← All rollout articles

Healthcare system: back-office automation with human oversight

Problem

Revenue cycle teams juggled dozens of desktop bots that broke when payer portals changed skin deep. Work backed up silently until someone noticed a red folder on a supervisor’s desktop.

Constraints

HIPAA logging, minimum necessary access, and union scheduling rules for after-hours coverage. Clinicians could not be pulled into IT theater; fixes had to reduce touches, not add screens.

Approach

We modeled work as durable workflows with explicit states: intake, validation, human decision, and posting. Bots became adapters only where no API existed, bounded by contracts and synthetic monitoring that mimicked real tasks.

Rollout

We sunset bots workflow-by-workflow, never “big bang.” Each workflow shipped with an exception queue, SLAs, and a single owner role. Shadow mode ran until false-positive and false-negative rates met thresholds agreed with operations leadership.

Risks mitigated

• Silent failures: end-to-end tracing from portal change to posting
• Audit gaps: immutable audit trail per decision with retention policy
• Moral hazard (“the bot will fix it”): capacity planning tied to human queue depth

Outcomes (illustrative)

Mean time to detect a broken integration dropped from days to minutes. Staff overtime associated with catch-up batches fell materially within two quarters after the first two workflows stabilized.

Lessons

The organization didn’t need more automation—it needed observable automation. Queues turned political debates into numbers everyone could agree on.