Ben Walker

Cold Calling AI Agent for Property Leads

May 2026 - Present • 4 mos

A boutique real estate investment firm had a list of thousands of off-market property leads sitting untouched because their small team could not call through them fast enough. I built an outbound AI cold calling agent using Retell AI, scripted to open with a soft, conversational tone rather than a hard sales pitch, qualify sellers on timeline and asking price, and log every answer straight into their CRM. The challenge was making the agent sound natural on a topic people are often guarded about, so I wrote and tested several script variations before landing on one that kept people on the line. Calls that qualify as a genuine lead are tagged hot and routed to a human closer within minutes through an n8n workflow, while cold leads are logged and scheduled for a later follow up call automatically. The firm went from manually dialing a few dozen leads a day to having their entire list worked through consistently every week.

CRM and Cart Recovery Automation for DTC Brand

May 2026 - Present • 4 mos

A direct-to-consumer skincare brand was manually exporting Shopify orders and abandoned carts into spreadsheets to follow up with customers, which meant most abandoned carts were never actually chased. I connected their Shopify store to HubSpot using Zapier and n8n, building automated flows that log every order and abandoned cart in real time, tag customers by purchase behavior, and trigger a three-step SMS and email follow up sequence through Twilio for anyone who leaves items in their cart. The main challenge was avoiding duplicate contacts across Shopify and HubSpot as customers reordered, which I solved by matching on email and phone number before any new contact record was created. The brand now recovers a meaningful share of abandoned carts automatically and has a clean, de-duplicated customer database they can actually use for future campaigns instead of a spreadsheet nobody trusted.

Multi Agent Support Workflow for SaaS Company

May 2026 - Present • 4 mos

A small project management SaaS company had one support inbox handling bug reports, billing questions, and feature requests, with no way to tell them apart without a human reading every single email first. I built a multi-agent workflow in n8n using OpenAI to classify each incoming ticket, route billing questions to their HubSpot pipeline, flag bugs into their engineering tracker, and draft a first response for straightforward questions that a human just approves and sends. The hardest part was getting classification accurate enough to trust, so I built a confidence threshold that routes anything ambiguous to a human queue instead of guessing. Support response time dropped noticeably because routine questions get an instant draft reply, and the team stopped losing bug reports in a crowded inbox since every one is now automatically tracked.