An explanation of benefits can look like a bill even though it is not one. A provider bill can arrive before a claim is final. A plan may show patient responsibility that does not match the provider's ledger, while a customer service agent can see only one of those records. This study protocol evaluates whether support teams identify the actual discrepancy, protect health information, and route the question to an owner who can resolve it.
The protocol does not decide whether a charge is legally valid, calculate benefits for an individual, or claim that every difference is an error. It is an evidence method for completed support cases. The core test is whether the record shows what the customer received, what the systems showed at that time, which authority could decide the issue, and what happened next.
Define the case around a specific claim
Use one claim or clearly linked set of claim lines as the unit. Record the claim number, date of service, service description, billed charge, allowed amount, plan payment, adjustments, and stated patient responsibility using study keys rather than public identifiers. Keep the EOB version and the provider bill version that prompted the contact. Later documents should not overwrite the original evidence.
CMS explains that an EOB is not a bill and describes the claim details and cost fields it commonly contains (How to read an EOB). That distinction should appear in the coding guide. It should not be used to dismiss a customer's concern, because the EOB can still reveal a denial, unexpected cost sharing, or network discrepancy that requires action.
Merge repeat contacts only when they concern the same claim state and requested outcome. If a reprocessed claim produces a new EOB and a new mismatch, preserve it as a later episode linked to the first. This makes it possible to measure recurrence without double-counting the original question.
Classify the discrepancy before selecting a queue
Create a taxonomy based on observable differences: EOB versus provider bill, duplicate claim, denied service, incorrect patient or provider information, network-status dispute, missing payment, coding question, coordination-of-benefits issue, deductible allocation, prior-authorization question, or a possible surprise-billing concern. Provide an unknown category so agents are not forced to guess.
CMS defines common terms such as allowed amount, balance billing, complaint, cost sharing, and EOB (Health insurance terms). Use those definitions to normalize language in the study. Do not infer a clinical code, coverage determination, or legal conclusion from the customer's description alone.
Record the customer's requested outcome separately from the discrepancy class. A person may want an explanation, a corrected bill, a claim reprocessing request, a payment plan, or a formal complaint. The same apparent mismatch can therefore require different next steps.
Reconstruct the evidence available to the agent
For each contact, inventory what the agent could see: EOB image, plan claim record, remittance advice, provider ledger, bill image, prior authorization record, network directory snapshot, notes from an earlier contact, and payment history. Mark each source as present, absent after search, or inaccessible to that role.
Compare the agent's explanation with the evidence available at send time. A later correction does not prove the first answer was unreasonable, but an unsupported statement such as "the provider billed you incorrectly" should be coded as premature if the agent had no provider ledger. Likewise, "the plan denied it" may be inaccurate when the claim was pending.
Preserve timestamps and version identifiers. Claims can move from received to pending, denied, adjusted, and paid. A static export collected weeks later can hide the state that shaped the original conversation. The study package should retain a point-in-time event history where systems allow it.
Map authority and routing rules
List which owner can explain benefit calculations, correct demographic data, amend a provider bill, resubmit a claim, review a coding issue, determine network status, accept a complaint, or evaluate a No Surprises Act question. Frontline support may gather and explain evidence without having authority to change every source record.
For each transfer, record the reason, receiving team, acceptance event, information passed, and stated response window. A referral that gives the customer only a general phone number is different from a warm handoff with claim context. Measure transfer loops, repeated document requests, and cases closed before the receiving owner accepted them.
CMS's overview explains that No Surprises Act protections apply to specified emergency and non-emergency circumstances and limit certain cost sharing (Overview of Key Consumer Protections). Use the source to define a specialist-review trigger. Do not let a frontline script declare that every out-of-network charge is prohibited or that every billing difference falls within the Act.
Evaluate privacy and minimum-necessary handling
The study should record authentication, authorization, channel, attachment handling, and which fields were exposed during each handoff. HHS guidance describes the HIPAA minimum-necessary standard and its role in limiting uses, disclosures, and requests for protected health information when applicable (Minimum Necessary Requirement). This informs evidence minimization but does not establish that every sampled organization or transaction is covered.
Review whether agents asked customers to send full medical records when an EOB page or claim reference would have been sufficient. Check whether free-text notes copied diagnoses or identifiers into systems with broader access than the source. Code both overcollection and evidence gaps, because indiscriminate minimization can also prevent a legitimate review.
Analytical extracts should replace names, member numbers, claim numbers, and provider identifiers with study keys. Keep source documents in a restricted location. Publish only aggregate findings and scrub quotations that could identify a person.
Measure completion rather than contact closure
Define outcomes precisely: explanation supported by evidence, corrected provider bill, corrected or reprocessed claim, payment posted, complaint accepted, specialist review pending, customer withdrew, or unresolved at cutoff. A ticket marked resolved is not a completed outcome unless the promised action can be observed.
Measure time to acknowledgment, time to correct owner, time to evidence-complete review, time to decision, and time to customer notification. Pause and calendar rules should be reported rather than hidden. Present medians and distributions by discrepancy type, because complex coordination cases should not be blended with simple EOB explanations.
Audit promises. If an agent said a revised EOB would appear within a period, verify whether it did. If a provider-billing callback was promised, look for an accepted task and completed contact. Report unsupported certainty separately from delays.
Test coding reliability and uncertainty
Two reviewers should independently code a varied sample. Compare agreement for discrepancy type, evidence sufficiency, decision owner, privacy handling, outcome, and communication accuracy. Adjudicate with source records and update the codebook before final analysis.
Run a sensitivity analysis that treats inaccessible evidence as incomplete, then reports it separately from observed process failure. If the headline conclusion changes, the missingness is a main finding. Stratify by channel, source-system access, discrepancy type, and whether more than one organization had to act.
This study can show where EOB questions lose context or receive answers beyond an agent's authority. It cannot determine medical necessity, interpret a plan for an individual, or establish legal liability. Teams building a healthcare administrative support function can connect the method to healthcare admin staffing, email and ticket support, and escalation and QA. The decision point is not simply how quickly to answer. It is whether the operating model can assemble the right evidence and deliver the case to the party empowered to act.