Refuse a savings claim with a measurement behind the refusal.

None of these is novel to an actuary. What is new is that each one is a call with a preserved corpus behind it, so the answer arrives with its sources and its vintage attached and can be re-run in front of the vendor.

1. 01

   What did they actually declare?

   The population, the baseline period, the comparison, the unit cost source and the attribution window, separated into what was measured, what was cited from somewhere else, and what the claimant simply declared. A figure computed from a mix inherits the weakest of its inputs, and the claim rarely says so.

   CALL declared_inputs, savings_methodology

2. 02

   How much of it is the mean coming back?

   A cohort selected because its costs were high will be cheaper next period whether or not anything was done to it. The expected regression is calculable from the selection rule and the spread, and the number it produces is usually a large fraction of the claimed saving.

   CALL regression_to_mean_expectation

3. 03

   Deferred, or avoided?

   A procedure that did not happen this year and happens next year is not a saving; it is a timing difference. The two get counted the same way in most vendor arithmetic, and separating them changes the sign of some claims.

   CALL deferred_versus_avoided

4. 04

   What is the exposure worth at all?

   Size the cost of care the claim is drawn against, over a declared population, line of business and funding type, before arguing about a percentage of it. A saving larger than the exposure is a stopping condition, not a negotiation.

   CALL cost_of_care_exposure

5. 05

   How hard does it have to be pushed to break?

   Move the inputs across their plausible ranges and see where the claim stops holding. A result that survives only at one corner of the input space is a result about that corner.

   CALL sensitivity, simulate

6. 06

   May the sentence be written?

   Put the sentence the vendor wants in the renewal memo to the checker. It returns the figures the sentence rests on, or refuses it and names what evidence would settle the question.

   CALL claim_check, judgement_check

Every figure carries its basis: measured, cited, or declared. Anything computed from a mixture inherits the weakest input, which is the rule most savings arithmetic quietly breaks. When the record will not carry a sentence, the answer is a refusal naming what evidence would settle it, rather than a hedged number somebody downstream will quote without the hedge.

There is no learned decision layer here. A reviewer can inspect the source, the transformation, the rule and the decision trace directly, which is what makes a finding usable in a room where the other side gets to argue. The 25 refusal classes are published in advance, so both sides know what is off the table before the meeting.

What is regression to the mean in a healthcare savings claim?

It is the tendency of a cohort chosen because its costs were unusually high to have lower costs in the next period regardless of any intervention. Because most care-management and digital health programmes enrol people at the top of a cost distribution, some of the drop that follows is arithmetic, not effect. The expected size of it can be calculated from the selection rule and the spread of the population, which turns an objection into a number.

What is the difference between a deferred cost and an avoided cost?

An avoided cost is care that does not happen. A deferred cost is care that happens later. Both look identical inside a twelve-month measurement window, and vendor arithmetic frequently counts them the same way. Separating them is often the single largest correction to a claimed saving.

What makes a saving hard rather than soft?

A hard saving reduces a cash outflow that appears in a financial statement: fewer paid claims, a lower unit price, a contract that costs less. A soft saving is an avoided hypothetical, a productivity estimate, or a utilisation change priced at an assumed unit cost. Both may be real. Only one survives a finance review, and a claim that does not label which it is has made the choice for you.

Do you need the vendor to co-operate?

No. The work starts from what the claimant has already published or supplied to you: the methodology document, the declared inputs, the reported figures. Where the claim rests on something the record does not contain, the finding is that the record does not establish it, which is itself a usable result at a renewal table.

Do you work for the vendor or for the buyer?

This work is aimed at the evaluator: payer vendor-management teams, actuarial consultants and investor diligence groups. A vendor wanting its claim strengthened before renewal and a buyer wanting a weak claim caught before purchase are opposed positions with different confidentiality rules, and we do not take both sides of the same claim.

What will it refuse to conclude?

The world refuses whatever the evidence does not carry, by name, with what would settle it. It will not tell you the vendor is wrong, only what their stated method does and does not support; it will not produce a paid-amount claim from published rates; and it will not attribute an effect to a programme from a before-and-after alone. There are 25 published refusal classes.

What does the output look like at a renewal or diligence meeting?

A short written finding with the figures it rests on, the vintage of every source, and the sentences the record declines to support. It is built to be handed to the other side, which is the only test that matters: a refusal you cannot show them is an opinion.

The useful first move is a real claim you are already sceptical of, with whatever the claimant has published about how it was computed. Twenty minutes, no charge, and the analysis is yours either way.

Bring us a claim → Docs →