Why AI Fails Without the Context Graph

The Dirty Secret of Enterprise AI

You can have the most sophisticated AI in the world, but if you ask it "Should I discount this deal?" without context, you get a useless (but dangerously confident) answer.

The AI needs to know:

What happened in the last three calls?
How has the champion's sentiment shifted? What did we do on similar deals?
Who else is involved? What's the competitive situation?

Foundation Capital's recent piece on "Context Graphs" captures this well, calling it the "missing layer": the decision traces, exceptions, and precedents that live in Slack threads, deal desk conversations, and people's heads.

We've been building exactly this for months. We call it the Common Customer Data Model (CCDM).

What Is the CCDM?

ServiceNow transformed IT operations with the Common Service Data Model (CSDM), a standardized schema that defines how every IT asset relates to every other. Before CSDM, IT data was scattered across tickets, CMDBs, and spreadsheets. After CSDM, it became a coherent system that was shared across all the various IT teams so issues can cascade automatically to impacted teams and services.

Go-to-market data is in the same pre-CSDM state today. Customer interactions live in Gong. Contact data lives in Salesforce. Email sequences live in Outreach. Meeting notes live in Notion. No single system captures how all of these relate to each other, which is why AI can't actually help.

The CCDM is our Context Graph for go-to-market: a living network of relationships between every entity and activity in your GTM motion, built on a standardized schema:

Accounts connected to Contacts connected to Opportunities
Activities (calls, emails, meetings) linked to all three
Transcripts attached to Activities with full conversation content
AI Assessments connected to the data that informed them
Workflow Executions that show what inputs led to what outputs
Historical Traces preserving how all of these evolved over time

When everything is connected in a coherent model, AI has context. When AI has context, it can actually help.

Why AI Doesn't Work Without a Context Graph

Imagine asking an AI: "Is this deal healthy?"

Without a Context Graph:

The AI sees: $500K, Stage 3, Close Date: March 15

The AI says: "Based on the deal size and stage, there's a moderate probability of closing."

You learn: Nothing you didn't already know.

With the CCDM:

The AI sees: $500K deal with 14 logged activities over 3 months. Champion mentioned budget concerns in the December call. Engagement dropped 60% in January. Three decision-makers were active early but only one has engaged in the last 6 weeks. Similar deals with this pattern closed at 34%.

The AI says: "High risk. Champion sentiment has declined since December. Multi-threading has collapsed. Engagement pattern matches churned deals from Q3."

What you learn: Exactly what's wrong and what to fix.

The difference is in the data model, not the AI model.

Building the CCDM: Why the System of Record Matters

Here's where most vendors get it wrong.

Conversation intelligence tools like Gong capture great data. They have transcripts, talk ratios, sentiment signals. But they're an analytics layer sitting on top of a broken CRM. They can tell you what happened on a call. They can't ensure that insight actually flows into the system that drives your workflows.

The result? Your CRM stays empty. Your forecasts stay unreliable. Your reps still spend hours on manual data entry.

We took a different approach. We built the CCDM as the system of record, not as an analytics layer, but as the source of truth that actually populates your CRM fields.

When a call happens, we don't just analyze it. It's critical to:

Identify all participants and match them to contacts (or create them)
Link contacts to their accounts
Associate the activity with open opportunities
Extract structured data: next steps, objections, stakeholder roles, competitive mentions
Write it back to Salesforce or HubSpot, automatically populating the fields that drive your workflows

The difference is accountability. We don't just tell you what's happening. We ensure it's captured in the system your business runs on.

Activities as the Connective Tissue

Every interaction (call, email, meeting) becomes a node in the CCDM. But a node alone is useless. What matters is how it's connected:

Which contacts participated in this call?
Which opportunity does this advance?
Which account does this belong to?
Who from our team was involved?
What came before this in the engagement sequence?
What did the AI extract from this conversation?

We don't just log activities. We weave them into the graph. When our system processes a calendar event, it identifies all participants, matches external attendees to known contacts, links contacts to their accounts, associates the activity with open opportunities, connects it to previous activities in the engagement timeline, and tags it with the AI insights extracted from any recording.

Now when you ask about an opportunity, the AI doesn't see a single record. It sees the complete web of relationships and history that the CCDM captures.

AI Assessments as First-Class Objects

Most systems treat AI outputs as fields, just data attached to a record. In the CCDM, they're first-class objects with their own relationships:

What inputs informed this AI assessment?
What transcript did it analyze?
What was the previous assessment?
What reasoning led to this conclusion?
Who triggered this analysis and when?

When an AI health score changes from 8 to 6, we don't just update a number. We preserve the old score and its reasoning, the new score and its reasoning, the activity that triggered re-analysis, the workflow execution with all inputs and outputs, and the timestamp so you can see the progression.

This means the AI can reference its own historical reasoning. "Last month I flagged timeline risk. The December call confirmed it: the customer pushed implementation to Q2."

Beyond Sales: The Cross-Functional Context Graph

Sales is where the pain is most acute, but the real power of the CCDM is what it enables across the entire GTM org.

Today, every team operates in isolation:

Marketing runs campaigns without knowing what messaging actually resonates in sales conversations
Sales closes deals without visibility into what support issues plagued similar customers
Customer Success onboards accounts without knowing what was promised during the sales cycle
Product ships features without knowing which customers need them or what pain points drove the request

The CCDM breaks down these walls. Because every interaction, from every team, connects to the same accounts, contacts, and opportunities, insights flow across functions automatically.

Marketing gets signal from the field. When prospects consistently ask about a specific use case on discovery calls, marketing sees it. When a competitor keeps coming up in late-stage deals, marketing knows to adjust positioning. The CCDM surfaces what's actually resonating, not what marketers hope is resonating.

Customer Success inherits full context. When a deal closes, CS doesn't start from zero. They see every conversation, every objection that was raised, every feature that was promised, every stakeholder's priorities. The handoff isn't a summary doc that gets outdated immediately. It's a living graph of everything that happened.

Product sees the real voice of customer. Feature requests aren't filtered through three layers of telephone. Product can trace a request back to the actual customer conversations where the pain was expressed. They can see which segments care most, which deals were lost because of the gap, which customers would expand if the feature shipped.

RevOps gets a single source of truth. No more reconciling data across six systems. The CCDM is the canonical model. Forecasts, territory planning, capacity models, all pull from the same connected graph.

This is what ServiceNow's CSDM did for IT: it created a shared language and shared data model that let every team operate from the same foundation. The CCDM does this for go-to-market.

The Compounding Effect

What makes the CCDM so powerful, is that it gets smarter with every interaction.

Every call processed adds new activity nodes linked to contacts and opportunities, transcript content connected to the activity, AI insights attached to opportunities, contacts, and accounts, historical traces showing how assessments evolved, and engagement patterns that inform future predictions.

After three months of building the graph, the AI can answer:

"How does engagement on this deal compare to our average won deal at this stage?"
"Which stakeholders have gone quiet compared to earlier in the cycle?"
"What objections have come up across multiple calls, and how were they addressed?"
"What patterns in this deal match deals we've lost?"

After a year, it becomes institutional memory:

"We've seen this exact situation before. Here's what worked."
"This competitor always gets brought in at this stage. Here's how we've won against them."
"Deals with this engagement pattern close 40% faster when we multi-thread early."

The AI becomes experienced because the CCDM allows for remembrance.

What the CCDM Captures

The "What" Layer

Accounts, Contacts, Opportunities. Activities (calls, emails, meetings). Transcripts and conversation content.

The "How" Layer

Relationships between entities. Engagement sequences and timelines. Changes over time (historical traces).

The "Why" Layer

AI reasoning and decision traces. Workflow executions with inputs/outputs. Human annotations and overrides.

The "So What" Layer

Patterns that predict outcomes. Anomalies that signal risk or opportunity. Precedents that inform future decisions.

The Structural Advantage

Foundation Capital argues that incumbents can't build Context Graphs.

We've seen this firsthand, and here's why they're right:

CRMs store current state. Salesforce knows what the opportunity looks like now. It doesn't preserve the context that led to changes. The model is flat, not rich. And it relies on humans to enter data that humans never enter.

Call recorders capture conversations. Gong has the transcript. But it doesn't see the email thread, the calendar context, the CRM stage changes, the previous calls in the sequence. Isolated nodes, disconnected. And critically, it doesn't write back. The insights stay in Gong while your CRM stays empty.

Warehouses receive data after the fact. By the time ETL runs, the decision context is gone. You can query what happened. You can't query why.

We capture relationships at decision time, when the activity happens, when the connections are clear, when the AI can synthesize across systems with full context and write structured data back to the systems that need it.

The CCDM isn't a feature we added. It's the architecture we built from day one.

Looking Forward

The question Foundation Capital poses (will entirely new systems of record emerge for decisions?) is one we've been answering in production. Will new platforms emerge that capture not just data, but the relationships that make data meaningful? We believe yes, and we're already seeing it work.

Because here's the truth: AI agents are only as good as the context they can access. An agent without context is just a chatbot. An agent with a complete Context Graph, with every relationship mapped, every decision traced, every pattern learned, can actually take action.

Not just summarize your calls. Populate your CRM fields. Not just flag deal risk. Tell you exactly which stakeholder to re-engage and why. Not just analyze patterns. Execute the playbook that worked last time.

That's what we're building.