What it is
Anthropic, publishing a customer story about Zapier, the workflow-automation company. Both sell the products the story features.
The story describes company-wide use of Claude across departments, with three concrete workflows. Marketing has the assistant draft content, save it to Google Docs and notify the team in Slack for review. Engineering triggers code work from an emoji reaction in Slack. A product manager researches event attendees from the CRM and the web before meetings.
The reported figures are adoption figures: 89% of employees using AI, more than 800 internal automations, and tenfold growth in usage year over year. There are no time-saved or revenue numbers, no dates on the page, and no description of what the review step checks.
What marketing teams should take from it
- Review happens where the team already works. The draft arrives in the document tool and the request in the chat tool, so nobody has to go looking.
- That is what makes it practical to keep the correction with the source or instruction it belongs to, the point finding 03.4 makes, instead of leaving it inside the final file.
- The pre-meeting research pattern, CRM record plus web search per attendee, is the same shape as the account brief in finding 01.3.
- “Adoption” and “agent” are not defined, so the 89% is as much a count of logins as of work.
Where to start
Copy the loop, not the numbers. For one content type, have drafts land in the document tool your team already uses and the review request in the channel they already watch, with a named reviewer. Record each correction against the claim or instruction that caused it.
For events, have the assistant build a one-page brief per attendee from the CRM record and public sources the morning of the meeting, and ask the account owner whether it helped the conversation.
Keep in mind
A vendor’s account of one customer, undated, with no measured outcomes.