AI automation
Sample the quality, then commit to the scrape
Our cold-outreach system: scrape a plain-English lead query into a validated list, then personalize it with a researched icebreaker line, one module at a time.
The problem
Cold outreach lives or dies on two things: whether the list is actually right, and whether the first line reads like it was written for that one person. Both usually cost either a lot of manual research time or a subscription to a black-box tool.
A scraped list also burns through a limited daily quota fast if the query wasn't specific enough, with no cheap way to sample quality before committing to a full pull.
What we built
The pipeline runs on two layers: a discovery layer that samples 25 leads first so the filters can be judged before anything expensive runs, and a production script that commits the full scrape straight to disk without routing thousands of rows through the AI's context window.
A second stage adds a personalized icebreaker to each lead: it researches the lead's website for a specific line where it can, about half the time on the first real run, and falls back to a rotating template otherwise, at zero cost by default.
The result
Sample the quality, then commit to the scrape
25
lead sample checked before a full scrape commits
~51%
of leads got a genuinely specific icebreaker line
$0
cost per icebreaker, by default