Your Payment Integrity Program Keeps Less Than Half of What It Reports. The Government Just Told You So.
There is exactly one payment integrity program whose economics are fully public, end to end: Medicare's own Recovery Audit program. In FY2022, it produced $506.7 million in corrections — and returned $220.7 million to the trust funds. Forty-four cents kept per corrected dollar, after contingency fees, administration, and appeal reversals, in the government's own accounting.
Now ask what your PI vendors reported last quarter, and at which stage of that waterfall their number was measured. Six lessons from building an instrument that asks.
1. Gross is a stage, not a result. Between "identified" and "kept" sit collection (6–9% of identified overpayments are never collected at all, per GAO), appeals and overturns, vendor contingency, your own administration, and rebilling. Every stage is real money. A program reported at identification value is a press release with a dollar sign.
2. There are two overturn stories, and honest people tell both. The famous statistics — ALJs reversing 56–72% of appeals — come from the 2010–2014 RAC era. The restructured program shows roughly 30–37% of appealed claims overturned. Different rules, different denominators, both true. A vendor quoting only the low number is marketing; a provider association quoting only the high one is advocacy. Your projections should say which era they're standing in, and why it transfers.
3. The fee clause is the incentive design. A vendor paid a share of findings, keeping its fee when the finding dies on appeal, is optimized for volume — you fund both sides of every bad finding. Medicare's own RAC contracts claw fees back on overturn. And here's a fact worth putting in every board deck: the only contingency rates ever made public were 9.0–12.5%, in 2009. Every current figure you've heard is rumor.
4. Extrapolation survives court and shrinks at the ALJ. Federal appellate courts consistently uphold Medicare extrapolation — the burden is the appellant's, the sampling is presumed valid. But at the ALJ level, one contested review procedure alone accounted for $42M+ in overturned extrapolations in two years. An extrapolated finding is an opening position. Budget for the ALJ round, not for vindication.
5. The analysis itself starts a clock. Under the amended 60-day rule, an overpayment is "identified" on the False Claims Act knowledge standard — and a data-mining analytic identifies things, including, sometimes, overpayments in your own favor that must be returned. Running the query creates legal obligations before the first recovery letter goes out. If there's no process for day 61, there shouldn't be a query on day one.
6. Shift left changes the law that applies. Moving an edit prepay converts a recovery into a denial — cheaper and cleaner, and now it lives inside turnaround clocks, appeal rights, and the prior-auth metrics published every March. Same dollar, different regulatory address. Half the "shift left" business cases we've seen never modeled the address change.
We built these questions into an instrument: the program inventory, the net waterfall beside the federal yardstick, the may-you gates, the vendor incentive flags, the extrapolation file checklist. It takes the reported figure as given and never recomputes it — it just asks whether the figure survives its own decomposition. Browser-only; nothing entered leaves the machine.
A recovery number without its waterfall is a wish with a decimal point.
What does your PI program actually keep per identified dollar — and who in your organization has computed it?