1,689 messages, 11 needed her
A small e-commerce company selling digital goods was buried in support mail. David Gewirtz describes building a mail system for his wife's business: Claude built a rulebook from an export of 88,777 past messages. A scheduled ChatGPT run now reads the rulebook each day, sorts and files new mail, and drafts template replies. An hourly script sends drafts that qualify, then labels and archives the originals. Urgent items go to the owner.
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 small e-commerce company selling digital goods was buried in support mail. David Gewirtz describes building a mail system for his wife's business: Claude built a rulebook from an export of 88,777 past messages. A scheduled ChatGPT run now reads the rulebook each day, sorts and files new mail, and drafts template replies. An hourly script sends drafts that qualify, then labels and archives the originals. Urgent items go to the owner.
In a tested backlog of 1,689 messages, only 25 were from real customers, and only 11 needed her response. Two earlier builds failed over about four weeks: their tests passed but never tested real unattended processing. The working system needed 15 rulebook revisions in its first two weeks. Cost is her $20 a month ChatGPT Plus plan, plus the builder's own Claude plan used to build it.
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.