Skip to content
Case study · In production (our own)

WeBuild's outbound agent — our own dog food

The cold-outbound research pipeline WeBuild runs on itself: an agent that researches prospects, enriches them, and drafts personalised sequences into Smartlead. If we won't run our own agents, why would you?

OutboundSmartleadResearchDogfood

The problem

Cold outbound done properly is research-heavy: who is this company, what do they run on, what would an agent actually fix for them? Done by hand, that's 15–20 minutes per prospect before a single line gets written — which is exactly the kind of repeatable, judgment-light work we tell clients to hand to an agent.

What it runs

The agent takes a prospect list, researches each company (site, footprint, tooling signals), enriches the record, scores fit against WeBuild's actual offer, and drafts a personalised opening sequence — which feeds Smartlead for sending. Drafts, not blasts: sequences are reviewed before a campaign goes live, and anything the agent can't ground in research it flags instead of inventing.

What we learned building it

The draft-approve loop isn't a compliance tax — it's where the agent gets good. Every edited draft is a training signal for the prompt and the playbook. And grounding rules matter more than model choice: the single biggest quality jump came from forbidding claims the research couldn't support — the same rule we apply to this website.

Where this stands — honestly

This one is fully in production — for WeBuild itself. It's shown as proof of craft, not as a client result. Ask to see it running on the intro call; a live walkthrough beats a screenshot.

Want to see it run?

Ask on the call — a live walkthrough of a production agent, not a slide about one.

Book an intro call →

30 minutes · no pitch · if AI isn't the answer, I'll say so