Skip to main content

Customer Intelligence Graph

The Customer Intelligence Graph is FunnelStory's organized context graph for B2B customer data. It connects customer entities — accounts, contacts, meetings, activities, and interactions — with internal workspace entities like teams, notes, and assignments, and continuously computes derived intelligence on top of that unified context. Every AI agent, copilot, and workflow in FunnelStory reasons against this graph rather than querying raw source systems directly — it is the layer that sits between your data and everything that acts on it.

Why the Graph Exists​

An AI agent is only as good as the context it reasons over. Point an agent at raw source data and it has to become a data analyst at runtime — querying, joining, and guessing at what fields mean, then hallucinating when the data is ambiguous or incomplete. Reliability isn't a property of the model; it's a property of the foundation the model stands on. The graph is that foundation: because intelligence is computed, validated, and cited before an agent touches it, agents return grounded answers instead of on-the-fly inferences over noisy data.

The graph also closes a gap that structured-only systems can't. The signals that most reliably precede churn and expansion live in unstructured data — a champion's tone shifting in email, a competitor mentioned favorably on a call, a pricing objection raised in a QBR, a commitment made and never kept. A warehouse or BI stack built on structured fields is blind to all of it. A core job of the graph is to turn that unstructured conversation into structured, queryable, source-cited intelligence and fuse it with the structured data you already have.

What the Graph Contains​

The graph is structured in layers that build on each other.

Knowledge Layer

The knowledge layer holds the raw and normalized facts about your customers and your workspace:

  • Customer objects — accounts, contacts, interactions, product usage events, support tickets, conversations, meetings, emails, and signals sourced from your connected tools
  • Workspace objects — teams, CSM assignments, workflows, recorded actions, notes, and internal configurations

Every object is timestamped, source-attributed, and versioned. When source data is incomplete or contradictory — a missing champion contact, duplicate account records, empty CRM fields — those gaps and conflicts surface explicitly as graph properties rather than being silently dropped. Wrong or empty fields are preserved with their provenance so you can see where data came from and why it looks the way it does.

Derived Intelligence Layer

The derived intelligence layer sits on top of the knowledge layer and holds everything FunnelStory computes from the raw context before any agent or user queries it:

  • Sentiment — tone and themes extracted from conversations, tickets, and emails
  • Events — detected moments of significance (a spike in usage, a drop in engagement, a contract milestone)
  • Predictions — churn risk, renewal likelihood, and expansion probability from models trained on your own historical outcomes, with confidence ratings
  • Needle Movers — leading indicators of risk and opportunity detected 3–9 months before a revenue event, drawn from unstructured conversation as well as usage
  • Relationship intelligence — team-to-contact coverage, engagement strength, and single-threaded and engagement-drop risk
  • Cohort factors — how an account compares to peers with similar profiles and histories
  • Norms and gaps — baselines for what "normal" looks like for an account, and deviations from those baselines

This separation is deliberate. By computing intelligence before the agent touches raw data, FunnelStory ensures that every query returns grounded, validated outputs — not inferences made on the fly from noisy source data.

Accrued Memory

Beyond point-in-time facts and scores, the graph accrues a durable, evolving memory of each account. The Account Dossier maintains a two-layer memory — recent activity and long-term context — folding new conversations in over time, keeping durable facts even after the conversations that surfaced them age out, and citing every fact back to its source. This is the wisdom a tenured CSM carries about an account, made durable and shared across your team and its agents. The graph is designed to get richer the longer an account runs on it.

How the Graph Is Built​

FunnelStory connects to your data sources through over 37 connectors spanning CRMs, data warehouses, product analytics platforms, support tools, communication platforms, and enrichment providers. Structured data (usage records, CRM fields, tickets) and unstructured data (call transcripts, chat logs, documents, notes) are ingested and fused into a single unified model at ingestion time — not joined at query time.

This patent-pending information fusion architecture is what makes the graph reliable for AI reasoning. Parallel data silos that must be reconciled at runtime introduce latency, inconsistency, and hallucination risk. By resolving everything at ingestion, the graph presents a single coherent view that agents can trust.

Handling Messy Enterprise Data

Enterprise data is rarely clean. CRM records have custom fields, duplicates, and partial entries that vary by tenant. FunnelStory's ingestion pipeline handles this through:

  • Schema alignment and normalization — source fields are mapped to canonical entity types and relationships regardless of how each system structures them
  • Entity resolution and deduplication — where connectors and rules allow, duplicate records for the same account or contact are resolved into a single entity
  • Provenance tracking — every record carries its source, timestamp, and version, so you can trace any value back to where it came from

The graph does not erase bad source data. If a CRM field is wrong, that wrong value is preserved with its source attribution. If a champion contact is missing, that gap is an explicit property of the account node — something agents and workflows can act on — rather than a silent omission.

Querying the Graph​

The graph uses a semantic database with a SQL interface and a federated query engine. It is designed specifically for agent-driven access: virtual tables present the graph in a familiar relational format while the engine manages the underlying complexity of fetching, joining, and computing data across different systems.

The database provides:

  • Workspace isolation — each FunnelStory workspace's data is fully isolated
  • Encryption — data is encrypted at rest and in transit
  • Embedding helpers — built-in support for document embeddings used in semantic search and unstructured data queries

Renari and AI Agents query the graph through this interface, retrieving pre-computed intelligence and raw context to answer questions, generate briefings, and take actions. Because intelligence is pre-computed at the derived layer, agents retrieve fast, validated answers rather than performing expensive joins or LLM-side reasoning over raw data.

The MCP Server exposes the graph's tools and data to external AI environments — Claude Desktop, Cursor, ChatGPT Enterprise, and any MCP-compatible client — using the same semantic database interface. External copilots get the same grounded context that Renari uses internally, under the same workspace isolation and permissioning.

What the Graph Gives You​

  • Agents you can rely on — grounded in validated, cited intelligence rather than raw data reasoned over at runtime
  • One shared source of truth — intelligence is computed once, centrally, so every teammate and every agent works from the same facts, with low latency and role-based access over what each can see
  • Visibility into unstructured signal — the conversation-borne indicators that structured-only analytics miss entirely
  • Memory that compounds — account context accrues and sharpens the longer you run on FunnelStory
  • Works with the AI tools you already use — the same grounded context is available to Renari and to any MCP-compatible copilot
  • How FunnelStory Works — how the graph fits into the three-layer platform architecture
  • Predictions — how prediction scores in the derived intelligence layer are computed and validated
  • Needle Movers — leading indicators drawn from the graph's unstructured data
  • Account Dossier — the accrued, cited account memory the graph maintains
  • AI Agents — Renari, agent automation, and the MCP Server
  • Data Connections — the connectors that feed data into the graph