Automated Payment Reconciliation Across 12 Hospital Facilities
Payment discrepancies only surfaced when someone reconciled by hand, often too late to fix. We automated reconciliation for twelve facilities on their own cadences, detecting six categories of discrepancy with the evidence attached.
- Service
- Business Transformation
- Industry
- Revenue Cycle & Finance
- Region
- United States
- Engagement
- SES (specialist engineers)
Outcomes
- 12 facilities reconciled automatically
- 6 discrepancy types with evidence
- Findings routed to owners
Stack
Python · Power Automate Desktop · SQL · Orchestration
Written byOriental Nippon Team
A US revenue-cycle operator collects on behalf of hospital facilities, and its payment records have to agree with the hospitals' own. When they don't, money is either not collected, not credited, or credited twice. None of those situations announces itself.
Before this work, differences surfaced only when someone sat down and reconciled by hand. Twelve facilities each had their own cadence, so the checks ran late, and the window to fix a problem was often already closing.
The problem
- Found late, if at all. Discrepancies appeared only during manual reconciliation.
- Twelve different rhythms. Each facility had its own schedule and its own data shape.
- Shrinking resolution windows. By the time a mismatch was found, there was less time to act on it.
- Unmeasured leakage. The revenue effect was real but nobody was counting what was being missed.
| Before | After | |
|---|---|---|
| Detection | Manual reconciliation | Scheduled, per facility |
| Discrepancy types | Informal | 6 agreed categories |
| Facilities covered | Whoever had time | All 12, on their own cadence |
| Findings | An exception count | Each discrepancy with the records behind it |
What we built
Per-facility reconciliation flows
Twelve automated flows in Python and Power Automate Desktop, each running on the cadence that facility needs.
Six discrepancy categories
Detected automatically and raised with the records that caused them, not as a bare number.
Routed findings
Each finding goes on a schedule to the people responsible for resolving it.
A repeatable pattern
New facilities are added to the same design as they onboard.
Every discrepancy now follows the same path, from detection to a recorded decision:
Technology and the role it played
| Technology | Role in this project |
|---|---|
| Python | Matching logic and discrepancy classification |
| Power Automate Desktop | Steps that depend on desktop-only systems |
| SQL | Querying both sides of each reconciliation |
| Scheduled orchestration | Running each facility on its own cadence |
How we delivered it
The first two weeks produced no code at all. Finance and operations had to agree on what counts as a discrepancy, because every later decision depended on that definition.
- Define discrepancy. Six categories agreed with finance and operations.
- Prove the pattern. Flows built for two or three facilities first.
- Roll out. Remaining facilities added with their own cadence and data shape.
- Run silently. Automated findings compared with manual reconciliation, with nothing sent to anyone yet.
- Switch on distribution. Only once the people receiving findings accepted the false-positive rate.
01Discrepancy definition
2 wks
Six categories agreed with finance and operations
02First facility flows
3–4 wks
Pattern established on two or three facilities
03Roll-out to remaining facilities
4–6 wks
Per-facility cadence and data shape
04Silent run and tuning
2–3 wks
Automated findings compared against manual reconciliation
05Distribution and hand-over
1–2 wks
Stakeholder routing and runbook
Risks and how we handled them
| Risk | How we handled it |
|---|---|
| Alert fatigue from false positives | A silent run compared findings with manual reconciliation before anyone received them |
| Each facility's data looks different | Per-facility flows built on one shared pattern |
| Disagreement about what counts as a discrepancy | Six categories agreed with finance and operations before any code |
| Findings with no owner | Every finding is routed to the people responsible for resolving it |
Outcome
Discrepancies now appear on a regular schedule and reach the people who can act on them. That reduces revenue leakage and removes hours of manual reconciliation every cycle.
How much leakage was actually recovered is the client's number to publish, not ours. We built the detection. They hold the ledger.
What the client keeps
- The reconciliation flows and their source
- The six discrepancy definitions, documented
- Per-facility configuration
- Distribution routing
Sources and further reading
- Introduction to desktop flows, Microsoft Learn
- Administrative simplification, Centers for Medicare & Medicaid Services
- The ACH Network, Nacha
Facing something similar? This is the kind of work our Enterprise & Digital Transformation team delivers every day. Talk to us about your stack.
