The agency side plans and publishes a month of social content for every client on its books. Research, drafting, briefing, approval and scheduling ran across a design tool, a scheduling tool and a chat thread, and the approval that actually mattered lived in somebody's inbox.
A back office the team works in as one document. It reviews last month against the platforms' own numbers, researches what is coming, drafts to each client's contracted package, scores the drafts against the brand, and compiles a monthly brief. Once a person approves that brief, it builds the calendar and publishes. Clients open their own calendar from a link with no login, comment on individual posts, and get their performance report as a PDF.
Publishing runs through the client's own accounts, and the step that puts a post in public is a coded gate rather than an instruction to an assistant. It refuses without a human tick, it refuses a second time for a platform already published to, and when something fails it records the error instead of marking the post as sent.
Measured from the system's own run records on those dates, including one full monthly cycle carried end to end for the following month, and a real approval given from a phone. The publishing gate is asserted by automated tests that each check something does not go out.
This page once carried claims that described the work inaccurately. They came down in an audit of our own copy, and each one went back up only when its numbers could be traced to something measured. A case study you cannot check is worth nothing to you, and one that turns out to be wrong costs us more than it ever earned.