Sales Intelligence and
Follow-Up Automation

The team was generating leads but losing them in the gap between interest and follow-up.

The sales team had inbound volume coming in across Instagram DMs, Facebook messages, and Slack. The problem was that every message sat in a queue. Someone had to read it, decide if it was worth pursuing, write a reply, log it somewhere, and then remember to follow up. That process was slow, inconsistent, and invisible to the team lead.

There was also no intelligence layer on top of calls. The team was doing discovery calls via Fathom but the recordings sat unreviewed. No one was extracting objections, identifying patterns in what was working, or flagging deals that had gone cold. The team was operating blind on their own pipeline.

The brief was to build something that monitored all three inbound channels, scored leads automatically, alerted the right person at the right time, and gave the team lead a daily view of what the pipeline actually looked like.

Three separate problems needed three separate workflows with one unified reporting layer.

I mapped the inbound channels first. Instagram, Facebook, and Slack each deliver messages in different formats, through different APIs, with different rate limits and authentication models. The naive approach was to funnel them all into one workflow. The better approach was to handle each channel separately and have them all write to a shared data layer that the reporting workflow could read from.

For lead scoring, I chose Claude over a rules-based classifier because the inputs were conversational and ambiguous. A rule that looks for the word "budget" will miss a message that says "we have room to invest in the right solution." Claude understands intent rather than just pattern-matching on keywords.

The Fathom integration was the most technically involved piece. Call recordings are long-form audio. I designed the workflow to receive the Fathom transcript rather than the raw audio, pass it to Claude with a structured extraction prompt, and produce a call summary that captured objections raised, buying signals given, and the agreed next step. That summary then fed back into the lead record automatically.

22 nodes across 3 workflows. Every inbound message scored, every call analysed, every morning a full pipeline brief.

• 01
  Multi-channel lead monitoring and scoring

  Webhook triggers fire on every new DM across Instagram, Facebook, and Slack. Each message is passed to Claude AI with a scoring prompt that evaluates urgency, intent, budget signal, and fit against the client's ideal customer profile. The output is a score from 1 to 10 with a one-line qualifier. HOT leads (score 8 and above) trigger an immediate Slack alert to the assigned rep with the full message and context. Warm leads are logged to Google Sheets for next-day follow-up.
• 02
  Call intelligence extraction

  When a Fathom recording is completed, the transcript is sent to n8n via webhook. Claude extracts three structured outputs: a deal summary, a list of objections raised during the call, and the agreed next step. These are appended to the corresponding lead record in Google Sheets and a short summary is posted to the team's Slack channel so the team lead sees every call outcome without needing to listen to recordings.
• 03
  Daily pipeline report

  Every morning at 8am, a scheduled workflow reads the Google Sheets data from the previous 24 hours, aggregates lead counts by channel and score tier, flags any leads that received a HOT alert but have not had a reply logged, and generates a plain-English daily brief. The brief is posted to Slack. The team lead starts every day knowing exactly where the pipeline stands without opening a CRM.

The team stopped losing leads to slow reviews and started acting on intelligence, not instinct.

The most significant change was visibility. Before the system, the team lead had no reliable picture of what was happening in the pipeline day to day. They were relying on individual reps to flag things. With the daily 8am brief, they had a consistent, accurate view every morning without asking anyone.

The HOT lead alerts changed the response behaviour on the floor. When a high-intent DM came in, the rep received a Slack notification with the message, the score, and the qualifier within seconds. Response times on high-value leads dropped from hours to under five minutes.

The call intelligence layer took longer to show value, but after three weeks the team lead had identified two consistent objections appearing across recordings that had not been visible before. They adjusted their call approach based on that data, something that was impossible to do when recordings sat unreviewed.