Built an AI-powered lead assistant that scored, routed, and prioritised inbound prospects using 9+9 research signals — and a feedback loop that kept making it smarter.
AI-Assisted Close Rate
Up from 52% baseline
Baseline Close Rate
Pre-scoring manual triage
Priority Threshold
Score to qualify as hot lead
Contact SLA
High-scoring leads
Inbound leads were being manually triaged by the BDR team with no consistent scoring framework. High-intent prospects sat in the same queue as low-quality inquiries. Response times varied wildly — a lead from a Fortune 500 CFO could wait as long as a student inquiry. There was no systematic way to prioritise by likelihood to convert. Close rate hovered at 52%.
The BDR team was oversaturated — hundreds of leads per week with no filtering mechanism. High-value leads from direct website inquiries and referrals were getting lost among cold inbound. Meanwhile, response time for hot leads drifted to 4-6 hours, sometimes longer. Every hour of delay reduced close probability by measurable margins.
Three months of lead data existed but had never been analysed for scoring patterns. The team operated on intuition — senior BDRs 'had a feel' for which leads would convert. But that intuition couldn't scale, couldn't be audited, and was lost whenever someone took a day off.
Audited 3 months of lead data. Two patterns changed everything.
Leads from direct website inquiries and referrals closed at 3x the rate of leads from cold channels. Source was the strongest single predictor of conversion — but the routing system treated all sources identically.
Leads responded to within 30 minutes were 4x more likely to convert than those responded to after 2+ hours. Time-to-response was the second-strongest predictor — but there was no SLA enforcement or escalation for slow responses.
Cross-referenced company size, industry vertical, and inquiry type against close outcomes. Built a weighted scoring model that assigned point values to each signal. The model predicted historical close outcomes with 83% accuracy in retrospective testing.
The data was conclusive: the existing manual triage was operating at a fraction of its potential. Not because the BDRs weren't skilled — because they didn't have the system to apply their skills where it mattered most. A scoring engine wouldn't replace the BDRs; it would let them focus on the 10% of leads that drove 60% of revenue.
Two-stage scoring engine. AI-powered. Human-verified. Built to learn from every interaction.
Two-stage scoring: Prospect Score (lead source, company fit, industry vertical, inquiry type, engagement signals) and Potential Score (budget indicators, decision-maker status, timeline urgency, purchase authority, integration fit). Combined into a 0-100 overall score.
High-scoring leads (75+) routed to senior BDRs within minutes with a pre-populated context brief — company name, industry, inquiry type, and recommended talking points. Low-scoring leads entered an automated nurture sequence with personalised email cadence.
Built a feedback loop where BDRs could flag scoring mismatches with one click — marking leads as over-scored or under-scored. The model ingested these flags weekly and adjusted signal weights. 200+ corrections in the first month, each one making the system more accurate.
Scoring Stages
Prospect + Potential
Feedback Corrections
200+ (Month 1)
Calibration Period
2-week shadow mode
Close Rate
52 → 71.63%
Response Time (Hot Leads)
Hours → <5 min
Scoring Accuracy
83% retrospective
BDR Focus Shift
All → top 10%
19.63 percentage point absolute improvement. The system paid for itself in additional closed deals within the first quarter.
Response time to high-scoring leads dropped from hours to under 5 minutes. The feedback loop captured 200+ scoring corrections in the first month, progressively improving the model's accuracy. BDRs reported higher job satisfaction — they were spending time on leads that converted, not leads that never responded.
AI didn't replace the BDRs. It gave them back their time.
The two-week shadow mode was the single best decision in the rollout. Running the scoring engine silently alongside the existing triage let us validate the model, catch edge cases, and build BDR trust before any lead was actually routed differently. Never skip the shadow mode.
The feedback loop was essential but I underestimated how quickly BDRs would adopt it. 200+ corrections in month one meant the model was improving weekly. Giving users agency over the system's accuracy created ownership — BDRs felt responsible for making the scoring smarter.
I should have started with a simpler rules-based system and added AI complexity later. The AI scoring was powerful, but 80% of the improvement came from the basic signals — source, response time, company size. The AI layer was the top 20%. Ship the simple version first, then layer intelligence on top.