Spotting the Visitor Worth Calling, While They Are Still on the Page
Analytics tools tell you what happened yesterday. By the time a sales team reads the report, the visitor who was ready to buy has been gone for a day and the moment has passed.
The Challenge
Conventional analytics answers questions about the past. A team opens a dashboard, reads what visitors did yesterday, and draws conclusions about people who have long since closed the tab. For a sales team that is close to useless, because the visitor who was worth a phone call was worth it while they were still on the pricing page. The product had to move the moment of insight from the morning report into the live session.
What We Built
We built HotLeads around scoring behaviour as it happens rather than aggregating it afterwards. Signals arrive continuously, get scored against what previous conversions looked like, and surface as a ranked answer to a specific question: which visitor should someone act on right now. The output is an instruction rather than a chart, which is the difference between a tool a sales team opens daily and one they stop opening.
Decisions Worth Explaining
BigQuery for depth, Firestore for the live path
Scoring a session cannot wait on a warehouse scan, and warehouse economics punish you badly for querying on every page view. Historical analysis lives in BigQuery while the live scoring path reads from a much smaller hot dataset, so the expensive store is used for the job it is good at.
Microservices split by data shape
Ingest, scoring and reporting have genuinely different load profiles: ingest is constant and spiky, reporting is bursty and rare. Splitting them meant a traffic spike on one customer site could not slow down the dashboard for everyone else.
The Result
The platform runs in production against tens of millions of events, scoring sessions live instead of reporting on them later.
Under the Hood
A Python and Node.js microservice backbone on Firebase Functions with BigQuery underneath, so behavioural scoring runs against tens of millions of events without the query cost of a warehouse scan on every page view.
Technology Stack
Want Similar Results?
Let's discuss how we can help solve your technical challenges.