Case Study / Agency / Lead Generation

Apex Growth Partners: AI Lead Qualification Engine

A growth marketing agency with 12 employees was drowning in unqualified leads: Tally collected 200+ submissions per month, sales spent 80% of their time on prospects with no budget, no urgency and no fit, and the close rate had dropped to 8%. The rebuild scores every lead in 3 seconds and routes only qualified ones to sales, cutting junk pipeline by 60% and lifting closes by 45%.

01 The problem

Apex Growth Partners, a growth marketing agency with 12 employees, was drowning in unqualified leads. Their Tally intake form collected 200+ submissions per month, but sales spent 80% of their time on Discovery calls with prospects who had no budget, no urgency and no fit. Close rate had dropped to 8% and the team was burned out on meetings that went nowhere.

The agency tried manual qualification , a sales assistant reviewing every form submission and tagging leads as hot, warm or cold. This added 4 hours of delay and was inconsistent. Some great leads sat untouched for a day. Some terrible leads got priority because the assistant misread the form.

02 The architecture

I built an LLM-powered qualification engine that scores every incoming lead in under 3 seconds. The system reads the free-text form responses, extracts pain signals (broken workflow, lost revenue, manual overhead), evaluates budget indicators (team size, current tool spend, growth stage), assesses urgency (when do you need this fixed) and maps fit against the agency's ideal client profile.

Each lead receives a score from 0-100 with a breakdown: pain intensity, budget readiness, urgency level and fit score. Leads above 70 are routed directly to the senior account manager with a pre-written context summary. Leads between 40-70 go to a nurture sequence. Leads below 40 receive a polite "not the right fit" response with a referral to free resources.

The step after the score is where most of this lives: my automated lead routing guide covers assignment, CRM writeback and the failure queue.

The scoring model was trained on 6 months of Apex's historical data , every closed-won and closed-lost deal, their form responses and the actual outcomes. This means the model learns what Apex's real buyers look like, not what a generic lead scoring template assumes.

Figure 1. The qualification pipeline, from form submission to routed lead.

03 Before / after

04 Results

Within 60 days, Apex's close rate went from 8% to 12% , a 45% improvement. Sales stopped wasting time on Discovery calls with unqualified prospects. The senior account manager reported working fewer hours while closing more deals because every meeting was pre-qualified and context-rich.

The system processes every lead automatically. No manual review. No delay. No inconsistency. The managing partner can see in real-time which intake channels produce the highest-scoring leads and has shifted ad spend accordingly.

05 What broke

Four things were broken in the qualification process before the build. Every one of them now has a named component behind it.

1. 200+ unqualified submissions a month reaching a sales team that spent 80% of its time on Discovery calls with prospects who had no budget, no urgency and no fit. Fix: an LLM scoring engine that rates every incoming lead in under 3 seconds before it reaches a calendar.
2. Manual hot, warm or cold tagging by a sales assistant, which added 4 hours of delay and was inconsistent. Fix: a 0-100 score with a breakdown of pain intensity, budget readiness, urgency level and fit score.
3. Some great leads sat untouched for a day while terrible leads got priority, because the assistant misread the form. Fix: a model trained on 6 months of Apex's historical data, every closed-won and closed-lost deal and their actual outcomes.
4. A close rate that had dropped to 8% and a team burned out on meetings that went nowhere. Fix: only qualified leads hit the sales calendar, and the close rate went to 12%.

06 The stack

07 Related case studies

FAQ Frequently asked questions