Case files — anonymized, logged, still running
Names are confidential; log files aren't. Every system below was designed and delivered by the consultant behind Thousand Cuts — some in-house, some on consulting engagements — and still runs today.
Automated brand and AP style review for the marketing team at a national eye-care group.
Creative intake for the creative ops team at a national health system.
Workfront and Azure DevOps kept in step for the marketing team at a top-5 U.S. credit union.
FIG. 1 — The case cards. Logs are representative reconstructions of each system's runs; the metrics are measured.
The marketing team's copy documents landed in monday.com and waited — sometimes days — for a first review round that was largely mechanical: brand terms checked against the guide, AP style enforced, risk language caught, internal jargon stripped before a customer could ever read it. The reviewers' judgment wasn't the bottleneck. The queue in front of their judgment was.
Our principal built the round-one reviewer, on a consulting engagement. Make watches monday.com; the moment a new copy document arrives, Claude reads it against the brand guide and AP style, flags risk language, and marks the internal jargon that shouldn't leave the building. The findings are back on the document within the minute — before a human has opened the file.
The humans still review. They just start from marked-up copy instead of raw copy, and round one stopped costing days of calendar for minutes of work.
One creative request rarely meant one project. A request with three assets meant a producer building three projects by hand — reading the form, picking the template for each asset, keying the same fields three times, and working out who should design and who should write. Multi-asset requests multiplied the re-keying, and the request sat in the queue while they did.
Our principal built the intake, on a consulting engagement. Workfront Fusion watches for new creative requests and reads the request's own metadata: one project per asset, each on the template that metadata calls for, designers and copywriters assigned from the same fields. Every action the automation takes writes a line to an error-logging project in Workfront — a standing audit trail — and anything that fails critically pages the team in Slack instead of failing silently.
A request now becomes staffed, templated projects in about a minute, and when something does go wrong, it goes wrong loudly, on the record, in front of the people who can fix it.
The marketing team planned and tracked its work in Workfront. The people delivering that work tracked the same work in Azure DevOps. So every request existed twice, and staying in step was somebody's job: a comment written in one system, re-typed into the other; a status changed here, remembered there; a due date that was accurate in one tool and stale in the other. The team measured the toll at roughly 3,600 staff hours a year, all of it spent re-typing what one of the two systems already knew.
Our principal built the connection — in-house, as staff. A work request entered in Workfront creates the project, and the matching Azure DevOps work items are created and linked to it. From that point the link is live in both directions — a comment, a status change, or a metadata edit made in either system appears in the other, without a person carrying it across.
Neither team switched tools, and neither team gave up the view it works in. They stopped keeping two copies of the same project by hand. The full architecture — service hooks, loop prevention, the tag-guard — is on record in the Workfront Azure DevOps integration walkthrough.
The Workfront team at a major retailer ran an instance of 500 users and 300+ active projects — and the questions an admin team actually asks (who has adopted what, which projects are healthy, where the platform is drifting) were the ones Workfront's native reporting drew worst. The data existed. The picture didn't.
The instance's data already landed in Snowflake, so our principal built the picture there — an in-house build: a Claude-generated audit dashboard over the warehouse — user adoption, project health, platform drift — refreshed on a schedule, with the charts native reporting can't produce.
The admins stopped assembling the view and started reading it: 500 users and 300+ projects on one screen, current on schedule, nothing exported by hand.