[Provider Grain]
Dispatch Match Rate & a Silent Outage
SMS and e-mail match rate measured at the provider's grain — and the audit that found a channel off while the dashboard stayed green.
The latency dashboard reports how many leads reach each dispatch platform. The SMS team compared it with their platform and the numbers didn't match.
Audited the metric, changed its unit, and investigated the incident end to end.
The dashboard counted events; the SMS platform counts people. And a drop in match rate with zero errors looked like a bug.
01 · Measure both units
About 1.1k events were about 850 people for one product × day — the unit was the gap. A global Events ⇄ Leads toggle followed.
02 · The provider's grain
Person (phone) × list × New York day for abandonment and cross-sell, with the cart attributed to the first event.
03 · Three readings, one decision
Three definitions of “finalized” were measured side by side over full history; the business chose the historical one and it was restored to the decimal.
04 · Follow the zero
Purchase-confirmation messages dropped to exactly 0% delivered. Not a failure: one provider had been switched off on purpose, and the parallel platform had no flow for those events.
! brokeExcluding deduplicated messages from “finalized” used LOCATE on a column that was NULL for months of history, collapsing old months to about 1%.
→ changedReverted to the historical definition and tested every candidate over full history before choosing.