Knowledge Graph = Autopilot for Product Research

Product research always fights the same battle: incomplete data. You need to understand the customer, the job the product does, the obstacles, the competitive landscape — but systematizing all of this requires interviews, spreadsheets, tagging, and summaries. It’s a lot of manual work.

AICPO handles this through two mechanisms: DP Tracker (26 data points from Product DNA methodology) and Knowledge Graph. Useful individually — together they create a system that identifies missing information on its own.

What Are the 26 Data Points?

Product DNA establishes that understanding a customer means understanding not just the job they’re hiring the product to do, but also switching forces, adoption barriers, solution economics, and trend context.

The 26 data points form a completeness checklist. Each represents a critical category of customer knowledge:

Segment and Context:

  • Customer identity (demographics, role, context)
  • When the need emerges (triggers)
  • Preceding situation (circumstances)

Work and Motivation:

  • Functional work (tasks to complete)
  • Emotional work (desired feelings)
  • Social work (how they want to appear)
  • Undesired outcomes (what they’re avoiding)

Alternatives and Barriers:

  • Current solutions in use
  • Satisfaction gaps with those solutions
  • Switching obstacles
  • Success criteria

Economics:

  • Payment willingness
  • Budget and purchase process
  • Decision influencers — makers, saboteurs, champions

Competitive Environment:

  • Competing solutions
  • Competitor weaknesses
  • Unique assets

Trends:

  • External forces reshaping customer work
  • Need evolution vectors

Adoption and Retention:

  • Previous unsuccessful attempts
  • Transition fears and resistance
  • Expected ROI and time-to-value
  • Previous solution abandonment reasons
  • Success vision
  • Cognitive load at first use
  • Catalyst event triggering the solution search

Each point is a boolean: true (data exists) or false (gap). AICPO tracks coverage in real-time.

How the Graph Relates to Data Points

DP Tracker starts with keyword matching — checking extracted facts against dictionaries for each data point. Found keywords close the item.

That’s the first layer. Knowledge Graph adds the second.

When “customer is a procurement manager in B2B” gets extracted, it enters the graph as:

Node(segment): "procurement manager"
  properties: { industry: "B2B", role: "procurement" }
  source_message: #12
  confidence: 87

When “difficult to coordinate with finance director” appears later:

Node(constraint): "coordination with CFO"
  properties: { type: "organizational" }
  source_message: #24

Edge: segment → constraint (type: faces, confidence: 82)

DP Tracker now checks not just for “decision-maker” text but for Node(segment) with organizational constraint edges. Keyword matching becomes semantic coverage.

Autopilot: What Happens Without Your Input

After each user message, a pipeline runs:

  1. FactExtractorService — LLM extracts structured facts
  2. DpCoverageService — maps facts to 26 data points
  3. Kg2::EpisodeService — creates an episode from facts
  4. Kg2::BuilderJob — EntityExtractor + EntityResolver + RelationDiscoverer
  5. BriefUpdaterJob — regenerates project summary

This entire pipeline runs asynchronously while the user keeps talking.

Result: after 20-30 minutes of interviews, you see a heatmap of 26 cells. Green = closed, red = gap. Next to each gap is a suggested question.

That’s autopilot. Not “what should I do next?” — “here’s what’s still unknown, here’s the question to ask.”

Why 26, Not 10 or 50?

10 is not enough. Classic frameworks like “problem/solution/market” miss critical zones: switching barriers, social work, trend vectors, abandonment reasons.

50 paralyzes. Researchers drown in details, lose focus, closure never happens.

26 is the minimally sufficient set for high-confidence product decisions. The first 19 cover the classics (segment through trends). Seven additional items address B2B, retention, and onboarding — zones that conventional frameworks overlook.

Skip “switching barriers” and you build products with excellent value propositions that users never actually adopt. “We’re used to Excel.” “Our contract runs through year-end.” “The team needs training.” These aren’t objections you can close with a better demo.

Skip “emotional work” and you solve the technical problem while missing the narrative. Users want to feel in control or appear professional — not “save two hours weekly.”

Data Points Connected to Artifacts

Each of 46 AICPO artifacts requires minimum coverage thresholds before it can be generated:

  • Value Proposition Canvas — functional + emotional work + undesired outcomes + success criteria + current solution (minimum 5 of 26)
  • Competitive Analysis — competitor identity + weaknesses + unique assets (minimum 3 specific points)
  • Pricing Strategy — payment willingness + budget + decision-maker (mandatory three)

Unclosed data points mark artifacts incomplete. The bot asks for missing information before generating.

This is different from “ChatGPT generates anything.” AICPO recognizes insufficient data and says so.

Graph as Cross-Session Memory

Research knowledge scatters across time. Monday interview, Friday follow-up, three weeks later another round — analysts reconstruct context from scratch each time.

Knowledge Graph holds everything. When a segment from interview three intersects with a pain point from interview one, the graph automatically creates the edge.

Data points work the same way: interview one closes items 1-10, interview two closes 11-20 — after the second session, DP Tracker shows 20/26 without reanalyzing the first. Research accumulates instead of restarting.

What You Actually Get

Standard researchers after 3-4 interviews manually code data: open spreadsheets, apply tags, find patterns. Two to three hours of work.

With AICPO this happens in real-time. After the final interview, you open the coverage map and immediately see:

  • Which data points are closed with high confidence
  • Which are closed but have graph contradictions
  • Which are still open with recommended closing questions

It doesn’t replace the researcher. It handles the mechanical coding and structuring — you keep the interpretation.