Thirty-eight signals tracked. Several turned out to predict nothing at all.
Signals were chosen on intuition and never audited. A funding-based trigger had been in use for two years on the assumption that it worked.
Reporting broke reply rate down per signal. The top signal returned 18.4%; the funding trigger returned 1.4% and was retired with the evidence attached.
Targeting now moves on measured performance. Three hundred and forty accounts were re-ranked after two weeks of reply data, without anyone editing a filter.
Thirty-eight signals tracked. Several turned out to predict nothing at all.
{{ sec.body }}