Case study 03 · AI Analytics Startup
AI-Assisted GTM Data Engine
Turning an ICP hypothesis into structured prospect data, CRM infrastructure, and outbound execution that could be tested and measured.
GTM Operations & Digital Marketing
HubSpot · AI Enrichment · CRM Data · Prospecting · Outbound Automation
The starting point
A hypothesis, not a database.
An early-stage B2B startup was validating who to sell to. Prospect data sat across spreadsheets and discovery tools. Records often lacked a work email or a consistent industry, enrichment could get expensive quickly, and information passed through several tools before it was usable for outbound.
The work sat where GTM strategy, data operations, AI-assisted enrichment, CRM structure, and outbound execution meet.
ICP → Discovery → Enrichment → AI classification → HubSpot → Outbound → Engagement
01 — Define the ICP
A targeting model you could actually query.
This was the working targeting framework for the engagement—a hypothesis to test, not proof that every matching company was an ideal customer.
02 — Build the prospect data pipeline
From scattered sources to CRM records.
- 01
ICP
Targeting framework
- 02
Prospect discovery
Clay · Google Maps · AI-assisted research
- 03
Data enrichment
Enrichment workflows
- 04
Classification / QA
AI-assisted, human-reviewed
- 05
HubSpot CRM
Structured prospect records
- 06
Outbound
Sequenced, controlled sending
- 07
Engagement data
Opens, replies, signals
03 — AI-assisted classification
AI did the sorting. People checked the work.
Inconsistent industry data made segmentation unreliable. I used prompt-driven workflows in Claude and ChatGPT to categorize 5,000+ raw contact records into 7 standardized industry groups.
This was AI-assisted data operations, not autonomous AI: classifications were reviewed before they were used.
Raw contact data
5,000+ records with inconsistent industry labels
AI classification
Prompt-driven categorization with Claude and ChatGPT
7 standardized industries
One consistent taxonomy for segmentation
Human QAHuman in the loop
Classifications reviewed before use
CRM-ready data
Segmentable records for HubSpot and outbound
04 — CRM & data quality
Auditing the wider dataset.
I audited a CRM dataset of approximately 9,000 records to identify segmentation, duplication, and data-quality issues.
- Inconsistent segmentation
- Duplicate records
- Personal email addresses
- Incomplete prospect information
05 — CRM integration
HubSpot as the destination.
I handled field mapping and sync planning so structured prospect data landed in HubSpot on consistent fields: Work Email, Industry, LinkedIn URL, Company Name, and Job Title.
Implementation status: a Freckle → HubSpot sync was still being developed and validated during the engagement, reaching roughly 75% functionality. It was not fully deployed.
06 — Outbound activation
Build, test, measure, learn.
Build
Sequences in Smartlead / HubSpot, plain-text messaging
Warm
Inbox warming and controlled sending volume
Test
Initial pilot to 81 target leads
Measure
Deliverability and engagement monitoring
Learn
Feed results back into targeting and messaging
81
Leads
37%
Open rate
0
Initial replies
I co-managed an initial cold email pilot to 81 target leads. It reached a 37% open rate through plain-text messaging and deliverability optimization. The initial test generated no replies—an early signal that targeting or messaging needed further iteration.
Hands-on implementation
What I actually built.
ICP Architecture
Target industries, company criteria, personas, and priority markets
Prospect Data Operations
Discovery, enrichment, standardization, and CRM preparation
AI-Assisted Classification
Prompt-driven categorization of 5,000+ contact records with human QA
CRM Infrastructure
HubSpot field mapping and structured prospect data
Outbound Operations
Smartlead / HubSpot sequencing, deliverability configuration, and pilot execution
Verified evidence
By the numbers.
5,000+
Contact records AI-categorized, with human QA
25
U.S. markets researched
7
Target industries standardized
37%
Open rate on the initial 81-lead outbound pilot
Next step
Turning an ICP into a working GTM motion?
If your prospect data is scattered and your outbound isn’t measurable yet, I’m happy to talk through where to start.