The problem

A client with a high lead qualification bar and a small sales team took multiple days to follow up with qualified leads, often missing their opportunity to book meetings.

The team had a time-intensive lead review process, no one with time to do it, and was leaving business on the floor.

What we built

A lead scoring agent that uses LLM reasoning to score leads. No rigid scoring formulas. It uses enrichment data it searches for itself. No third-party tools. It operates in an automated system that flags high quality leads immediately and puts the rest in a queue.

Tools used

  • Claude and Claude Code Routines
  • Salesforce
  • Slack
  • Google Drive
  • Google Sheets

How it works

AI inbound lead analysis

We conducted an analysis of this client’s qualified leads over the past year with a high-reasoning model and produced a scoring model for our Lead Scoring Agent to use. The model assigned numeric values to inputs on the lead form (like country) and attributes determined during web search (like funding).

It relied on the AI agent’s reasoning for most of the lead’s score.

Separately, we had previously created a knowledge graph for this client. It gives all their agents consistent, updated context about the business, competitors, ICP, and martech stack (more on that in a future post).

This allowed our AI agent to reason about a lead’s potential fit using context about the client’s business in an automated workflow.

The workflow

  1. A Claude Code Routine runs hourly, checking for new inbound leads since its last run.
  2. For each new lead, the agent enriches via web search — company data (funding, headcount, geography, products) and lead data (LinkedIn, title).
  3. Claude reasons about fit using the enrichment data and a business knowledge graph, then produces a final score combining its reasoning score with values assigned to form inputs.
  4. The agent sends a Slack message if any leads were deemed high priority.
  5. The agent writes scores and enrichment data to a CSV and uploads it to Google Drive.
  6. An Apps Script in Google Sheets detects new CSVs, parses them, and adds rows to the sheet.
  7. A Sheets extension syncs new rows into Salesforce where leads are assigned to sales reps.

Routing and assignment

Leads that are determined to be high priority are posted to a dedicated Slack group for the team to respond to immediately.

High quality leads that were previously lost in a queue for review are surfaced with:

  • A summary of why they are a fit
  • Firmographic data about the lead
  • The lead’s message and timezone
  • Visibility and a sense of urgency for someone to respond

Other leads are put into a lead review queue in Salesforce, rank-ordered by their score.

Results

Average response time for qualified leads dropped from 4 days to 2 hours.

Why it worked & what’s next

The workflow rode existing tools, and the Lead Scoring Agent was equipped with the context it needed to score leads effectively. The score itself allowed the agent’s lead-specific reasoning to carry weight.

Surfacing only the best leads in Slack built trust with the sales team and provided visibility into their response times among leadership stakeholders.

Now that we’ve fixed this client’s leaky inbound lead funnel, we’re launching lead acquisition campaigns and scoping an AI SDR for the future.

This is an anonymized account of work performed by Lobo Growth for a client engagement.