The small balances nobody works
Nobody decides to write off a forty-dollar claim. The queue never reaches it, the filing deadline passes, and an adjustment code closes it. The write-off is real, it repeats, and it appears nowhere as a choice.
The rule that produces it
The rule is rational and nobody stated it. A claim costs the same to work whatever it is worth, so at some balance the expected recovery falls below the cost of the touch, and below that line working the claim loses money. Sorting a worklist by balance implements that rule automatically, without anyone having to name the threshold or defend it.
The trouble is not the rule. It is that the threshold is invisible. Nobody knows where it currently sits, whether it is the same for every payer, or what it costs, because abandonment happens by omission and omissions generate no records.
Make the omission visible
There is a number worth building, and building it is not hard: claims that reached their filing deadline without ever being touched.
Cut it by payer, by balance band and by denial reason. What comes back is the threshold your operation is actually running, as opposed to the one anyone would say it was running. It is usually higher than expected, and it usually varies by payer for no principled reason — one payer's small claims get worked because they sit next to large ones in the same worklist, another's do not because they arrive in a batch of their own.
That report is a decision you are already making. Publishing it changes none of the arithmetic. It changes whether the arithmetic is being applied on purpose.
It also arrives at the wrong time to be useful, which is worth fixing separately. By the moment a claim is written off, everything that could have been done is out of reach. The same cut, run against claims that still have time left, names the group you can still act on. One version of the report is a post-mortem; the other is a worklist, and they are the same query with a different date filter.
Small balances are not scattered
The second thing the report shows is that abandoned claims cluster. They share payers, service categories, fee schedules and denial reasons, because low balances come from low-priced services and low-priced services are concentrated.
The clustering is the useful part. If the abandoned set were random, working it would mean paying the full cost of a touch on every claim in it. It is not random, so it does not.
Where a group of claims shares a payer and a denial reason, the analysis is done once. The first claim costs a full touch: read the remittance, work out what the code means for this payer and this service, establish what fixes it. The second claim and the fortieth cost the correction only. The average cost across the group can fall well below the cost of the first, which is exactly the movement that changes which side of the threshold the group sits on.
That is why the small-balance question is a batching question rather than a triage question. Triage asks which of these claims is worth a touch. Batching asks how few touches this group needs.
Where the threshold should sit
Once the cost of a touch can be reduced for a group, the threshold stops being a constant and becomes something set per group. A category where the resolution is known and the correction is mechanical has a low threshold: it is worth working down to small balances. A category where every claim needs its own investigation has a high one, and small balances there genuinely are not worth the time.
Two thresholds applied deliberately will collect more than one threshold applied by accident. And the difference is not found by working harder through the worklist. It is found by asking what the abandoned claims have in common — which requires having a list of them.