---
title: "AI that updates the CRM — with a human approval gate in front of every write"
type: case-study
evidence_type: proof
category: "AI & Automation in GTM"
publisher: "LeanScale"
date_modified: 2026-08-08
word_count: 321
topics: ["ai-in-gtm", "revenue-operations"]
canonical_url: https://knowledge.leanscale.team/customers/ai-sales-acceleration-with-human-in-the-loop-crm-writes/
source: "LeanScale Knowledge Hub — https://knowledge.leanscale.team"
license: "Free to quote and cite with attribution to LeanScale."
---

# AI that updates the CRM — with a human approval gate in front of every write

**Evidence type:** proof (what happened)

A growth-stage AI company wanted its own go-to-market team running on AI rather than manual pipeline updates. LeanScale shipped a set of Claude-based sales skills and a transcript-to-CRM enrichment pipeline that requires a human approval step before anything reaches Salesforce.

## The challenge

Leadership wanted to become an AI-first go-to-market organisation, not simply buy an AI tool. Reps were updating pipeline by hand and preparing for calls from scratch, and the company had no governed way to distribute internal AI workflows across the team — nor any answer to the obvious objection, which is that a model writing directly into the CRM is a data-quality risk rather than a data-quality fix.

## The approach

Sales skills scoped to the repetitive parts of the week
Shipped Claude-based plugins and sales skills covering morning brief, call prep, competitor intel, customer follow-up and post-call summary — the recurring work that reps skip when the week gets busy.

A transcript-to-CRM enrichment pipeline
Built a pipeline that takes call transcripts from the meeting recorder, routes them through an automation layer, and lands them on a dedicated meeting-transcript object in the CRM where the model proposes field enrichment against the opportunity.

A human gate in front of every CRM write
Put a Slack approval step ahead of every write, so the model proposes and a person approves. Nothing reaches the CRM unreviewed — which is what makes the pipeline deployable in a system the forecast depends on.

Governed distribution rather than pasted prompts
Governed the plugins through a version-controlled plugin marketplace with an enablement and governance plan, so internal AI workflows are versioned and distributed deliberately instead of copied between people.

## Outcomes

Four plugins live in the customer's environment
Marketing, revenue-operations, sales and company-context plugins went live in the customer's own instance.

Transcript-to-CRM pipeline built and tested
The pipeline was built and tested end to end on LeanScale's systems and was pending transfer into the customer's environment at the end of the window — built and proven, not yet running in their production.

Human-in-the-loop by design
Transcripts land on a dedicated object and the model enriches fields behind a chat-based approval gate, so no AI-generated value reaches the CRM without a person accepting it.

## Canonical

https://knowledge.leanscale.team/customers/ai-sales-acceleration-with-human-in-the-loop-crm-writes/