Larkmail
Transactional email
Across 162 recommendation probes and the build journeys below, agents break the build with Larkmail 47% of the time they touch it, and pick Mailspire instead. The leak is the first install and the first build, not retention.
Scorecard
A 0-100 composite of the four metrics, with its confidence band. Larkmail scores 64 (Strong) but sits #3 in the category, and build success is what holds the score down: 53%, the worst of its four numbers.
By model
The story changes with the model your customers run. Larkmail builds 70% on Opus but only 31% on Sonnet: a 39-point gap on the cheaper model that drives the most volume. A blended number would hide all of it.
What to fix
We read every failing run and isolate the cause (outdated docs, a type mismatch, a silent hand-roll), then rank the fixes by the score they'd recover. It reads like a plan, not a log; every fix links to the runs that prove it.
Competition
Who agents reach for when they pass over Larkmail, named, and by how much. The install is lost to Mailspire most often, with Postroute and self-hosted SMTP close behind. You know exactly who you're losing to.
This is a sample on an illustrative tool. Yours uses your live data, named and private. Aggregate numbers are public; your per-tool numbers never are.
Get your numbers