A warehouse back injury, ninety days off work and a bill that does not match the injury

A warehouse picker lifts a carton wrong and feels something go in the lower back. Workers' comp claims like this one are where most of a comp desk's money goes: not the catastrophic injuries, but the ordinary ones that run longer than they should, with treatment that drifts away from the injury and bills that nobody reads line by line.

This post follows one such claim through InsightCW — the team assignment, the indemnity facts, the medical review agent's findings, the reasonable-value range and the referral that fires when time off work passes ninety days. The worker, the employer and the amounts are illustrative, and so are the workers' comp schedules: InsightCW says so wherever it shows them, and they must be checked against the state's current schedule.

Where comp claims go wrong

Workers' comp is unusual among lines because the carrier pays medical bills directly and for as long as treatment is reasonable and related. That second condition — related — is where cost builds. An injury to the lumbar spine does not explain a shoulder MRI. Twelve weeks of passive therapy with no change in function does not explain twelve more. A provider billing the same procedure code twice on one date of service is either an error or a pattern.

None of these steps is negligent on its own. The problem is order: money moves first and the question “does this bill match this injury?” is asked afterwards, if at all. InsightCW changes the order.

Who talks to whom

Step by step

The reasonable-value range is a formula, not a model

Bodily-injury value is where AI in claims is most often oversold. InsightCW does not ask a model what an injury is worth. The injury worksheet is a transparent formula that anyone on the desk can re-run by hand:

# the injury worksheet's value range (cw_injury_service.value_range)
low  = (specials + medical x multiplier_low ) x liability_share x venue_factor
high = (specials + medical x multiplier_high) x liability_share x venue_factor
# defaults: multipliers 1.5 and 3, liability 100%, venue 1 — each one editable, each capped at the policy limit

The same helper serves the evaluation screen and the medical review agent, so the two can never disagree.

The medical review agent explains the range: which specials it counted, which it left out as unrelated, and why. The model writes the explanation; the formula writes the numbers. A model sentence carrying a figure the worksheet did not produce is thrown away.

What the screens show

What the agents said, and what each answer rests on

Edge cases, and how authority stops them

What happens	What InsightCW does
No body part recorded	The medical review agent cannot say a bill is unrelated to an injury it does not know, so it says what is missing and prepares nothing.
The provider is on a watch list	The provider-flag indicator raises the fraud score, and the fraud agent explains it beside the medical review — two findings, two sources, one file.
A manager sets medical review to L4	No effect. Sending bills to medical review is capped at L2.
A bill is paid and later flagged	The flag still appears, with the payment named; a void goes through the ledger and tells InsightCede (claim.payment_voided).
The worker returns to work at day 60	TTD stops accruing; the 90-day referral never fires; the reserving agent's next run reads the new facts.

The decision record for this claim

On day 91, the Analyze run that followed the long-TTD referral read as follows. Notice how much of it is not about the referral: the record shows the whole state of the file at that moment, so the supervisor decides with everything in view.

The verdict is refer, because the line rule refers and refer outranks warn. The supervisor agent's brief is built from the same run, so the brief and the record cannot disagree.

Ask the copilot

Tuning it for your book

• Replace the schedules. The workers' comp schedules InsightCW ships are illustrative. Load the states you write before any of them drives a payment.
• The long-TTD threshold. WC_LONG_TTD refers at 90 days off work. Some desks refer at 60 for certain injury types; express that as a condition on the rule.
• The value-range defaults. The injury worksheet's multipliers (1.5 and 3), liability share and venue factor are defaults per evaluation, editable on the worksheet. Agree the defaults with your bodily-injury leads; the agent explains whatever the worksheet holds.
• Assignment. ASSIGN_WC routes every comp claim to the Workers Comp team. Split it by state or injury type if your teams are organised that way.

Variations on this claim

A fatality

SEV_FATALITY makes the claim critical at once, assignment routes it to senior adjusters, and the reserve almost certainly exceeds the adjuster's authority on day one — so the supervisor agent prepares the approval with the authority gap and precedent. The medical review agent has little to say; the correspondence agent's letters matter more, and a person writes them.

The worker reaches maximum medical improvement

Recording the MMI date and a PPD rating changes what the reserving agent reads: indemnity moves from temporary to permanent partial, and the reserve's basis changes with it. The schedule that turns a rating into weeks is state-specific and, as shipped, illustrative.

Medicare entitlement

When the injured person may be a Medicare beneficiary, the injury record tracks the Section 111 obligations — the beneficiary query and the TPOC or ORM report. In this build those go through a stand-in transport: InsightCW prepares them and nothing leaves until a carrier connects its reporting agent.

A time-limited demand

On a liability claim with a bodily injury, a time-limited demand at or over the limit that the evaluation can reach is a policy-limits exposure: an urgent diary before the deadline and the supervisor told. Comp rarely produces these; general liability and auto produce them often.

Rolling it out on your desk

1. Before go-live: replace the illustrative schedules with the states you write, and agree the injury worksheet's default multipliers with your bodily-injury leads.
2. Week 1: require the injury and body part on every comp claim. The medical review agent cannot judge relatedness without them, and it will say so rather than guess.
3. Weeks 2–6: medical review at L2, so flagged bills arrive as decisions. Compare what the agent flagged with what your nurse reviewers would have flagged; the Change and Reject answers are the evidence.
4. Month 3: look at medical paid per claim and bills reviewed before payment against the months before. Both come from InsightCW's ledger.

What the person still decides

• Compensability. Whether the injury arose out of and in the course of employment is a legal judgement.
• Which bills go to review. The agent prepares the list; the adjuster or nurse sends it.
• The treatment plan conversation. An IME, a second opinion, a return-to-work offer — these are relationships, not rules.
• The reserve. The agent recommends with a basis; the person posts it, inside their authority.

What to measure

Medical paid per claim against the line's own history; the share of bills reviewed before payment rather than after; days to first indemnity payment; TTD duration; and reserve changes per claim, which falls when the first reserve rests on evidence. All are read from InsightCW's ledger and clocks; see measuring it.

Keep reading