Customer story · anonymized

A youth sports
e-commerce operator's
back office, run on
three workflows

Anonymized. Real, live, measured. The same workflows we offer clients are the ones we run for ourselves first. Numbers below are as of the last reporting cycle.

Drafts processed
1,942
Adoption rate
59%
Rewrite rate
35%
Live agents
3
Articles ingested
141 KB

The setup

The customer is a mid-market e-commerce operator in youth sports. Their team processes hundreds of daily support tickets through Front, reconciles deposits across multiple sales channels, and publishes scheduled finance reports into Slack. Their stack: Slack, Front, ShipStation, PlanetScale, Contentful.

01
Customer ops

Customer service drafting

Drafts replies in Front, grounded in 141 KB of ingested help articles. Human reviewer approves or edits; edits become few-shot examples for the next draft.

02
Finance ops

Daily deposit reporting + user lookups

Logs into the banking dashboard, pulls yesterday's deposits, generates a CSV, uploads to storage, posts a Slack summary by 9am. Handles ad-hoc user lookups from the database with a browser fallback for apps without APIs.

03
Orchestration

Routing + supervision

Supervisor pattern: routes work between the CS and finance specialists and the human team. Owns escalation policy, handoff context, and audit logging so nothing falls through the cracks.

The numbers (live)

Selina's draft adoption rate
Adoption rate, 59% 59% approved · 41% rewrote

1,942 drafts processed to date. 59% approved without rewrite — the human reviewer hit send as-is. 35% edited for tone, missing context, or policy detail. The rewrite rate trends down week over week as the workflow absorbs each edit.

What humans edited
Edit category breakdown
  • Tone and phrasing 42%
  • Missing context 28%
  • Policy detail 18%
  • Other 12%

When the team edits, they most often touch tone and phrasing. Those edits are the highest-leverage training signal. Every edit becomes a few-shot example for the next draft.

Why it works

The edit loop is the moat. Most AI customer-service tools try to go 100% autonomous on day one, fail, and quietly get turned off. These workflows start in shadow mode, graduate to drafting, and earn the right to approve their own responses as the rewrite rate drops. The team never has to trust a black box. They just stop editing.

The 59% number is what progress looks like in real production, not a demo.

Hire

Want numbers like this for your ops?

Join the waitlist See how it works
Preview build. Not the live site. (env: production)