Agentic prospecting system
How ChatWalrus doubled its LinkedIn network without touching it
An AI training company selling to brands like HexClad and Warby Parker had hit the ceiling of manual prospecting. We built an agentic system that finds, filters and engages the right decision-makers — without ever putting the account at risk.
- 2×network growth in month one
- 1000sof leads processed weekly
- 0LinkedIn violations
About the client
ChatWalrus helps retail, CPG and food & beverage brands — from $10 million to $5 billion in revenue — build genuine AI fluency across their teams. It is trusted by brands including HexClad, Ridge, Bombas and Warby Parker.
Founder Craig Foldes built the company on a conviction that AI confusion inside enterprises is an operational problem with a practical solution. Capturing fast-growing demand meant reaching the right decision-makers before competing vendors did.
The challenge
- Network capped at ~1,500 connections
- Only 50–100 qualified prospects a week
- Hours spent sourcing instead of closing
- Direct automation risked the account itself
As enterprise demand for AI training accelerated, manual LinkedIn prospecting hit a hard ceiling: roughly 1,500 connections and 50–100 qualified prospects a week — far below what the pipeline required.
Finding engagers on relevant posts, filtering for founders and C-suite, checking prior contact and routing prospects into outreach consumed time that should have gone to closing.
The obvious fix — automating LinkedIn directly — risked restriction or suspension of the account the business depended on, ruling out most off-the-shelf tools.
The system we built
How it works.
Agentic prospecting: what comes in, what the AI does with it, where people stay in control and what goes out.
Comes in
- LinkedIn post URLs
- Craig’s personal posts
- Prospect replies
Agentic prospecting
- Isolated extraction (off-account)
- Founder & C-suite filtering
- Global de-duplication
- Prior-conversation check
- AI intent classifier
Human in the loopHuman approval before any message from Craig’s personal presence
Goes out
- Qualified prospects into outreach
- Booking link sent in seconds
- Live campaign dashboard
Step by step
What we built, in order.
Shipped in 3 weeks, with weekly demos on real data along the way.
- 01
Extract without risk
Given a post URL, extraction runs on isolated infrastructure — completely separate from ChatWalrus’s account — capturing every engager without leaving a footprint.
- 02
Keep only decision-makers
Prospects are filtered to founders and C-suite, then checked against a historical registry and prior conversations so nobody is contacted twice.
- 03
Protect the founder’s voice
For Craig’s personal posts, stricter filters, live connection checks and a human approval step run before any message is sent.
- 04
Turn replies into meetings
An AI intent classifier reads each reply and sends a booking link within seconds when interest is genuine.
- 05
See everything live
A custom dashboard shows every campaign, lead status and approval queue in real time.
Results
What changed.
- 2×network growth in the first month — from 1,500 to 3,000+ connections with no manual prospecting
- 1000sof leads processed weekly, up from 50–100 by hand
- 0%duplicate contact rate, protecting sender credibility
- Secondsfrom interested reply to booking link, down from hours
- 0LinkedIn violations — all extraction kept entirely off the account
What this means for you
Growth channels have limits when people run them by hand. Agents remove the ceiling — and good engineering removes the risk.
Built with
- Multi-agent orchestration
- Isolated data extraction
- LLM intent classifier
- Outreach platform
- Next.js dashboard
“Awais is a wonderful partner! He is a true entrepreneur, who’s hungry to over-deliver & add incredible value. He cares deeply about ensuring the work output meets your expectations & is done fast. He asks all qualifying questions to ensure he & his team have everything they need & I enjoy working with him immensely!”
Craig FoldesCEO & Founder, ChatWalrus
