AI content that clears review on its own, 92% of the time
Context
A multi-state hospital network serving 1.4M+ portal users, with marketing running across four markets (Georgia, Kentucky, Tennessee, Texas): copy, imagery, and policy updates for each one.
Problem
They wanted AI in their marketing. They had no way to govern it.
The content team was ready to build pages and copy with AI, but nothing defined what AI could publish on its own, who handled what it got wrong, or how a decision could be traced later. For healthcare content in four states, with patient privacy, clinical claims, and state rules in play, that gap was the risk.
Approach
Build the team first, then the model, then put it inside the tools people already used.
What I did
Identified the governance team with the marketing lead, drafted the governance model, and took it through approval. Consulted on the design of the three agents and weighed in on each round of tuning, working with the internal build team so governance decisions fed straight back into the agents.
Ran the rollout and the biweekly sync. Content that passed the Risk Agent published on its own, with the decision logged. Failures and Intake Agent tickets landed in one JSM queue, auto-assigned to the right staff, with first contact inside 1 business day and a decision inside 3. The Legal Agent started by reviewing every item to build trust, then narrowed to failures only.
Outcome
Over 3 months of tuning, the share of content clearing risk review on its own rose from 38% to 92%, and legal review narrowed from every item to failures only. The intake queue was trending down and starting to level off when the contract ended, with the model, the queue, and the sync in place for the client’s team to run.
What I would do differently
Training and enablement came in a month before rollout. They should have had a seat at the first meeting, so the people working the queue helped shape the policy instead of learning it at the end.