The problem this solves
Everyone can see the funnel leaks; almost nobody can see where. Traffic arrives, some of it converts, some of the conversions become pipeline, and between each stage a percentage vanishes for reasons that live in the gap between two systems. GA4 knows what happened on the site, HubSpot knows what happened to the records, and the leak hides precisely at the seam where one system's story hands off to the other's.
The manual version of this analysis is a spreadsheet safari: export landing-page conversions from one tool, lifecycle counts from the other, and try to join them on definitions that were never designed to match. Sessions do not map cleanly to contacts, conversion events do not map cleanly to lifecycle changes, and after a day of wrangling the analyst has a directional guess held together by assumptions nobody wrote down. The next person to ask the same question starts from zero.
A diagnosis worth acting on names the two biggest leaks specifically - which stage transition, which segments, how much volume - and quantifies them with stated assumptions and figures that trace to their source pulls. Then it proposes fixes proportionate to each leak and puts the whole thing where the team plans work, which is what turns a finding into a fix.
How the mission runs
- Map the site-side funnel from GA4. GA4 provides the top of the story: landing pages, the paths users take, and where conversion events fire, segmented by source and page group. The output pins which events were treated as conversions, because that definition is where funnel analyses usually go quietly wrong.
- Map the record-side funnel from HubSpot. HubSpot provides the rest: how contacts move through lifecycle stages, where they stall, and how long each transition takes by cohort. Stage definitions and date windows are stated explicitly, so the two halves of the funnel are joined on documented assumptions rather than hopeful ones.
- Join the funnel and find the leaks. The agent connects site behavior to record progression and computes stage-to-stage flow across the whole funnel, then isolates the two transitions losing the most volume relative to what comparable segments achieve. Two is deliberate: a diagnosis naming every leak at once prioritizes none of them.
- Quantify and explain each leak. For each of the two leaks: the volume lost, the segments where the loss concentrates, and the likeliest mechanisms given the evidence - a page problem, a handoff problem, a follow-up timing problem. Each figure traces to the GA4 or HubSpot pull behind it, and the reasoning is written for a skeptical reader.
- Write the diagnosis into Notion. The full diagnosis lands as a Notion page: the funnel map, the two leaks with quantification and mechanism, and a proposed fix per leak with expected recovery. Structured for planning rather than archiving, it slots into the workspace where the team already decides what to build next.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A Notion page holding the complete funnel diagnosis: the joined GA4-to-HubSpot funnel map with its assumptions stated, the two biggest leaks quantified by stage, segment, and lost volume, the likeliest mechanism behind each, and a proportionate fix proposal per leak. Every figure traces to the data pull that produced it, so the follow-up debate is about which fix to ship first rather than whether the numbers are real.
Make it yours
- Scope the analysis to one acquisition channel when you suspect the leak is channel-specific, and the diagnosis will confirm or clear the suspicion with numbers.
- Re-run it monthly after shipping a fix, holding definitions constant, so the same analysis that found the leak becomes the measurement of whether the fix worked.
- Extend past the two-leak limit into a full leak inventory when planning a quarter of funnel work, keeping the top two as the recommended starting points.
Frequently asked questions
How reliable is the join between GA4 and HubSpot?
It rests on stated assumptions rather than hidden ones, which is the honest version of every funnel join. Sessions and contacts are different creatures, so the diagnosis documents exactly how they were connected and flags where the mapping is soft. You can challenge an assumption and re-run with your correction.
Why only two leaks?
Focus. Every funnel leaks at every stage, and a report listing all of them becomes a wall nobody acts on. Two named, quantified leaks with proposed fixes is an actionable week of decisions. The full-inventory variation exists for planning cycles that genuinely need the complete picture.
Does this change anything in GA4 or HubSpot?
No. The mission reads both systems and writes only the Notion page. Fixes are proposals for humans to prioritize and ship, and if a fix later warrants CRM changes, that is its own mission with its own review gate rather than a side effect of a diagnosis.