Solution

A knowledge base for your practice's policies, billing, and SOPs

An AI knowledge base for healthcare practices answers the administrative questions that eat a front desk's day — what a payer's prior authorization policy requires this quarter, what a specific plan covers, how the practice's own SOP handles a returned check, where the manual for the new scanner is. It is not clinical decision support: it does not interpret symptoms, does not suggest a diagnosis or treatment, and it will not answer a clinical question. Every answer comes with the source document attached, so staff verify before acting instead of trusting a confident-sounding guess.

· Reviewed by Artur Horimoto, Founder & CEO

Where the answers already exist but nobody can find them fast enough

A payer's prior authorization policy is a PDF, and it changes without much warning — a procedure that needed no prior authorization in the spring can need one by fall, and the fax number for submitting the request can be different again by winter. Multiply that by every plan your practice bills, and the honest answer to "does this need authorization" often lives in someone's memory of the last time they dealt with that specific payer, not in a document anyone could quickly pull up.

The same pattern repeats everywhere administrative work touches paper. Billing and coding guidance sits in a mix of payer bulletins, internal cheat sheets, and whatever the coding lead remembers from the last denial. Internal SOPs — how to handle a returned check, what to do when a patient shows up without ID, how to process a records request — exist somewhere, but the version on the shared drive is not always the version anyone actually follows. Vendor and equipment documentation for the new scanner or the phone system sits in a manual nobody has opened since installation day. And a front desk with real turnover means the person who used to know all of this by heart may not be the person answering the phone anymore.

None of that is a clinical problem. It is a findability problem, and it has a specific, expensive symptom: staff stop looking things up and start walking over to interrupt the one coworker who has been there long enough to know the answer — and that coworker's own work stops every time someone asks.

What this is not

Before anything else: this is administrative and operational knowledge, not clinical support. It does not interpret symptoms, does not suggest a diagnosis, does not recommend a treatment, and it refuses a clinical question rather than guess at one. Nothing here replaces a clinician's judgment or a coding professional's final call on an ambiguous claim. What it does is put the paperwork, policy, and process material your practice already has in writing in front of the person who needs it, with the source document attached so they can check it themselves before acting. If a question falls outside what the indexed material covers, the system says so instead of filling the gap with something that merely sounds plausible.

What the system does, day to day

Once it is built around your practice's actual documents, the system works the same way for every question: it searches the indexed material — payer policy PDFs, your SOP library, coding and billing references, onboarding material, equipment manuals — finds the passage that actually answers the question, and returns an answer built from that passage with the document cited. Nobody needs to know which folder something lives in or what the file is called; they ask the question in plain language, the way they would ask a coworker.

Access mirrors whatever permission structure your practice already has. A new front-desk hire sees front-desk material; a question that touches something scoped to billing or compliance only surfaces for someone with that access, the same way it would if they had to go find the physical binder themselves. Every retrieval is logged against the person who asked, so there is a record of who looked up what and when.

The index stays current on a schedule matched to how often the material actually changes — a payer's policy library might sync more often than an SOP that only changes twice a year. When a document is retired or replaced, the old version comes out of the index rather than sitting there waiting to be surfaced as if it were still current.

Three ways this shows up on a normal week

A prior authorization question mid-shift

A scheduling coordinator is booking an MRI for a patient covered by a payer the practice deals with less often. She doesn't remember whether this specific plan requires prior authorization for imaging, and the coworker who usually knows that payer is out. She asks the system directly. It returns the answer from the payer's current policy document — yes, prior authorization is required for this procedure under this plan, along with which form the payer wants and where to submit it — with the policy PDF cited so she can open it herself before she calls the payer. The booking moves forward the same day instead of waiting for a colleague to come back from lunch.

A new front-desk hire's first week

A practice with real front-desk turnover just onboarded someone with no institutional memory of how anything is handled. On her third day, a patient shows up asking to update the insurance on file without her card in hand. Rather than interrupting the office manager mid-task, the new hire asks the system what the practice's own procedure is. It returns the actual SOP — the steps, the form, what to do if the patient can't produce a card that day — cited to the internal policy document, not a guess at what seems reasonable. She follows the real procedure on day three instead of learning it by making a mistake.

A billing question during a coding review

A biller is working through a batch of claims and hits a code she hasn't billed in months, unsure whether the payer's documentation requirements changed since the last time she used it. She asks the system, which surfaces the current billing and coding guidance for that code and that payer, cited to the source bulletin, and flags that the requirement changed at the start of the year. She corrects the claim before submission instead of after a denial comes back weeks later, and doesn't need to pull the coding lead away from her own queue to confirm it.

How it connects to what you already run

The system indexes the documents your practice already has rather than asking anyone to re-enter anything. Payer policy PDFs, prior authorization requirement sheets, SOP documents on a shared drive, coding and billing references, onboarding material, and vendor or equipment manuals are all reasonable sources, wherever they currently live — a shared drive, SharePoint, or a folder on someone's desktop that really should have moved to a shared drive years ago. Most practice management and EHR platforms also expose an API or export that can feed structured reference data into the same index where that's useful, though the administrative material described here typically lives outside the EHR itself.

Because this system touches documents that can contain sensitive material even when none of it is clinical — billing details, vendor contracts, internal policy — it's built with the same discipline any system touching practice data needs: access control scoped to what each person's role actually requires, audit logging detailed enough to show who asked what and when, and data minimization so the system isn't holding more than it needs to answer the questions it's actually asked. Where your practice's arrangement with a vendor requires a business associate agreement, that gets put in place before anything goes live. None of this amounts to a formal "HIPAA certification" — no vendor can hold one, because the obligation sits with the practice as the entity of record — but the access boundaries and logging are built to support the compliance discipline your practice already has to maintain regardless of which tools it uses.

This is one piece of the knowledge systems we build for document-heavy, permission-sensitive parts of a business — the use case for searching an internal knowledge base walks through the mechanics in more general terms, and the glossary entry on AI knowledge bases is worth a look if some of the terminology here is new. It also sits alongside the other systems Calfy builds for healthcare practices — the voice AI that covers your phones solves the same kind of coverage problem for calls that this solves for paperwork, and the full range of AI systems for healthcare covers where each one fits.

Frequently asked questions

Is this a clinical decision support tool?

No. It does not interpret symptoms, does not suggest a diagnosis, and does not recommend a treatment. It answers administrative and operational questions — payer policy, prior authorization requirements, internal SOPs, billing and coding guidance — and every answer cites the document it came from so staff can verify it before acting. A clinical question gets refused, not guessed at.

Is this HIPAA compliant?

There is no formal "HIPAA certification" a vendor can hold — that obligation belongs to the practice as the covered entity, not to a tool. What we build is access control scoped to each role, audit logging detailed enough to show who asked what and when, data minimization, and a business associate agreement wherever your setup requires one.

How does it keep up with payer policies that change often?

The index syncs on a schedule matched to how often each source actually changes — payer policy libraries that update frequently sync more often than an SOP that changes twice a year. Retired or replaced documents come out of the index rather than staying there to be surfaced as if they were still current.

Will it work with the documents we already have?

In most cases, yes. PDFs, shared drives, SharePoint, internal wikis, and exports from a practice management or EHR platform are all reasonable sources. Bring the list of where your policies, SOPs, and billing guidance actually live to the first call — that's usually enough to scope what's realistic.

How much does this cost and how long does it take to go live?

It depends on how many document sources it connects to and how complex the permission model is. We scope every engagement and agree clear pricing before build work starts, and most systems are answering real questions within weeks, starting with the sources causing the most interruptions first.

Bring the payer policy, the SOP, or the onboarding binder your front desk keeps re-explaining from memory. A free 30-minute strategy call is enough to tell you whether a knowledge system fixes it, and roughly what that would take.

See what this looks like for your business

Thirty minutes, your actual workflow, and a straight answer on whether this is worth building.

Free 30 minutes. No pitch deck. You leave with a plan either way.