---
title: AI Deal Enrichment for HubSpot | Case Study | Ramach Ops
description: How reps describe a deal in plain English and 30+ HubSpot fields fill themselves in, with every change to an existing deal approved by a person first.
---

[![Ramach Ops](https://ramachops.com/hs-fs/hubfs/Ramach%20Ops%20Logo%20(2000%20x%20500%20px)%20(2).png?width=170&height=43&name=Ramach%20Ops%20Logo%20(2000%20x%20500%20px)%20(2).png)](https://ramachops.com/?hsLang=en)

[Services](https://ramachops.com/services?hsLang=en)

[HubSpot Systems Audit](https://ramachops.com/services/hubspot-systems-audit?hsLang=en) [Fractional RevOps](https://ramachops.com/services/fractional-revops?hsLang=en) [Software Integrations](https://ramachops.com/services/software-integrations?hsLang=en) [Custom Tools & AI Automation](https://ramachops.com/services/custom-tools-ai-automation?hsLang=en)

[My Work](https://ramachops.com/my-work?hsLang=en) [About](https://ramachops.com/about?hsLang=en) [Schedule a Consultation](https://ramachops.com/schedule-a-consultation?hsLang=en)

Case study · Custom Tools & AI Automation

# AI Deal Enrichment for HubSpot

Reps describe a deal in plain English and thirty-plus structured HubSpot properties fill themselves in. 

HubSpot UI extensionServerless functionsn8nClaudeStructured outputs

The problem

Keeping deals updated meant clicking through dozens of fields and dropdowns, so records drifted away from reality.

What I built

An AI tool that fills in the properties based on a natural language input.

The result

A paragraph in, 30+ structured fields out, with a person approving every change to ensure it gets it right.

The problem

## Deal records were too much work to keep up

Deal records only work if they're up to date. Keeping them that way meant reps working through dozens of properties and dropdowns when they created a deal, and again every time something changed. Most of the time, they didn't. The CRM drifted from reality, along with every report based on it.

What I built

## An AI tool that keeps the data updated

A rep types (or pastes) a deal description in plain English in a sidebar card on the deal. The card returns a field-by-field diff showing the current value, the proposed value, and whether each change sets, adds, removes, or clears a value. Nothing is written until the rep approves.

[View the full walkthrough (PDF)](https://ramachops.com/hubfs/AI%20Deal%20Enrichment%20for%20HubSpot%20-%20Jason%20Ramach.pdf?hsLang=en)

Design decisions

## Built so the team can trust what the AI writes

### Constrained to the real schema

The model can only use fields and dropdown options that exist on the deal. Anything it can't confidently map is surfaced as unmatched, not guessed.

### Reasons against the current deal

It reads the deal's current state, so relative edits like “add 25 seats” work correctly.

### Writes exactly what was approved

Approval writes the exact change set the rep saw. The model isn't consulted again at write time.

### Proportional change notes

A small edit gets a one-line note. A bigger change gets a fuller summary.

### One source of truth for the schema

A single field-mapping file drives the response schema and prompts. A build script generates the workflow code, so nothing drifts out of sync from hand edits.

Why it matters

## Built to fit how your team already works

### Works right inside HubSpot

Reps use it directly on the deal record, so there's no new tool to learn, no extra login, and nothing to switch between.

### Easy to see and adjust

Every step of the process is laid out visually, so it's easy to understand, troubleshoot, and change as your sales process evolves, with nothing extra for your team to maintain.

### Clean, reliable data

The AI fills in fields in exactly the format HubSpot expects, so records stay consistent and your reports stay accurate.

The result

## Complete records without the clicking

Reps describe deals the way they'd explain them to their manager. Reps save time, records stay updated, and leadership can better trust deal reporting. 

[More about Custom Tools & AI Automation](https://ramachops.com/services/custom-tools-ai-automation?hsLang=en)

## Want something like this for your team?

Tell me where your reps lose time to data entry. I'll tell you how I'd take it off their plate.

[Schedule a Consultation](https://ramachops.com/schedule-a-consultation?hsLang=en) [More work](https://ramachops.com/my-work?hsLang=en)

Ramach Ops

RevOps and GTM systems for B2B teams running on HubSpot. Based in Denver, Colorado.

[Schedule a Consultation](https://ramachops.com/schedule-a-consultation?hsLang=en)

Services

[HubSpot Systems Audit](https://ramachops.com/services/hubspot-systems-audit?hsLang=en) [Fractional RevOps](https://ramachops.com/services/fractional-revops?hsLang=en) [Software Integrations](https://ramachops.com/services/software-integrations?hsLang=en) [Custom Tools & AI Automation](https://ramachops.com/services/custom-tools-ai-automation?hsLang=en)

Explore

[My Work](https://ramachops.com/my-work?hsLang=en) [About](https://ramachops.com/about?hsLang=en) [Contact](https://ramachops.com/schedule-a-consultation?hsLang=en)

Connect

[LinkedIn](https://www.linkedin.com/in/jason-ramach/)

© 2026 Ramach Ops · [Privacy Policy](https://ramachops.com/privacy-policy?hsLang=en)