How We Automated Lead Qualification for a $10M SaaS Company

The Problem: 25 Hours of Manual Work Per Week

Our client — a B2B SaaS company doing $10M ARR — had a problem that's painfully common: their sales team was spending more time sorting leads than actually selling. Every morning, three account executives spent the first 90 minutes of their day manually reviewing HubSpot, checking LinkedIn, and updating deal stages.

Across three reps, that's 22.5 hours per week — nearly a full working week of lost selling time every single month. We were brought in with one mandate: eliminate the manual work entirely.

The Architecture: What We Built

Rather than patching the problem with another Zapier workflow, we designed a proper event-driven architecture: a system that acts on data changes in real time, enriches leads automatically, scores them with an AI model, and routes them to the right rep — all within 90 seconds of a form submission.

The tech stack consisted of three main layers: a data ingestion and enrichment layer, an AI scoring engine, and an action layer connected directly to HubSpot and Slack.

Phase 1: Data Ingestion & Enrichment

Every inbound lead triggers a webhook that fires to our enrichment pipeline. Before any scoring happens, we fill the gaps in the lead record — pulling in company size, tech stack, funding status, and industry.

• Company headcount pulled from LinkedIn API in real time
• Tech stack detection via BuiltWith integration
• Funding rounds and estimated ARR from Crunchbase data
• Lookalike scoring against their top 50 existing customers

This enrichment step alone took the average data completeness rate from 34% to 91%. You cannot score a lead accurately on a half-empty record — and most companies are trying to do exactly that.

Phase 2: AI Scoring Engine

Once enriched, the lead record is passed to our scoring engine — a fine-tuned classification model trained on the client's own historical closed-won and closed-lost data. The model outputs a score from 0 to 100 across three dimensions: fit, intent, and timing.

Critically, we did not use a generic lead scoring model. Generic models fail because every business has a different ICP. We trained ours specifically on 24 months of historical data — 1,200 closed-won and 4,800 closed-lost records.

Phase 3: Automated Routing & Actions

Based on the score output, the system routes leads automatically and takes action without any human in the loop. A score above 75 triggers immediate priority routing — the lead is assigned to the right rep, a Slack notification fires, a follow-up sequence is enrolled, and a Calendly link is sent automatically.

• Score 75–100: Priority route → Rep notified on Slack → Auto-sequence enrolled
• Score 40–74: Standard route → Added to nurture queue → Weekly digest
• Score 0–39: Disqualified → Moved to marketing list → Tagged for re-engagement

The Results: Numbers That Matter

After 90 days of running live, the numbers were clear. This wasn't a marginal improvement — it was a step-change in how the sales org operated.

The 25 hours of manual work per week didn't just disappear — it was converted into selling time. Within the first quarter, the team booked 40% more demos without hiring a single additional rep.

Key Takeaways

If you're considering automating your lead qualification process, here is what this project taught us:

• Data quality is the foundation. Enrich before you score — always.
• Generic scoring models don't work. Train on your own historical data.
• Speed matters enormously. A 90-second response time vs. 4-hour response time is a conversion multiplier.
• Measure rep capacity, not just leads. The goal is more selling time, not just more leads.
• Build in a disqualification path. The time saved on bad leads is as valuable as time spent on good ones.