AI prior authorization and insurance verification

Started by mille125
This forum made possible through the generous support of SDN members, donors, and sponsors. Thank you.
Get help with your application

Use all the free resources available to you from SDN: articles, guides, expert advising, forums discussions, and school research.

Browse articles Compare schools My application plan Read Interview Feedback
Practice for interviews Ask an Advisor
M

mille125

Full Member
15+ Year Member
Advertisement - Members don't see this ad
I am now very happy with my billing and collections but I think that my office is not optimal on prior authorization and insurance verification. I am still seeing a lot of undercollection of patient payments. For a recent procedure the EOB shows a patient owing $1350 but we only collected $825 because that is what our estimations showed.

In todays world there are quite a few AI platforms for insurance verification, patient authorizations, or both. I am going to demo droidal this week.

Does anyone here use and AI platform that you like for prior auths (which are more numerous now) and verification of benefits?
 
mille125 said:
I am now very happy with my billing and collections but I think that my office is not optimal on prior authorization and insurance verification. I am still seeing a lot of undercollection of patient payments. For a recent procedure the EOB shows a patient owing $1350 but we only collected $825 because that is what our estimations showed.

In todays world there are quite a few AI platforms for insurance verification, patient authorizations, or both. I am going to demo droidal this week.

Does anyone here use and AI platform that you like for prior auths (which are more numerous now) and verification of benefits?
Click to expand...
anyone use phreesia for insurance verification and other stuff. I am in the process of getting a demo
 
Advertisement - Members don't see this ad
mille125 said:
I used it in my old company and I found it to be underwhelming
Click to expand...
how so? I am looking to use it for insurance verification and benefits check instead of my staff calling or using availity. how much were you paying them?
 
Historically I have had my billing company do the verification and prior auths.

If its for scs, minute man, mild, pns, si fusion...usually the vendor will do this however your staff has to upload items into their portal
 
jsaul said:
how so? I am looking to use it for insurance verification and benefits check instead of my staff calling or using availity. how much were you paying them?
Click to expand...
It was a long time ago like 2014 or 2015. I dont remember what was paid. I just remember that the insurance verification was spotty. Maybe it has changed.
 
PinchandBurn said:
Historically I have had my billing company do the verification and prior auths.

If its for scs, minute man, mild, pns, si fusion...usually the vendor will do this however your staff has to upload items into their portal
Click to expand...

Prior auths haven't been much of a problem even though that take awhile. Insurance verification even using the EMR tools has not been great in my mind even though it might be average compared to other practices.

My understanding is that many of the new AI insurance verification and prior auth platforms can rapidly decrease time for prior auths from minutes to seconds and can be much more accurate with insurance benefits. This frees staff for other things. In the future I want my staff to focus on customer service and face to face patient needs and I want artificial intelligence to do paperwork and authorization related tasks.

I was wondering if anyone is using any of the AI platforms for insurance benefits. If you are not, do you think that your staff is getting the copays/deductibles right upfront at the time of service instead of billing them later? I only know because I have been looking at EOBS and comparing it to the amount collected and I have seen disparities. Most are getting billed the difference and we get paid later but I dont think that is efficient. This is happening about 20% of the time. Maybe that is industry standard but if it is I think AI can help us do better.
 
I've made a lot of progress with my AI scribe in the last two months. The big gains have come since Claude Code gained the ability to easily manage database updates, code accurately, test its own code and then update in a self-contained feedback loop. I've been able to build much more complex tools.

Just yesterday I added a "Prompt Builder" function that ingests raw auth criteria (like the PDF put out by Carelon), and it extracts, processes, and generates LLM friendly prompts that can be used to generate an audit-defense ready addendum to your notes in real time. And I do mean in real time- I'm presently working on a loop process that reviews visit context, then transcript chunks while the visit is in progress, then gives you realtime feedback on what criteria has been met/unmet/not yet addressed/NA. That way you can "talk" your way around the red lights to make them green.

I'll probably be doing alpha testing soon. It's got HIPAA compliance, user registration/authentication from the git go. Right now it's a mac os based swift project, but eventually will transition to a web app.

As for PA, I still have my remote admin doing these manually. I'd be curious to hear what an AI could do and at what cost.
 
powermd said:
I've made a lot of progress with my AI scribe in the last two months. The big gains have come since Claude Code gained the ability to easily manage database updates, code accurately, test its own code and then update in a self-contained feedback loop. I've been able to build much more complex tools.

Just yesterday I added a "Prompt Builder" function that ingests raw auth criteria (like the PDF put out by Carelon), and it extracts, processes, and generates LLM friendly prompts that can be used to generate an audit-defense ready addendum to your notes in real time. And I do mean in real time- I'm presently working on a loop process that reviews visit context, then transcript chunks while the visit is in progress, then gives you realtime feedback on what criteria has been met/unmet/not yet addressed/NA. That way you can "talk" your way around the red lights to make them green.

I'll probably be doing alpha testing soon. It's got HIPAA compliance, user registration/authentication from the git go. Right now it's a mac os based swift project, but eventually will transition to a web app.

As for PA, I still have my remote admin doing these manually. I'd be curious to hear what an AI could do and at what cost.
Click to expand...
Very interested in a few projects like you’ve described but have been stymied by my own lack of coding knowledge. Do you already know how to code, or has Claude been able to guide you through the whole process?

When you say in real time, did you also build a bidirectional interface between your program and your EMR? I’ve daydreamed about having something exactly as you describe but have no idea how to pull it off in a smooth fashion.
 
callmeanesthesia said:
Very interested in a few projects like you’ve described but have been stymied by my own lack of coding knowledge. Do you already know how to code, or has Claude been able to guide you through the whole process?

When you say in real time, did you also build a bidirectional interface between your program and your EMR? I’ve daydreamed about having something exactly as you describe but have no idea how to pull it off in a smooth fashion.
Click to expand...

I was never trained to code, but I picked up a lot when trying to reverse-engineer a script I paid a guy to write for me a few years ago and with LLM assistance my ability to adapt code expanded dramatically. Now that the LLMs can write code reliably, it's been another exponential expansion. So when I converse with GPT about something I want to build, it's with an understanding of how code works, capabilities, limitations, lingo, etc. As for the actual writing, there's almost no point anymore. Claude or Codex will do it better, faster, and more accurately. That said- experience helps a lot. You start to learn the ongoing limitations and account for them when planning. For example, say you build a window that has a function that generates some text for you. You assume that there will be a button somewhere to save, close, clear, whatever. Or that if you click off that window and back on, your text will still be there. Then you test it out and see you needed to specify things like that. With time you start to integrate gotcha items into your planning.

So my process is to have a discussion with GPT about what I want to build, go back and forth for a while. Make a phased plan. Then go to Codex and have it formalize the plan and make it relevant to my existing codebase (if it exists). Then rewrite the plan. Then I might have Claude implement it, and have Codex check the work. That process works incredibly well if you really want to get it done. I'm actually teaching this to my kids now and even got one of them their own subscription to Claude Code. She's working on an app that helps you write a book.

Now... EHR integration. That comes later. It will be easier to integrate once the product is fully grown. Athena will want to see a fully fleshed out app before I can plug into their API. Right now you can cut and paste context into a box, or use the customized context extraction system (add your files, it will extract facts, store them, then you can call up relevant context anytime you write a note at the press of a button). Then you scribe-- that's the point where the procedure auth loop engages. You tell it what procedure you need an audit defense for, then it outputs your progress to a separate window you can monitor while the visit is ongoing. Haven't tested it out yet, but this should be working soon.

This is an enormous amount of work, and I had to go through 3 simpler versions of my app to arrive at the big boy I'm building now. Start with small projects, learn, and then build bigger ones.
 
Last edited:
powermd said:
I've made a lot of progress with my AI scribe in the last two months. The big gains have come since Claude Code gained the ability to easily manage database updates, code accurately, test its own code and then update in a self-contained feedback loop. I've been able to build much more complex tools.

Just yesterday I added a "Prompt Builder" function that ingests raw auth criteria (like the PDF put out by Carelon), and it extracts, processes, and generates LLM friendly prompts that can be used to generate an audit-defense ready addendum to your notes in real time. And I do mean in real time- I'm presently working on a loop process that reviews visit context, then transcript chunks while the visit is in progress, then gives you realtime feedback on what criteria has been met/unmet/not yet addressed/NA. That way you can "talk" your way around the red lights to make them green.

I'll probably be doing alpha testing soon. It's got HIPAA compliance, user registration/authentication from the git go. Right now it's a mac os based swift project, but eventually will transition to a web app.

As for PA, I still have my remote admin doing these manually. I'd be curious to hear what an AI could do and at what cost.
Click to expand...

My main issue is undercollecting copays because the estimation of benefits yields a number smaller than what pops up on the EOB. I am lucky that my patients are paying after they are billed but obviously it is most efficient to collect 100% of this upfront. I only realized this because I initiated a self audit about 3 weeks ago and I found that 20% of the time the estimation was wrong and many times it was $300-$500 off.

Does your AI do anything related to that?

I am demoing droidal tomorrow and I will post my thoughts.
 
mille125 said:
My main issue is undercollecting copays because the estimation of benefits yields a number smaller than what pops up on the EOB. I am lucky that my patients are paying after they are billed but obviously it is most efficient to collect 100% of this upfront. I only realized this because I initiated a self audit about 3 weeks ago and I found that 20% of the time the estimation was wrong and many times it was $300-$500 off.

Does your AI do anything related to that?

I am demoing droidal tomorrow and I will post my thoughts.
Click to expand...

Not on my radar because Athena's eligibility check gives me accurate copays and coinsurance on almost everyone, plus I insist on CCOF for everyone who will ever need to self-pay anything. Can't make it past my front desk with out that card to catch all incidentals. No copay today? Reschedule-- and pay the late r/s fee. All patients are told they will need a credit card to check in before they are scheduled the first time. Occasionally I get folks who insist on cash, but if there's coinsurance --> card, or go somewhere else and pay those awesome ASC/HOPD fees.

My AI would need to hook into the EHR billing system to do what you're saying. We would need to find a way to gather all the relevant data electronically via APIs. If the EHR already has the data, or it's available through an easily accessible API, no problem (ha ha... just hours of work figuring out how to talk to all these things).
 
powermd said:
Not on my radar because Athena's eligibility check gives me accurate copays and coinsurance on almost everyone, plus I insist on CCOF for everyone who will ever need to self-pay anything. Can't make it past my front desk with out that card to catch all incidentals. No copay today? Reschedule-- and pay the late r/s fee. All patients are told they will need a credit card to check in before they are scheduled the first time. Occasionally I get folks who insist on cash, but if there's coinsurance --> card, or go somewhere else and pay those awesome ASC/HOPD fees.

My AI would need to hook into the EHR billing system to do what you're saying. We would need to find a way to gather all the relevant data electronically via APIs. If the EHR already has the data, or it's available through an easily accessible API, no problem (ha ha... just hours of work figuring out how to talk to all these things).
Click to expand...
Maybe Athena is better at it than ECW. My office does the same. I have very good collectors up front but if they have the wrong info it is less effective. I never thought that it was a problem until I started looking at EOBs and comparing them to what was collected based on the eligibility check.

There was much more data in Jan and Feb because a lot of people haven't met deductibles.
 
You guys are better than me! I do have them collect at the front. But sometimes that's awkward for the front staff. I usually bill them later. I know we SHOULD collect upfront as once they leave you never know...

no doubt I have been burned a few times. but most people pay . I think it's a win some lose some strategy....
 
powermd said:
I was never trained to code, but I picked up a lot when trying to reverse-engineer a script I paid a guy to write for me a few years ago and with LLM assistance my ability to adapt code expanded dramatically. Now that the LLMs can write code reliably, it's been another exponential expansion. So when I converse with GPT about something I want to build, it's with an understanding of how code works, capabilities, limitations, lingo, etc. As for the actual writing, there's almost no point anymore. Claude or Codex will do it better, faster, and more accurately. That said- experience helps a lot. You start to learn the ongoing limitations and account for them when planning. For example, say you build a window that has a function that generates some text for you. You assume that there will be a button somewhere to save, close, clear, whatever. Or that if you click off that window and back on, your text will still be there. Then you test it out and see you needed to specify things like that. With time you start to integrate gotcha items into your planning.

So my process is to have a discussion with GPT about what I want to build, go back and forth for a while. Make a phased plan. Then go to Codex and have it formalize the plan and make it relevant to my existing codebase (if it exists). Then rewrite the plan. Then I might have Claude implement it, and have Codex check the work. That process works incredibly well if you really want to get it done. I'm actually teaching this to my kids now and even got one of them their own subscription to Claude Code. She's working on an app that helps you write a book.

Now... EHR integration. That comes later. It will be easier to integrate once the product is fully grown. Athena will want to see a fully fleshed out app before I can plug into their API. Right now you can cut and paste context into a box, or use the customized context extraction system (add your files, it will extract facts, store them, then you can call up relevant context anytime you write a note at the press of a button). Then you scribe-- that's the point where the procedure auth loop engages. You tell it what procedure you need an audit defense for, then it outputs your progress to a separate window you can monitor while the visit is ongoing. Haven't tested it out yet, but this should be working soon.

This is an enormous amount of work, and I had to go through 3 simpler versions of my app to arrive at the big boy I'm building now. Start with small projects, learn, and then build bigger ones.
Click to expand...
This sounds amazing, great work! I have been playing around with something similar myself. Let me know if I can help at all, I have a busy practice that utilizes athena, happy to beta test for you or share notes if that would help.
 
Advertisement - Members don't see this ad
powermd said:
Not on my radar because Athena's eligibility check gives me accurate copays and coinsurance on almost everyone, plus I insist on CCOF for everyone who will ever need to self-pay anything. Can't make it past my front desk with out that card to catch all incidentals. No copay today? Reschedule-- and pay the late r/s fee. All patients are told they will need a credit card to check in before they are scheduled the first time. Occasionally I get folks who insist on cash, but if there's coinsurance --> card, or go somewhere else and pay those awesome ASC/HOPD fees.

My AI would need to hook into the EHR billing system to do what you're saying. We would need to find a way to gather all the relevant data electronically via APIs. If the EHR already has the data, or it's available through an easily accessible API, no problem (ha ha... just hours of work figuring out how to talk to all these things).
Click to expand...
Sounds complicated
Yes how do we all beta test this
 
I did the droidal demo.

It is so good that is is scary. AI actually goes thru our records and even referral records to do a task. If a procedure is denied because of lack of 6 weeks of conservative therapy, it actually goes in to the chart notes and finds this info in all of the notes and does it in seconds.

Price is expensive. I was quoted just over $4000 per month. That is too high for me but if the cost comes down, I would definitely sign on.
 
mille125 said:
I did the droidal demo.

It is so good that is is scary. AI actually goes thru our records and even referral records to do a task. If a procedure is denied because of lack of 6 weeks of conservative therapy, it actually goes in to the chart notes and finds this info in all of the notes and does it in seconds.

Price is expensive. I was quoted just over $4000 per month. That is too high for me but if the cost comes down, I would definitely sign on.
Click to expand...

Oh for F's sake... This is basically Claude Code attached to your EHR. Do you have a recording of their presentation?
 
powermd said:
Oh for F's sake... This is basically Claude Code attached to your EHR. Do you have a recording of their presentation?
Click to expand...
I am ignorant of Claude Code but I thought it was mainly for coding.....which I dont need. I code all of my visits and I had a coding audit where they only questioned 3 of 100 records.

If Claude Code is inexpensive, I might be interested.
 
I ran my scribe all day today in the office- it's way better now than ever before. Notes done before the next patient starts. I found some things to tighten. The HPI needs some beefing up. I ran my "loop PA" scanner-- that definitely needs work. It posted in its window line by line the auth criteria in excruciating detail and none went green. Have to simplify this.

Would you guys on Eclinicalworks (or other EHRs) be willing to share a few cut and paste items with me? You could obscure PHI with placeholders, of course. Specifically I'm interested in a rough cut and paste of your day's schedule, and another of a patient registration (that has all demos, insurance, etc). I want to make sure my data capture/parse can handle other popular EHRs.
 
Between fomo and feeling like the proverbial lamplighter u guys are making me question my practice mechanics

It’s a good thing. Thanks.
 
The "running loop" idea is challenging. It's one thing to prepare an audit defence worthy summary from a completed note, but from the raw transcript it's a lot to ask. I'll probably look for some shortcuts next just to prompt the user with criteria on demand, and leave the audit defence as a post processing step. It doesn't take long, and my note generator is pretty fast now.

What othe rdocumentation would you like to see automated? The app has a "document studio" where you can put together a variety of practice documents. I'm working on an M1 generator now. It picks up the patient's WC info from the database (which you captured already. with a simple cut and paste from your ehr), then you select the visit note of interest to base it on.. out pops your M1. Different states may have slightly different formats, so I may need to add the capability to adapt it.

I know the PCP community wants FMLA paperwork automated, the problem there is the forms can vary from employer to employer.
 
mille125 said:
I did the droidal demo.

It is so good that is is scary. AI actually goes thru our records and even referral records to do a task. If a procedure is denied because of lack of 6 weeks of conservative therapy, it actually goes in to the chart notes and finds this info in all of the notes and does it in seconds.

Price is expensive. I was quoted just over $4000 per month. That is too high for me but if the cost comes down, I would definitely sign on.
Click to expand...

powermd said:
Oh for F's sake... This is basically Claude Code attached to your EHR. Do you have a recording of their presentation?
Click to expand...
I just paid for the Pro plan.
I saw their page about Prior Auths and using LCDs, but I am not sure how to do that? I asked CoWork to create a PA checker tool, but it requires me to paste the note each time.

I havent used Code yet since it requires Git and I gotta figure out wtf that is and then Ill go from there.
I use eCW so itll be interesting.
 
Status Sciaticus said:
I just paid for the Pro plan.
I saw their page about Prior Auths and using LCDs, but I am not sure how to do that? I asked CoWork to create a PA checker tool, but it requires me to paste the note each time.

I havent used Code yet since it requires Git and I gotta figure out wtf that is and then Ill go from there.
I use eCW so itll be interesting.
Click to expand...
Pro Plan for droidal?
 
For those of you wondering WTF Claude Code can do that's not "write me code", today I fed it copies of my Quickbooks transactions, as well as PDFs from my bank and credit card statements. I told it to extract the transactions from the PDFs to a single CSV, then reconcile against my QB output, and put the unreconciled transactions in a separate file. No problemo. Now- the reconciliation is turning out to be enough of a PITA that I just hired a VA to do it, but at least I was able to hand her all the raw material she needs.

On the Scribe front, I'm adding my voice agent from a different project. This is a conversational task management tool. Next feature of the app is ability to run your PA generator against your note, and if it flags any missing data, it can then search older context and visit notes to find what's needed and add it to your note such that the PA justification will now pass and append to your note.
 
Advertisement - Members don't see this ad
powermd said:
For those of you wondering WTF Claude Code can do that's not "write me code", today I fed it copies of my Quickbooks transactions, as well as PDFs from my bank and credit card statements. I told it to extract the transactions from the PDFs to a single CSV, then reconcile against my QB output, and put the unreconciled transactions in a separate file. No problemo. Now- the reconciliation is turning out to be enough of a PITA that I just hired a VA to do it, but at least I was able to hand her all the raw material she needs.
Click to expand...
Does this data stay local with Claude Code or does Claude upload it to it's cloud? Does it hold onto it for a specified time and wipe from memory?
 
Screenshot 2026-04-01 081129.png
Screenshot 2026-04-01 081623.png
Screenshot 2026-04-01 081706.png






Denied. Self pay. I do not use AI.
 
callmeanesthesia said:
It’s because you did not explicitly state there is no evidence of radiculopathy. I’d suggest having a template or dot phrase saying “patient meets the following:” then lists the general insurance medical necessity criteria as a bulleted list.
Click to expand...
agreed
 
You must log in or register to reply here.

Similar threads

T
insurance saying - No Authorization Required - then deny
Replies
28
Views
5K
Ducttape
D
S
Procedure prior authorization time?
Replies
31
Views
4K
pmrmd
P
G
Performing Different Type of Epidural Than Initial Authorization
Replies
9
Views
997
No_Pain_No_Gain
No_Pain_No_Gain
hyperalgesia
Could AI replace all administrators?
Replies
28
Views
3K
hyperalgesia
hyperalgesia
I
AI scribe?
Replies
30
Views
7K
DOctorJay
DOctorJay