From Mental Model to Graph: Building the Fundamentals of a Central Customer-Intelligence System
What single factor most improves marketing (resonance) and product development (relevance)?
A holistic understanding of customers—functional and emotional progress in context (journey phases, triggers, constraints, habits). If we could capture qualitative customer research in a way that any team can query on demand, we’d ship better ideas faster.
What if those qualitative insights lived in a graph—and everyone could access them, as connected evidence, anytime?
1) The origin: a mental model of “customer progress”
In marketing & innovation work we used a board called The Wheel of Progress®—a practical way to map interviews to forces and outcomes (pushes, pulls, pains, anxieties, habits, trigger events, jobs-to-be-done, desired outcomes, gains…).
This mental model matured into the 12 Elements of Customer Progress Design®: a domain language for how customers make progress over time.
2) From mental model → data model
We formalized the 12 Elements with clear definitions, syntax, and allowed relations. Early on, we created partially connected nodes “by hand” to learn the shape of the data (what belongs where, and why).
That schema became our ontology for customer progress:
-
Node labels: e.g.
JTBD,Pain,Pull,DesiredOutcomeetc. -
Relationship types (directed, typed):
CAUSES,HINDERS, etc.
We attach evidence to everything:
source_id,case_id, etc.
3) Into Neo4j: schema, triples & import
Once the language stabilized, we automated extraction from interviews (GenAI-assisted classification → human verification), produced probable triples, and imported to Neo4j.
4) Why a graph? Because customer reality is relational
The value appears when you traverse motifs like:
Pain → Push → First Thought → ActiveSearch
These paths are explanations you can hand to humans and machines.
5) What we can do now (Graph → GenAI → output)
Because the graph already encodes who, what, why, when (with evidence), we can prompt GenAI with directed, typed, evidence-weighted paths instead of vague personas or scattered bullet points. This is a huge feature of using LPGs- the right (connected) data at the right time.
-
Persona maps (derived from evidence, not invented): Cluster by motifs and phases; produce on-demand summaries that stay linked to sources.
-
Marketing messaging: Generate headlines, CTAs, and narratives that explicitly address high-weight
Push→Pain→JTBD→DesiredOutcome→Solutionchains. -
Innovation ideas: Ask for features that relieve a
Constraintand enable aDesiredOutcome, with acceptance criteria and likely trade-offs. -
Sales enablement: Objection handling from
Anxiety/Habit HINDERS Solutionedges, with counter-stories anchored in real cases.
This “graph-first” RAG approach increases precision, reduces hallucination, and keeps explainability (“why did the model say that?” → follow the path).
6) What surprised us
-
We consistently see fewer iterations and higher first-draft relevance when prompts carry paths + evidence weights (vs. text chunks alone).
-
Teams align faster because the graph encodes shared language + provenance.
-
Unstructured qualitative data—once trapped in presentation decks—becomes a living, queryable asset.
7) What’s next
-
Stronger evidence pipelines.
-
More automated motif detection.
-
Fine-tuning graph-aware RAG
Why share this here?
We moved from a domain mental model → data model → ontology/schema → machine-linked graph → triple import and now use Neo4j to power customer-intelligent outputs with GenAI. It’s a pragmatic pattern you can reproduce in other domains where qualitative context matters.
We are still in an early stage but one thing is clear already: applying graph technology/Neo4j is a game-changer for us.
Thanks for reading—and thanks to the Neo4j community for pushing graph thinking where it has the most human impact: making better decisions, faster, with real customer reality.
