SaaS · Case Study

AI Lead Qualification for a SaaS Sales Team

A funded SaaS company was losing deals because leads were scored by hand, so follow-up took around four hours. I led a team that built an OpenAI-powered agent to score, categorise, and route every inbound lead in real time, wired into their existing CRM over webhooks. Response time fell to five minutes, conversion improved by 35%, and the team closed 40% more deals in the first quarter.

Results

4 hrs → 5 min Lead response time
+35% Conversion rate
+40% Deals closed (first quarter)

The problem

A SaaS client was losing deals because manual lead scoring meant slow follow-ups. Sales team was overwhelmed.

The approach

I led a team to build an OpenAI-powered agent that scores, categorizes, and routes leads in real time. Integrated into their existing CRM via webhooks.

  1. Mapped the qualification criteria

    Before any model work, we wrote down what actually made a lead good for this business, so the agent had an explicit contract to score against rather than a vague prompt.

  2. Built scoring as structured output

    The agent returns a structured score and category the CRM can validate and act on, instead of free text that a downstream system has to guess at.

  3. Integrated over webhooks

    Routing plugged into the existing CRM through webhooks, so the sales team kept the tools they already used and there was no migration to sell internally.

  4. Measured against real follow-up

    Success was judged on response time and closed deals, not on model benchmarks — the metrics the client already tracked.

The outcome

Response time dropped from 4 hours to 5 minutes. 35% conversion improvement. 40% more deals closed in the first quarter.

Stack

Laravel OpenAI API Python Webhooks MySQL

Questions

How long does an AI lead qualification build take?

The scope above — scoring, categorisation, CRM routing, and monitoring — is typically a matter of weeks rather than months, because the integration surface is small and the criteria are agreed up front.

Does it replace the sales team?

No. It removes the manual triage step so the team reaches good leads faster. Routing and prioritisation are automated; the selling is not.

What happens when the model is unsure?

Low-confidence leads are routed for human review rather than silently mis-scored, which is what keeps the sales team trusting the system.

Building something similar?

If you are shipping AI features or scaling a remote engineering team, I work with founders across the UK, EU, US, and MENA.

Get in touch →