Vendor jargon, decoded.

A tool finds the relevant parts of a document, such as a payer policy, and has an AI model interpret them for each question. Useful for reading, but the interpretation can change from one run to the next.

Ask: “If I run the same case twice, do I get the same answer, for the same reasons?”

The kind of AI model behind ChatGPT, Claude, and Copilot. Very good at reading and writing text; its answers are generated, not looked up.

Ask: “Which parts of the decision does the model make, and which are fixed rules?”

Software that uses an AI model to carry out multi-step tasks on its own, such as checking a portal, filling a form, or drafting an appeal.

Ask: “What exactly can it do without a person approving it, and how do we see what it did?”

When an AI model states something that isn't in its source, such as a criterion the policy doesn't contain or evidence the chart doesn't show.

Ask: “Can I click from every answer to the exact sentence in the policy and the chart?”

Rules that always give the same answer for the same input. For payer policy, the same case against the same policy version gets the same result every time, with the reason attached.

Ask: “When the policy is ambiguous, how is that shown instead of silently decided?”

A number a tool attaches to an answer to say how sure it is. It describes the model, not whether the payer will agree.

Ask: “What happens to low-confidence cases, and who reviews them?”

The share of cases a tool finishes with no person involved. Vendors often quote it as a headline number.

Ask: “Measured on whose cases, and how many of the touchless ones were later denied?”

Software that pulls facts out of written text, such as diagnoses, dates, and treatments from clinical notes. Older NLP matches patterns; newer tools use LLMs.

Ask: “How does it handle negation, like "no history of" or "patient denies"?”

Structured data sits in fields a system can read directly, like codes and dates. Unstructured data is free text: notes, scanned letters, faxes. Most medical-necessity evidence is unstructured.

Ask: “Does it read our notes and scans, or only discrete EHR fields?”

Turning scanned pages, faxes, and handwriting into text that software can read. Accuracy on messy handwriting varies a lot between tools.

Ask: “Can I see how it reads our worst scanned or handwritten notes?”

A person reviews or approves what the software proposes before it is used. The useful question is which cases reach a person, and what they see when they do.

Ask: “What does a reviewer see: just the answer, or the evidence behind it?”

A record of what was decided, when, by what rules or model, and on what evidence, so a decision can be explained later to a payer, an auditor, or a patient.

Ask: “Can you reproduce a decision from six months ago exactly as it was made?”

Running a new tool beside the current process on past cases and comparing its answers with what actually happened, before it touches live work.

Ask: “Can we trial it on our own past cases without changing anything live?”

The standard way modern systems like Epic exchange clinical data through APIs.

Ask: “Which data would you read from our EHR, and would you write anything back?”

A standard way to launch a third-party app inside an EHR such as Epic, so it opens in context of the patient the user is already looking at.

Ask: “What does our Epic team need to configure, and does it depend on our Epic version?”

Submitting and tracking prior authorization requests electronically instead of by portal, phone, or fax. The X12 278 transaction is the long-standing format; newer FHIR-based APIs are replacing it.

Ask: “Which of our payers does it actually support electronically today?”

FHIR standards for prior authorization: CRD tells the provider whether auth is needed, DTR collects the documentation the payer requires, and PAS submits the request.

Ask: “Which of these do you support, with which payers, and in production or pilot?”

A CMS rule, finalized in January 2024, that requires Medicare Advantage, Medicaid, and CHIP plans and plans on the federal ACA marketplace to speed up prior authorization decisions and to support prior authorization APIs, with most API requirements taking effect in January 2027.

Ask: “What changes for us, and which payers does it actually cover?”

Medicare's own coverage rules: national decisions from CMS and local ones from each Medicare Administrative Contractor. Medicare Advantage plans generally must follow them too.

The HIPAA contract a vendor signs before a health system shares patient information with it.

Ask: “Where is our data processed and stored, by whom, and for how long?”

A payer's published document describing when it considers a service, drug, or device medically necessary, and what documentation it expects. Most are PDFs, and payers revise them regularly.

Whether a service is reasonable and needed for the patient's condition under the payer's criteria. Payers decide it by comparing the chart against the criteria in their medical policy.

The specific conditions inside a medical policy that a case must meet: an age, a diagnosis, a failed prior treatment, an imaging finding, a duration of symptoms.

Licensed sets of clinical criteria that many payers and hospitals use for utilization review, alongside or instead of criteria written into their own medical policies.

A group of related medical policies, such as a payer's policies for spine procedures, infused oncology drugs, or advanced imaging. Denials tend to cluster by family.

The version of a medical policy in force on a given date. A claim is judged against the version in force on the date of service, not today's.

Getting payer approval before a service is performed. The request has to show that the case meets the payer's clinical criteria.

Confirming, before the date of service, that a scheduled visit is covered: eligibility, benefits, referrals, and any prior authorization the payer requires.

A program where a payer exempts a provider from prior authorization for certain services because of a strong approval record. Rules vary by payer and by state.

A denial because the payer decided the service didn't meet its criteria, often because the chart didn't show a criterion clearly, not because the patient didn't qualify.

Standard codes on a remittance explaining why a claim was denied or paid differently. They say which category a denial falls in, but rarely which criterion was missed.

A conversation between the ordering clinician and the payer's medical reviewer to discuss a denial or pending request, often before a formal appeal.

A request for the payer to reverse a denial, usually by pointing to the chart evidence that meets each criterion the denial turned on.

The work of making sure the chart accurately reflects the patient's condition and care, which is what payers and coders read.

Making sure a health system is paid correctly for the care it provides: charges, coding, and documentation that hold up against payer rules.

Checking that claims are paid according to the contract and policy: catching underpayments, overpayments, and incorrect denials.

The code sets used to describe procedures, supplies, and drugs on a claim. J-codes are the HCPCS codes for many injected and infused drugs. Two codes, or two drug names, can describe the same thing.

A service that checks and routes claims and payer responses between providers and payers, and returns remittances and rejections.

The payer's electronic explanation of how a claim was paid or denied, including reason codes. It is the usual starting point for analyzing denials.