The Challenge
A B2B SaaS company with a product-led growth motion was generating 2,000+ new trial signups per month. Revenue was growing, but sales efficiency was declining — and the team couldn’t figure out why.
The core problem: sales reps had no way to distinguish high-intent trial users from casual signups. Their CRM showed every new trial as a lead, but the only data available was firmographic — company name, title, company size. Nothing about what the user actually did in the product.
The consequences were expensive:
- Reps called everyone. With no signal about product engagement, SDRs worked the list top-to-bottom, spending equal time on users who signed up and never came back as on users who’d completed onboarding, invited their team, and were hitting plan limits
- Conversion rates were declining. As trial volume grew, sales couldn’t keep up. Response times increased, and engaged trial users waited days for a call — by which point many had moved on
- The “qualified” in MQL was meaningless. Marketing qualified leads based on form fills and email opens. But a CMO who downloaded a whitepaper and never logged into the product was scored higher than a developer who’d been using the product daily for two weeks. The scoring model was backwards
The VP of Sales was frustrated: “We know product usage matters, but we can’t see it. We’re calling people blind.”
What We Found
Rich behavioral data existed — it was just trapped. The warehouse had detailed product events: logins, feature usage, workspace creation, team invitations, API calls, data volume, and more. The data team had already modeled these into clean, tested tables. But none of it reached the CRM.
Historical conversion patterns were clear. When we analyzed the last 12 months of trial-to-paid conversions, strong predictive patterns emerged: users who completed three specific onboarding steps, invited at least one teammate, and used a core feature within the first week converted at 8x the rate of users who didn’t.
The existing lead scoring was counterproductive. The marketing-based MQL score had a 0.12 correlation with actual conversion. Product engagement signals had a 0.67 correlation. The team was optimizing for the wrong signal.
What We Built
- Product-qualified lead scoring model — a gradient-boosted classifier (XGBoost) trained on 12 months of trial conversion data, using 23 product behavioral features: activation milestones, feature adoption depth, usage frequency, collaboration signals, and data volume trends
- Daily scoring pipeline — the model runs daily in the warehouse, scoring every active trial account and producing a 0-100 PQL score with a human-readable explanation of the top contributing factors
- Reverse ETL sync to Salesforce — PQL scores, contributing factors, product usage summary, and key milestones synced to the lead and account records via Census, refreshed every 6 hours
- Sales workflow automation — Salesforce workflows that route high-PQL accounts (score > 70) to the appropriate rep based on segment, trigger Slack alerts for score spikes, and auto-update lead status
- Rep-facing context cards — custom Salesforce component showing the prospect’s product journey: what they’ve done, what they haven’t, and what behaviors historically predict conversion
The Result
Qualified pipeline from inbound trials increased 40% in the first quarter. Reps were calling the right people at the right time — users who had demonstrated product engagement and were ready for a sales conversation. Conversion rates on PQL-flagged accounts were 4.2x higher than the overall trial base.
Average response time for high-intent prospects dropped from 3 days to 4 hours. The routing automation ensured that accounts crossing the PQL threshold got immediate attention instead of sitting in a queue.
SDR productivity nearly doubled. Reps stopped spending time on dead accounts. The number of meaningful conversations per rep per day increased from an average of 6 to 11 — not by working harder, but by knowing who to call.
Marketing-sales alignment improved. The PQL score gave both teams a shared language. Marketing could optimize campaigns for product engagement (not just form fills), and sales could give marketing concrete feedback on which user behaviors actually predicted revenue.
The model improves over time. As more conversion data accumulates, the model retrains monthly with updated feature importance. The data team manages this independently — no ongoing dependency on our team.
Talk to Us About Data Activation