The problem this solves
Every lead-gen advertiser eventually meets the quality death spiral: the campaign optimizes toward form fills, the cheapest fills come from the least qualified people, the algorithm learns to find more of them, and each week's leads are a little worse than the last. Budget rises, quality falls, and the standard remedies - tighter targeting, longer forms - fight the optimizer instead of teaching it.
The teaching mechanism exists and is criminally underused: send a conversion event only when a lead is actually good, and the algorithm reverse-engineers what good looks like. This is the qualified-lead pattern the Conversions API was designed for. Almost nobody runs it, because it requires scoring every lead promptly and feeding the result back on a cadence - exactly the kind of diligent loop that human teams start and abandon.
This mission is the loop, running on a schedule. Each week it pulls the new leads, scores them against your ICP with the reasoning shown, and stages conversion events for only the qualified ones - your approval on every batch. LinkedIn stops hearing "we want form fills" and starts hearing "we want people like these."
How the mission runs
- Pull the week's leads. Every Lead Gen Form submission in the window arrives with its form answers and source campaign - the raw material for scoring, fresh enough that the feedback reaches the algorithm while its choices are still recent.
- Score against the ICP. Each lead's company is enriched and the lead is scored: industry, size, seniority, and what their answers signal about intent. Every score carries its reasoning, and the threshold for "qualified" is yours to set and adjust.
- Stage events for the qualified only. Qualified leads become conversion events against your qualified-lead rule - emails hashed at staging, timestamps set to the qualification moment. The unqualified simply produce no event, which is itself the signal.
- Approve the batch. The week's batch parks in the Approvals queue with the scored list attached, so the approval is informed: you see who qualified and why before anything streams.
- Watch the mix shift. Acceptance is reported per batch, and the weekly quality mix becomes the metric to watch: as the algorithm learns, the share of qualified leads per hundred fills should climb - the death spiral running in reverse.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A standing quality feedback loop: weekly scored leads with reasoning, conversion events staged for the qualified only, your approval on every batch, and the acceptance report - plus the week-over-week quality mix that shows the algorithm learning what your buyer looks like.
Make it yours
- Start with a strict threshold and loosen deliberately; teaching with your top quartile first gives the optimizer the sharpest picture.
- Add deal values to events for qualified leads that later convert, layering revenue weight onto the quality signal.
- Pair with the form-and-campaign teardown so targeting changes and algorithmic learning pull in the same direction.
Frequently asked questions
Why not just send every lead as a conversion?
Because that reproduces the form-fill signal with extra steps. The entire mechanism is selectivity: the algorithm learns from the difference between leads that produced an event and leads that did not. Feeding it everything teaches it nothing.
How fast does the optimization improve?
LinkedIn needs volume to learn - expect visible movement over weeks of consistent batches rather than days. Consistency matters more than size: a steady weekly signal beats an occasional large one, which is exactly why this runs as a schedule.
What if our ICP definition is wrong?
Then the scoring reasoning will look wrong in the approval review, which is the point of keeping a human on the loop. Adjust the ICP in workspace memory, rerun, and the next batch teaches the corrected lesson - the loop is steerable at every cycle.