AI Adoption GuideHealthcareIntake
Insurance benefits explainer
NLP agent answers patient questions about copay, deductibles, and out-of-pocket estimates from live payer data before the visit starts.
Healthcare processAccessIntakeAssessDiagnoseTreatDischargeBillFollowup
By Don, DoneThat’s AI coach · updated
Pull the eligibility payload before you name a dollar
Name a dollar only after the current eligibility response is loaded for this member, this payer, and this date of service. A benefits explainer that starts from a fee schedule, a last-visit copay, or what the patient "usually pays" is already off the payload.
Real-time eligibility verification is the upstream step. The explainer reads that response. It does not run a second eligibility check and it does not invent a second source of truth.
Confirm three matches before any patient-facing sentence: the member on the response is the patient on the appointment (not only the subscriber, not a sibling dependent), the date of service is the scheduled visit, and the service type on the inquiry is what the clinician will bill. A copay cited from an office-visit service type does not explain an imaging visit.
Clearinghouses and practice systems (Waystar, AKASA, Epic, athenahealth) already return eligibility as a class of vendor tools. The explainer maps returned fields into language a patient can follow. It does not scrape a payer portal, backfill a missing remaining deductible, or round a number because the visit starts in ten minutes.
If eligibility has not returned, the patient-facing output is that coverage details are not available yet. It is not a rounded copay.
Cite the payer field or leave the line blank
Quality means every dollar the agent speaks is tied to a named field on that eligibility response. Copay from the copay field. Deductible from the deductible field the payer sent. Out-of-pocket from the out-of-pocket field the payer sent.
Empty stays empty. If the payer omitted remaining deductible, the agent does not compute one from last year's plan, from a sibling's visit, or from a "typical PPO." The line stays blank. The counselor sees the blank and decides what to say.
Citing has to be visible, not implied. A counselor reviewing the transcript should be able to point at the field that produced each sentence. "Your specialist copay is $40, from the specialist copay amount on this eligibility response" is an explanation. "$40 copay" with no field behind it is a number the practice cannot defend when the statement prints a different amount.
Out-of-pocket estimates follow the same rule. An estimate is allowed only when the payer returned the components that estimate uses. If coinsurance is present and allowed amount is not, do not multiply a clinic charge by coinsurance and call it today's out-of-pocket. Name what is known. Leave the rest for the counselor.
One visit: copay, deductible, and what stays unsaid
A patient messages before a Tuesday follow-up: "What will I owe?"
The agent loads today's eligibility for that member and that date of service. Specialist copay is present: $40. Deductible applies to this visit type, but the remaining deductible field is empty. Out-of-pocket maximum remaining is present: $1,200. Coinsurance for this service type is not returned.
The agent answers only what it can cite. It says the specialist copay on the eligibility response is $40. It says remaining deductible was not returned by the payer, so it cannot state a deductible amount. It says remaining out-of-pocket maximum on the response is $1,200, and that this figure is a remaining maximum, not a prediction of today's bill. It does not invent coinsurance.
The counselor then talks to the patient. They can repeat the cited copay, say deductible still has to be confirmed because the payer left it blank, and make clear that today's total is not a quote. If the patient needs the same conversation in another language, hand off to a multilingual patient navigation agent rather than asking the benefits explainer to paraphrase coverage it did not cite.
This is an illustration of the rule, not a clinic result. No visit volume, no collection lift, no minutes saved.
Three ways an explainer becomes a billing fight
A dollar with no payer field is the usual failure. The model has seen copays in training data. The clinic cash-pay rate is sitting in a prompt. Someone pastes $40 because last month's patients paid $40. None of those are the eligibility field. If the field is missing, the output is missing. A fabricated copay is worse than a blank: the patient plans around it, then the statement arrives different.
Treating the chat as a quote is the second failure. Patients screenshot the thread. Front desk repeats it. A supervisor later treats the transcript as a good-faith estimate. The explainer is a reading of today's eligibility fields. It is not a price commitment, not a contract, and not a substitute for the estimate your organization is required to give. The counselor's spoken caveat belongs in the same encounter as the chat. If the practice needs a formal estimate, generate one through the estimate workflow. Do not promote the chat to that document.
Inventing a deductible is the third. Deductible is the field staff most often fill by habit. Remaining deductible is frequently absent, stale, or returned as a plan-level number that does not match this member. Computing "you've probably met it by September" is invention. So is copying the annual deductible when remaining is blank. So is assuming family deductible when the response only shows individual. Leave the line empty. Let the counselor explain the uncertainty instead of letting the model erase it.
Prior authorization is a separate question. If the service may need auth, do not fold "you're covered" into the benefits chat. Point that work to autonomous prior authorization and keep the explainer on the fields it actually has.
The counselor, not the chat, closes the conversation
The outcome is quality of the answer, not replacement of financial counseling. A counselor still talks to the patient. The agent prepares cited language and flagged blanks. The counselor interprets plan quirks, hardship, deposits, and what happens if the payer later reverses eligibility.
Review the transcript the way you would review a new counselor's first week. Every dollar has a field. Every blank is still a blank. No sentence promises a total the payer did not support. If a sentence cannot point at a field, delete the sentence. Do not "clarify" it into a new number.
When the visit is over, charging is a different system. Do not ask the explainer to justify a charge from the note. That is automated charge capture from notes, downstream of the encounter. Mixing intake explanation with post-visit coding is how a copay chat turns into a documentation argument.
Adjacent intake work this does not replace
Eligibility must succeed before explanation. If eligibility is down, failed, or matched to the wrong member, stop. Do not explain yesterday's response as if it were today's.
Language access is not a benefits feature. Translating a cited field is allowed. A translation that adds a deductible the source payload lacked is still invention.
Auth status, medical necessity, and network direction sit beside this page, not inside it. The explainer may say the eligibility response did not return an auth-required flag, or that the field is present and set to yes. It may not approve the service.
Keep the patient conversation in intake. Keep the dollar tied to the payer. Keep the blank a blank. Keep the counselor in the room.
Is this worth automating for you?
Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other.
DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.
Measure the baseline first