Project Description

For Teams · Sales intelligence

Every sales call carries intelligence — competitor mentions, feature requests, objection patterns, buying signals — but at hundreds of calls a week it gets lost in the noise. CallForge is an 8-workflow AI system that turns raw call recordings into structured, actionable intelligence.

Built to integrate with Gong, it ingests transcripts, runs specialized AI pipelines, and routes the results to the teams that need them — objection reports to sales, competitive intel to marketing, feature trends to product — delivered automatically into Salesforce and Notion.

Key features

8-Workflow Architecture

Each workflow handles a specialized task — modular, maintainable automation at scale.

Competitor Intelligence

Automatically detects and tracks competitor mentions across every sales conversation.

Multi-Team Routing

Insights delivered to sales, marketing, and product — each in the tool they already use.

Objection Pattern Analysis

AI surfaces recurring objections and trends to sharpen your sales playbook.

Built with

n8nGong APIOpenAI GPT-4SalesforceNotion APIWebhook orchestration

The full build · part of The Last Layer series

The goldmine nobody has time to mine

Every sales call carries competitive intelligence — competitor mentions, feature requests, objection patterns, buying signals. At hundreds of calls a week, it’s sitting in Gong and nobody has the bandwidth to mine it.

1,247 calls last quarter. Every one recorded. Not one of them mined.

Eight workflows, one pipeline

AI in the middle, on rails: structured context goes in, structured intelligence comes out — never a raw transcript and a prayer.

1 · The Filter

Queries Salesforce for opportunities in the stages that matter and pulls only those Gong calls. The AI budget stays predictable.

2 · The Prep Stage

Pulls each transcript, enriches it with deal context from Sheets, and stages the call in Notion — context, not a raw dump.

3 · Transcript Processing

Cleans Gong’s formatting, chunks long calls, and writes the non-AI signals — talk-time, duration — straight back to Salesforce.

4 · The AI Analysis Core

A structured prompt extracts competitors, feature requests, objections, buying signals, and risk flags as scored JSON.

5 · Azure Triangulation

A parallel model scores sentiment and technical depth. Two models agreeing raises confidence; disagreeing flags a human.

6 · Aggregate + Human Check

Rolls insights up across every call into Notion dashboards. A human approves before anything is distributed.

7 · Sales Routing

Enriches the Salesforce opportunity and pings the rep in Slack — on the exact deal they’re working right now.

8 · Product & Marketing

Feature trends route to product, competitor positioning to marketing — each with enough context to act on.

What the AI pulls from every call

Competitor mentionsFeature requestsObjectionsBuying signalsRisk flags

Why eight workflows, not one

Same principle as microservices, applied to automation — single responsibility, clear interfaces, independent evolution.

Modularity

Gong changes its API? I update Workflow 1. The other seven never notice.

Debugging

In an 8-system integration, something always breaks. I isolate it to one workflow — not a hunt through hundreds of nodes.

Independent scaling

The AI-heavy workflows retry and scale on their own schedule, without touching the simple data-routing ones.

The AI is in the middle, not at the edges

Three workflows prep and constrain the AI; three more aggregate, human-check, and route what it produces.

The AI is powerful but unreliable. The rails make it reliable. I’ve swapped the model twice — the pipeline never changed. The rails are the durable investment.

Results that compound

Two quarters in — all from calls that were already happening. No new data collection, no new process, no behavior change.

Product

Three feature gaps that showed up in 20%+ of technical evals. Two hit the roadmap; the third was a docs problem — the feature existed, prospects couldn’t find it.

Marketing

Caught a competitor’s launch being positioned against us in a way the messaging didn’t answer. Battle cards updated within a week.

Sales

Pricing objections in enterprise tracked with deals where the champion wasn’t in commercial talks. Include the champion → win rates improved.

Leadership

A quarterly competitive-landscape briefing generated from the data — the kind that used to need a consulting engagement.

Build the rails. The intelligence follows.

Ready to build your last layer?

Stop paying for tools nobody uses. Let’s build the layer where your team already works — on infrastructure you own.

Book a 20-min callSee the method →