70% of tickets, closed with drafts
A Swedish brand sells consumer hardware direct to customers on Shopify. A small support team handles every conversation in Zendesk. Each ticket now runs a fixed pipeline. The system reads the live Shopify order and shipment data. It reads the written policy for that request type. It drafts a reply from those facts only. A separate checker pass tests every claim in the draft against the sources. A failed draft goes back for a bounded retry. A person approves the final draft in Zendesk before it sends. Nothing reaches a customer without that approval.
Start with one job only. Fill in the three lines. Keep the bot on that job.
The one job this bot does: ______________________________
What it may do on its own: ______________________________
What it must ask me first: ______________________________
Two lines from this case, side by side. No new figures.
A Swedish brand sells consumer hardware direct to customers on Shopify. A small support team handles every conversation in Zendesk. Each ticket now runs a fixed pipeline. The system reads the live Shopify order and shipment data. It reads the written policy for that request type. It drafts a reply from those facts only. A separate checker pass tests every claim in the draft against the sources. A failed draft goes back for a bounded retry. A person approves the final draft in Zendesk before it sends. Nothing reaches a customer without that approval.
In the pilot, 70% of incoming tickets closed with the pipeline draft as the reply, approved with or without light edits. The AI cost averaged about €0.30 per ticket across all model calls. The cost per resolved ticket was about €0.43. The other 30% stayed human work: complaints, disputes, request types the policy does not cover, missing data, or drafts the checker would not pass. The brand is not named, and the source calls the pilot a draft-only test, not a controlled trial. The source says the written support policy mattered more than the prompt: where the policy was vague, the drafts were vague.
The after line is the source's claim, not our measurement.
smalloffice.space · Get a case like this in your inbox each morning. Subscribe free
Read this case on your phone now, and keep it for later. Type your email. We send this case to that address, and you get the next cases each morning too.