Most Cost-of-Care Savings Numbers Are Produced by Hand, Once — and Then Argued About Forever
Ask how a reported medical-cost savings number was produced and you'll usually get a spreadsheet with one author, no version history, and a methods tab nobody has opened since the number went into the board deck. Then the number gets contested, and there's nothing to contest it with. Five lessons from building a set of cost-of-care instruments designed to be argued with.
1. Decompose the trend before you claim any of it. A cost trend has arithmetic inside it: price, utilization, mix, membership. A "bridge" that walks from last period to this one, category by category, turns "costs are up 9%" into four specific conversations — and usually reveals that the intervention being credited touched one of the four.
2. Savings claims die at realization, not at identification. Every portfolio of initiatives has a ladder: identified, validated, implemented, realized, banked. Money evaporates at every rung, and the evaporation rate is the most honest number in the portfolio. A pipeline reported at identification value is a wish list with a currency symbol.
3. Small populations make noise look like results. Regression to the mean plus a small denominator will manufacture a 20% "improvement" out of nothing, reliably. Before believing any before-and-after result, ask how much movement pure chance produces in a population that size — there's a floor, and if your result is under it, you've measured weather.
4. Some recoveries were never yours to make. Reclassifying an admission after payment, downgrading a level of care after prior authorization — federal rules constrain both, and an authorized service generally can't be denied later for medical necessity without good cause. A recovery program that doesn't check the may we question before the how much question is manufacturing appeals — and the overturn statistics say the appeals win.
5. Provider risk deals need a compliance floor, not just a spreadsheet. Physician incentive rules cap what risk can be transferred without stop-loss protection. A gainshare that pencils beautifully and violates the incentive rules is not an opportunity; it's a finding.
The pattern across all five: the number that survives is the one produced by a stated method, from disclosed inputs, with its own limitations printed on it. So we built the instruments that way — a trend bridge, a realization ladder, a risk-arrangement ladder, a recovery guard, and a noise floor that sits under the other four and says how much of any measured movement was chance. Every threshold cited, every report carrying its own caveats, everything running in the browser with nothing sent anywhere.
The uncomfortable one is the noise floor. Run it before believing any number — including ours. A tool that can't flag its own output isn't an instrument; it's marketing.
The thread through it all: a savings number is a claim, and claims deserve due process. The organizations that get durable value out of cost work are the ones whose numbers survive being checked — because they were built to be.
What's the largest reported savings number in your organization that no one has re-derived from the inputs?