The Marketer Graph for Insurance

Marketing has always had a difficult relationship with correlation vs. causation. On a Monday morning a marketer logs into their analytics dashboard that (mostly) reliably shows spend went up last week but leads went down, two seemingly related numbers moved in opposite directions. The Slack hits an hour later - "Why are the leads down? MBR is tomorrow, we need to know what happened, what the impact is and what we're going to do to fix it ASAP." So the why, the actual cause, behind multi-million dollar acquisition decisions gets passed to a few brave marketers armed with 9 different source systems that don't speak to each other and require tickets to extract the data they need, some spreadsheets, ChatGPT that confidently concludes a minor tracking issue "explains everything", maybe some SQL, 10 Slack threads with data, engineering, agency partners and a deadline to answer the cause with a full build-up, the fixes and an implementation plan with confidence by COB.

From Day 1 Fount has been focused on closing that gap: turning the mess of insurance marketing and customer data into a verifiable structure, a single connected graph of how a business actually fits together to measure cause and effect. From click to bind and beyond. We have gone through many iterations of this, through messy warehouses, overconfident agents, broken pipelines, identity resolution challenges, hallucinations, broken tracking and attribution issues.

We recently took a big step forward in solving this problem, both in how we represent the cause and effect of marketing actions and, importantly, in how we get our agents to leverage that understanding to take actions themselves in a trusted, verifiable and governed manner.

This development is summarized as the Marketer Graph, representing the world of insurance growth, from paid campaigns, ad creative copy (yes, the actual images and videos as well), leads, quote information, click events, call transcripts, insured assets, not just as a tightly connected series of identity-resolved relational data sources, which is where we started, but as a causal graph that agents can traverse and selectively reason over without running full queries. This layer enables Quinley Graph, the latest version of our agent harness which, compared to Quinley Base and Quinley Prime, shows a 40%+ performance improvement on verifiable tasks, mostly driven by significantly stronger performance on complex multi-hop tasks, in our internal insurance marketing evals while reducing average context usage by 4.3x, creating efficiency and leaving significant headroom for other agent tasks and tools.

This enables every number, suggestion and conclusion an agent produces to be traced to its source and verified against many prior related instances with confidence scores and so that governance can be imposed on conclusions and actions. This structure is the foundation for where we are really headed, which is true causal AI for marketing: an agent that understands what happened, anticipates what a change will do before it is made, and then makes the change itself, under governance, with the confidence required when a single budget decision moves millions or tens of millions of dollars.

In this series

This is the overview. The five parts below go deeper on each piece of how the Marketer Graph works and where it is taking us:

1. The causal growth marketing problem - why "why are the leads down?" still eats a marketer's week.
2. The evolution of Fount's agent harness - the three versions of Quinley, from Quinley Base to Prime to Graph, and what the evals show.
3. Turning marketing into a graph you can search - the structure underneath the agent, and how it finds the right knowledge by meaning.
4. Trust, rails and how we know it's right - governed execution, traceability, and how we verify every answer.
5. Causal AI and governed action - from correlation to cause, and from answering to acting.