Case Study 2 — The Queue That Worked Perfectly: Providence, Washington's Charity Care Law, and the Machine Pointed the Wrong Way

A real matter, from the public record: the Washington State Attorney General's 2022 lawsuit against Providence Health & Services, contemporaneous investigative reporting, and the publicly announced 2024 resolution. Allegations are labeled as allegations; where Providence disputed characterizations, that is noted; reported figures are as publicly announced and should be verified at the source. The subject is the failure mode §31.10 exists to prevent: collection machinery — competent, measured, and diligent — run against patients the law said should never have been in it.

Background

Section 31.10 gave the sequencing rule in six words: determine, then pursue — never the reverse. Bad debt is a balance the patient could have paid; charity care is a balance the patient could not; and the only way to know which one you are holding is to screen before collecting.

Washington State wrote a version of that rule into law decades ago. Its charity care statute requires hospitals to notify patients that financial assistance exists and to screen for charity care eligibility before undertaking collection, with income thresholds the legislature has since expanded substantially. In Washington, §31.10's sequencing is not a best practice. It is a legal element of the collection process itself.

Providence is one of the largest nonprofit health systems in the country, headquartered in Washington, operating dozens of hospitals across several states. Which is what makes the case instructive: this is not a story about an understaffed office that had no process. It is a story about a sophisticated revenue cycle organization whose process worked — measured by every number in this chapter except the one that mattered.

What the state alleged

(Allegations from the Attorney General's complaint, filed February 2022, supplemented by investigative reporting published later that year. Providence contested aspects of the state's characterization while acknowledging problems and describing corrective steps.)

The program had a name and a designer. Beginning around 2018, Providence engaged a major consulting firm to build a revenue initiative — internally called "Rev-Up" — aimed at increasing point-of-service and patient collections. Training materials quoted in the complaint and in reporting instructed registration and collection staff to ask every patient for payment, every time, to persist past initial refusals, and to follow scripts that did not lead with — and in practice tended away from — the information that financial assistance existed.

The machinery downstream ran exactly as Chapters 24 through 31 describe machinery running. Balances that went unpaid moved through statement cycles into follow-up. Accounts that aged out were placed with collection agencies — tens of thousands of them, per the complaint — including accounts belonging to patients whose incomes qualified them for free or discounted care under Washington law, and, the state alleged, accounts belonging to Medicaid enrollees, a population categorically likely to qualify.

The state's legal theory tracked §31.10's sequencing rule almost exactly: patients were pursued, and placed, without the charity care screening and notice the law requires to happen first. The suit alleged violations of the charity care statute and the state Consumer Protection Act.

What the reporting added

A national newspaper investigation in September 2022 put faces and documents to the pattern: patients who qualified for free care describing collection pressure at the point of service and collectors afterward; internal training materials with scripts; employees describing discomfort with what the scripts asked of them. For this book's purposes the details matter less than the shape: the information that would have stopped the machine existed at the front desk — a question away — and the scripts were engineered to not ask it.

The resolution

In early 2024, the Attorney General announced a resolution requiring Providence to forgive roughly \$137 million in patient medical debt and refund roughly \$20 million — approximately \$158 million in total relief, reported as reaching on the order of a hundred thousand patients — described at the time as the largest charity-care enforcement resolution in the country. Providence, which did not concede the state's characterizations wholesale, had by then also made changes: revised policies, expanded assistance, refunds in progress. (Figures as publicly announced; verify at the source.)

What it shows

First — and this is why the case sits in this chapter rather than Chapter 32 — every AR mechanism in the story was working. The queues drained. Follow-up followed up. Placement happened on schedule. Point-of-service collection rates presumably improved, because that is what the program was built to improve and asking every patient every time does improve it. A dashboard built from §31.11's SPEED and YIELD sections would have glowed. The failure was invisible to every collection metric because it was not a collection failure. It was a classification failure: balances that were charity care by law were processed, from the first script to the final placement, as presumptive bad debt.

Second, the missing number was a missing category — Chapter 29's Case Study 1 at statute scale. A root-cause list with no row for payer error could only ever blame the practice; a patient-AR process with no screening step can only ever conclude that an unpaid balance is a collection problem. The measurement system did not report "unknown eligibility." It reported the nearest available answer: pursue. The countermeasure is the same in both cases — build the category, then put a number on the dashboard that goes red when the category is skipped: accounts placed without a completed FAP screen is as computable as days in AR, and almost nobody computes it.

Third, the scripts are this book's configuration thread applied to human beings. A billing macro that appends modifier 59 (Chapter 21), a template that answers its own question (Chapter 29's Case Study 2), and a collection script that persists past "I can't pay this" are the same object: a configuration that makes an assertion — here, "this person can pay and should be asked again" — on every encounter, without anyone deciding it in the individual case. §31.10's 📞 taught the individual counter-move; this case shows why the counter-move cannot be left to individual virtue when the script points the other way.

Fourth, the finding came from outside — again. An attorney general and a newsroom saw the pattern before any internal metric surfaced it, joining the thread that runs through Chapter 26's Case Study 1 (detectable from outside before inside) and the commercial-purpose findings Chapter 37 §37.8 has been promised. A pattern of placements against low-income ZIP codes and Medicaid enrollees is visible in data a system's own analysts hold; what was missing was not the data but a question anyone was paid to ask of it.

And fifth, the case prices the reputational arithmetic Chapter 32 §32.9 will formalize. Set aside the resolution's dollars: the durable cost is that a charitable institution's name became a national shorthand for hounding the poor — over balances that, individually, were exactly the small sums §31.7 teaches practices to decline to chase when mere postage is at stake. An organization that will not spend \$2.95 to chase \$4.15 should need no statute to stop it from spending its name to chase a few hundred dollars from a patient who qualified for free care.

The lesson

Collection machinery has no opinion about where it is pointed. The same queues, cadences, placements, and dashboards that responsibly retrieve money a solvent payer owes will, run against the wrong population, efficiently manufacture harm — and the metrics that prove the machine is working are structurally incapable of noticing. The protections are all upstream and all boring: a screening step that is a hard gate, not a script option; a written FAP applied uniformly; placement files reconciled against eligibility, every account, every time; and one dashboard line — placed without screening — whose only acceptable value is zero.

Discussion questions

1. Reconstruct the dashboard Providence's program would have produced under §31.11's SPEED, YIELD, and COST headings. Which numbers improved because of the conduct at issue? What single line would have caught it, and in which section of the dashboard does it belong?

2. Washington's statute makes screening a legal precondition to collection; most states' laws are weaker, and physician practices sit outside §501(r) entirely. Argue for or against this claim: "§31.10's sequencing rule is good operations even where no law requires it." Use the cost and reputational material from §31.7 and this case.

3. The scripts told staff to ask every patient, every time — which is also, nearly verbatim, legitimate point-of-service collection advice from Chapter 24 §24.8. Where exactly is the line? Write the two-sentence version of the script that keeps the collection benefit and loses the violation.

4. Compare this case's missing-category mechanism with Chapter 29's Case Study 1. In each, what did the measurement system report instead of "unknown," and who absorbed the error? What does the comparison suggest about where to look for the next missing category?

5. The relief was reported as roughly \$158 million spread across on the order of a hundred thousand patients. Compute the rough average per patient, and then say what the distribution behind an average like that probably looks like. Against §31.7's arithmetic — where the practice's own pursuit costs make sub-\$10 balances not worth chasing — what does it tell you that balances of this general size, pursued at scale, produced the largest charity-care enforcement resolution in the country?

6. Providence made changes before the resolution and disputed parts of the state's account. What would you need to see, as an outside auditor two years later, to conclude the fix was real? Name three artifacts, at least one of which must be a number on a page someone reads.