A qualification pipeline that combines third-party enrichment, internal CRM history, and GPT-4o reasoning to score every inbound lead on fit and intent, then notifies the right AE in Slack within seconds.
The Challenge
This Series B B2B software company faced major sales inefficiencies because Sales Development Representatives (SDRs) triaged inbound leads on a first-in, first-out (FIFO) basis. Under this legacy setup, high-intent enterprise buyers sat in the queue for an average of 26 hours, while the SDR team spent valuable hours chasing low-fit, free-trial signups. This long lead response latency caused the company to lose valuable enterprise opportunities to faster competitors. Furthermore, SDRs wasted approximately 40% of their business day performing manual firmographic and technographic research on prospect organizations rather than engaging in high-yield outreach. The lack of standard, quantitative scoring criteria meant that every sales representative evaluated leads differently, creating massive data discrepancies within the Salesforce customer relationship management (CRM) database and leading to lost sales pipeline velocity.
Pain points we set out to solve
- ×26-hour average response time on top-of-funnel leads
- ×SDRs spent 40% of their day on lead research, not outreach
- ×No consistent scoring - every SDR triaged differently
- ×Enterprise leads and trial signups hit the same queue
Objectives
- 01Cut average lead response time to under 5 minutes for qualified inbounds
- 02Auto-enrich every lead with firmographics, tech stack, and news signals
- 03Produce a defensible fit-and-intent score the sales team trusts
- 04Route scored leads to the right AE in Slack with full context
Approach
How we delivered — phased, with clear checkpoints and evidence at each step.
- Week 1-2
ICP and signal definition
Workshopped the ideal-customer profile with Sales and Marketing. Mapped the firmographic, technographic, and behavioral signals that correlate with closed-won in the last 12 months of Salesforce data.
- Week 3-5
Enrichment and scoring chain
Built a LangChain pipeline that calls Clearbit, Apollo, and BuiltWith to enrich each inbound, then passes the enriched record to a GPT-4o scoring chain that returns fit, intent, and a one-paragraph rationale.
- Week 6-7
Salesforce and Slack integration
Wrote scores and rationales back to Salesforce as custom fields, and built a Slack bot that posts high-score leads to the right AE channel with a one-click Accept button that creates the Opportunity.
- Week 8-9
Calibration and rollout
Back-tested the scoring model on 9 months of historical leads to calibrate thresholds. Ran a 4-week parallel pilot where SDRs still triaged manually, then switched over once the AI matched or beat human scoring.
The Solution
The engineered solution is an automated lead enrichment and scoring pipeline built on the LangChain framework that qualifications and routes every inbound lead in under two minutes. The pipeline enrichment node automatically queries Clearbit, Apollo, and BuiltWith to compile comprehensive corporate profiles, including technographic data and technographical news signals. An advanced scoring chain powered by GPT-4o then runs a Retrieval-Augmented Generation (RAG) query over a vector store of the company's historical closed-won and closed-lost deal history in pgvector. This allows the system to generate a highly structured fit-and-intent score grounded in real sales precedents rather than arbitrary rules. Scored opportunities are written directly to Salesforce and pushed to a dedicated Slack channel using the Slack Bolt framework, enabling account executives to accept leads and create sales opportunities in one click.
Multi-source enrichment
Combines Clearbit firmographics, BuiltWith technographics, and recent news signals into a single lead record before scoring.
RAG-grounded scoring
Scoring prompt pulls similar closed-won and closed-lost deals from a vector store of prior CRM history, so each score is grounded in precedent.
Slack routing with AE match
High-score leads post to the right AE channel in Slack with context, rationale, and a one-click Accept that creates the Opportunity.
Self-tuning thresholds
A weekly job re-calibrates score thresholds based on the last 30 days of SDR feedback and deal outcomes.
Technology stack
Picked for latency, cost, and long-term maintainability — not for novelty.
AI / Agent
Enrichment
CRM / Messaging
Infra
Results
Business impact
The sales team stopped triaging and started selling. Pipeline-sourced revenue per SDR climbed 27% in the first quarter, and the AI-first triage layer is now the default for all new lead sources the company onboards.
Key takeaways
- Fit scoring fails when grounded in generic ICPs - RAG over your own closed-won data is what makes it defensible
- Speed-to-lead is a bigger multiplier than prettier enrichment - optimize the end-to-end latency first
- Keep a human accept step in the loop for the first quarter: it builds trust and generates the feedback data you need to tune thresholds
Visual reference
