Marketing Intelligence System
How multi-agent research designed a three-score attribution model that quantified previously invisible marketing touches.
Enterprise Software, 8 weeks (research + implementation)
The client
A global technology company whose marketing team's revenue contribution was systematically undervalued due to last-touch attribution.
The challenge
The marketing team had no lead scoring model. No segmentation beyond basic lists. Attribution was last-touch only, which meant the channels that closed deals got all the credit while the channels that generated awareness and consideration got none. The result: budget decisions based on incomplete data, marketing's contribution to revenue underreported, and no way to distinguish a high-fit, high-engagement prospect from someone who downloaded one whitepaper six months ago. The team needed scoring, segmentation, and attribution built from the ground up.
The approach
Designed a three-score model separating Fit (firmographic match to ICP), Engagement (behavioral signals across touchpoints), and Product (usage signals indicating commercial intent). Used W-shaped attribution to distribute credit across four key moments: first touch, lead creation, opportunity creation, and close. The entire architecture was designed through a multi-agent research methodology: 7 research sessions across two phases, producing 50,000+ words of analysis before writing a single line of implementation code. Each session used 5-8 specialized research agents working in parallel. The scoring model was implemented in HubSpot using existing infrastructure at zero additional software cost.
Built with HubSpot, Claude AI multi-agent research, W-shaped attribution, and custom scoring algorithms.
How the engagement ran
Observe
Audited existing attribution, scoring, and segmentation. Found: no scoring, no segmentation, last-touch only attribution.
Decompose
Separated the problem into three independent scoring domains (Fit, Engagement, Product) to avoid the monolithic scoring trap.
Design
7 multi-agent research sessions over 2 phases. Each session produced structured analysis. Architecture decisions documented with tradeoffs.
Build
Implemented scoring in HubSpot. MQL workflow automated. Attribution model designed for the existing CRM infrastructure.
The results
First lead scoring model live in HubSpot. MQL workflow automated with score-based routing. PQL experiment designed with product data integration (pending data engineering access). W-shaped attribution model quantifying previously invisible marketing touches across the full buyer journey. The research methodology itself became a reusable pattern: structured multi-agent analysis as a prerequisite to implementation, producing better architecture decisions than traditional consulting approaches.
Is attribution undercounting your marketing?
Tell us about your operation. We will tell you what a system like this would take to build.