What is AI prior authorization and how much can it save a multi-site portco?

AI prior authorization uses large language models, OCR, and payer-rule retrieval to read the clinical chart, extract evidence against the payer's medical necessity criteria, and pre-populate the authorization request with cited source notes. Typical savings on a multi-site portco run $200K to $1.2M per year, driven by faster cycle times (Availity reports under 90 seconds on average), higher first-pass approval rates, and 3 to 6 FTE redeployed from authorization queues to higher-value work.

Build versus buy: when should a healthcare-services portco build AI in-house instead of buying?

Buy for clinical documentation, prior authorization, and patient engagement. The category leaders (Abridge, Nuance, Innovaccer, Availity) have payer-rule libraries and EHR connectors that take 18 to 24 months to replicate. Build for revenue cycle anomaly detection, no-show prediction, and scheduling optimization if the portco has a 3-plus person data team on Snowflake or BigQuery. These three use cases run on the portco's own claims and visit data, where in-house wins on cost and customization at scale.

AI Advisory: PE Healthcare Services Portcos

Why AI moves EBITDA in healthcare services.

AI advisory for healthcare services is the practice of giving a private equity operating partner an outside operator who's stress-tested AI vendors across multi-site healthcare portcos and can size which interventions actually move EBITDA, which ones look good in a deck and stall at integration, and what the all-in 24-month TCO really is. It's not a tooling recommendation. It's the read on which two or three deployments will compound at a 20-site portco and which seven won't survive contact with Epic.

Walk a clinic on a Wednesday. The front desk is on hold with a payer trying to verify benefits for a patient who's been sitting in the waiting room for 25 minutes. A medical assistant is chasing a prior authorization that was submitted six days ago and is still "pending." Two clinicians are charting after hours because the EHR template ate 40 minutes of every appointment slot. The billing team is working a denial queue from claims that went out three weeks ago. Every one of those moments was the shape of problem current AI is finally good at. The category leaders (Abridge for ambient documentation, Availity AuthAI for prior auth, Innovaccer for workflow orchestration) have moved from pilot to production at PE-backed multi-site operators in the last 18 months.

Prior authorization is the cleanest example. The Deloitte survey shows 93% of health plan executives expect AI to add value by automating PAs, and the AI build-side has caught up. AI prior authorization reads the chart, extracts evidence against the payer's medical-necessity criteria, and pre-populates the request with cited source notes. Availity reports recommendations rendered in under 90 seconds on average. The operating math: a 20-site portco running 300 PAs per week per site saves 3 to 6 FTE, recovers $200K to $1.2M per year in margin, and lifts first-pass approval rates by 8 to 14 points. None of that is theoretical. New Mountain's Smarter Technologies platform (formed by combining Access Healthcare, Thoughtful.ai, and SmarterDx) is built on exactly this thesis.

Clinical documentation is the next layer. Abridge, Nuance DAX, and Suki have converged on the same product shape: ambient capture, structured note generation, EHR write-back. The honest read is that clinician adoption was the hard part for the first 24 months and is now solved. The harder question for a PE-backed portco is the integration cost. The sticker price ($200 to $500 per clinician per month) misses the change-management budget, the integration partner fees on non-Epic stacks, and the 90-day productivity dip while clinicians adapt. The 24-month all-in TCO on a 50-clinician deployment is closer to $600K to $1.1M than the line-item suggests. The payback still works because each clinician recovers 5 to 8 hours per week, but the math only works if the operating partner runs it honestly upfront.

Five AI use cases moving EBITDA right now.

Pulled from current retainer engagements with multi-site healthcare portcos in the $20M to $300M revenue band across dental, vet, behavioral health, primary care, and women's health. Vendor names are mentioned where the category has converged on a credible build-on-top option. None of these are speculative. All five are in production at multiple PE-backed multi-site operators as of Q2 2026.

Prior authorization agent that submits and tracks across payers.

PA queues swell every quarter and headcount can't keep pace. A medical assistant runs three to four authorizations per hour on a good day. Cycle times of 4 to 9 days are normal. Denials and "more information needed" responses send the same work through twice. At a 20-site portco running 6,000 PAs per week, the labor cost alone is $1.4M to $1.9M per year, before the revenue lost to delayed care.

The fix is an AI agent that reads the chart, extracts evidence against the payer's medical necessity criteria, pre-populates the request with cited source notes, and tracks status across portals. Availity AuthAI and Innovaccer are the category leaders. Recommendations render in under 90 seconds. First-pass approval rates lift 8 to 14 points. The MA keeps the override on every submission.

Sized ROI $200K to $1.2M per year, on a multi-site healthcare portco

Implementation 10 to 14 weeks. First payer live by week 6, fleet of payers by quarter end.

Ambient clinical documentation with HIPAA-compliant transcription.

Clinicians spend 25 to 35% of their day charting. The after-hours work that drives burnout is mostly note completion. EHR templates that were supposed to fix this added structured-data clicks instead. Specialty groups report 5 to 8 hours per clinician per week lost to documentation that doesn't reimburse for a single dollar more.

Ambient AI scribes (Abridge, Nuance DAX, Suki) capture the visit, generate a structured note, and write back to Epic, Athena, eClinicalWorks, or NextGen with the clinician's review. Adoption was the hard part 24 months ago and is now solved. Each clinician recovers 5 to 8 hours per week. The math works if the integration partner cost is sized honestly upfront.

Sized ROI 5 to 8 hours per clinician per week, plus 10 to 15% throughput lift

Implementation 6 to 10 weeks for the pilot cohort. 90-day adaptation curve.

Patient no-show prediction with same-day rebook outreach.

Patient no-shows run 12 to 22% across multi-site operators. Empty chairs in scheduled blocks cost $80 to $300 per slot depending on the specialty. The standard response (overbook the schedule, send a reminder text) leaves money on the table because it treats every patient as equal risk when the data already says they aren't.

A model scoring no-show probability on visit history, time-of-day, weather, distance, and payer mix, then triggering a high-touch outreach for the top-quartile risks and a same-day rebook offer when the cancellation lands, recovers 6 to 15% of lost utilization. Notable Health and Luma Health ship this as a product; the build-from-scratch version is straightforward on the portco's existing visit data.

Sized ROI 6 to 15% utilization lift, typically $40K to $80K per location per month

Implementation 6 to 8 weeks for the model. Rebook workflow takes the longer 30 days.

Revenue cycle anomaly detection for denials and underpayments.

Clean-claim rate is the metric the sponsor never sees until diligence. At most multi-site portcos it sits 8 to 14 points below where the same payer mix supports. Denials are worked reactively, underpayments are caught a quarter late, and the AR team carries the burden of issues that should have been caught at submission.

A model that scores each claim before submission, flags the likely denials, and surfaces underpayment patterns by payer across the portfolio recovers 3 to 7 points of clean-claim rate. AKASA, CorroHealth, and Infinx ship this. New Mountain's Smarter Technologies platform is built on this thesis at scale. In-house builds work if the data team has 90 to 120 days of capacity.

Sized ROI +3 to 7 points clean-claim rate, typically $400K to $2M per year recovered

Implementation 10 to 16 weeks. Pilot at one payer relationship before fleet-wide.

Clinician schedule optimizer balancing acuity and payer mix.

Operating room and clinical schedules are built three to six weeks out and rebalanced manually. The wrong patient in the wrong slot leaves a clinician under-utilized in one block and over-stretched in the next. Payer mix drift across the day is invisible until the month closes. Average utilization across multi-site operators sits 10 to 18 points below what the patient panel actually supports.

A schedule optimizer that balances acuity, payer mix, clinician preference, and patient demand, then surfaces the next-best swap when a cancellation lands, lifts utilization 10 to 15%. The scheduler keeps the override. The model carries the cognitive load. Build-on-top works fine on top of Epic Cadence, Athena, or NextGen scheduling modules.

Sized ROI 10 to 15% utilization lift, plus 4 to 7 points payer-mix margin improvement

Implementation 8 to 12 weeks. Shadow mode 30 days before the scheduler acts on it.

Five questions to ask before approving an AI purchase at a healthcare portco.

The vendor pitch in this category got polished fast in 2024 and 2025. The questions below are the ones the polish doesn't survive. Ask any one of them on a vendor call and the honest answers separate the real solutions from the deck-only ones.

Question 01

How does AI integrate with Epic, Athena, eClinicalWorks, or NextGen EHRs?

Real EHR integration in healthcare services means FHIR-based read/write access to the encounter, problem list, medication history, and claims, validated against the EHR version the portco actually runs. Most vendor decks show an App Orchard logo and imply parity across the stack. The honest answer is usually one of three things: a CSV export, a read-only HL7 feed, or "we're in the validation queue."

Why most vendors get this wrong: they have a working integration with one EHR (typically Epic via App Orchard or Athena via Marketplace) and use that screenshot to imply parity with the legacy stack. They don't have a working integration with eClinicalWorks 11 or NextGen 6.x. They have a roadmap item.

Right answer pattern: a working list of named multi-site customers running on the same EHR version as your portco, plus a specific named integration partner if the work is done by a third party. If the vendor can't name two customers within a phone call, the integration story isn't real yet.

Question 02

What's the HIPAA and BAA reality when your AI vendor trains on patient data?

A signed Business Associate Agreement is table stakes. The question that actually matters is whether patient-derived data flows into a shared model that other customers benefit from. HIPAA permits de-identified data use under Safe Harbor or Expert Determination, and most AI vendors lean on that clause to justify training on the aggregated corpus. For a PE-backed portco being prepared for exit, that's a real competitive asset leaking out.

Why most vendors get this wrong: they conflate "we sign a BAA" with "your data doesn't train our model." Those are different statements. The first is about access and breach liability. The second is about model weights. Many SaaS contracts permit the second under "aggregated and de-identified" clauses without the portco realizing it.

Right answer pattern: a contractual line stating model weights derived from your data stay in your tenant and don't propagate to the shared base model, plus an explicit data-deletion clause on contract termination. If the vendor pushes back with "that's not how we work," the answer is the answer.

Question 03

What's the all-in TCO for AI clinical documentation including change management and integration partner fees?

AI scribe pricing runs $200 to $500 per clinician per month on the sticker. The real TCO on a 50-clinician deployment is $600K to $1.1M over 24 months. Add change management ($30K to $80K for training and adoption), integration partner fees ($50K to $150K on non-Epic stacks), and the 90-day productivity dip while clinicians adapt to new workflows. None of this is in the deck.

Why most vendors get this wrong: the per-clinician SaaS line is the only one with their name on it. They have no commercial reason to surface the integration partner's scope until you've already signed. By that point, the negotiating room is gone.

Right answer pattern: a single TCO worksheet covering the SaaS line, the integration partner line, the internal change-management line, and the realistic ramp curve to full deployment. Ask the vendor to put their name on a 24-month all-in number. If they won't, you don't have a TCO. You have a teaser.

Question 04

What's the regulatory exposure on AI-assisted clinical decisions?

The FDA, OCR, and state medical boards have been quiet so far on AI clinical-decision support, but the silence won't last. CMS has signaled it expects payers and providers to maintain a clear audit trail on AI-assisted prior authorization decisions starting in 2026. The Medicare Advantage prior-auth rules tightened in 2024 and tighten again on January 1, 2027. A vendor whose audit trail is "the model said yes" will not survive the first regulatory inquiry.

Why most vendors get this wrong: their compliance team is staffed for SOC 2 and HIPAA, not for the emerging AI-specific regulatory layer. They don't have a clinician of record on each AI-generated recommendation, and they can't produce the source citations a regulator will eventually ask for.

Right answer pattern: per-recommendation source citation (which note, which lab, which payer rule), human-in-the-loop sign-off on every clinical decision, and a 7-year audit log retained in the customer's tenant. If the vendor can't produce a citation trail for a single past recommendation on demand, they're not ready for the regulation that's coming.

Question 05

Why are we buying this AI tool instead of building it on the data team we already pay for?

Many multi-site healthcare portcos already have a 3-to-5-person data team running on Snowflake or BigQuery, a BI layer, and reasonable engineering bench. For three of the five use cases above (no-show prediction, scheduling optimization, RCM anomaly detection), the in-house build is the right answer if the team has 90 to 120 days of capacity. The vendor is selling time-to-value, not capability the in-house team can't match.

Why most vendors get this wrong: they pitch "healthcare AI is hard, you need us" when the honest answer is "the model side got 10x easier in 18 months, and your existing data team can do three of these five." The vendor sale is a time-to-value sale, not a capability sale, and that changes the negotiation entirely.

Right answer pattern: a build-versus-buy worksheet comparing 24-month TCO of the SaaS path versus a named in-house build, including the opportunity cost of the data team's time. For clinical documentation and prior authorization, buy wins on time-to-value. For no-show prediction, scheduling, and RCM anomaly detection, build is increasingly the right answer at scale.

AI Advisory for Private Equity Portfolios in Healthcare Services.

Why AI moves EBITDA in healthcare services.

Five AI use cases moving EBITDA right now.

Prior authorization agent that submits and tracks across payers.

Ambient clinical documentation with HIPAA-compliant transcription.

Patient no-show prediction with same-day rebook outreach.

Revenue cycle anomaly detection for denials and underpayments.

Clinician schedule optimizer balancing acuity and payer mix.

Five questions to ask before approving an AI purchase at a healthcare portco.

How does AI integrate with Epic, Athena, eClinicalWorks, or NextGen EHRs?

What's the HIPAA and BAA reality when your AI vendor trains on patient data?

What's the all-in TCO for AI clinical documentation including change management and integration partner fees?

What's the regulatory exposure on AI-assisted clinical decisions?

Why are we buying this AI tool instead of building it on the data team we already pay for?

Bring an AI advisor into your next healthcare diligence call.

Why AI moves EBITDA in healthcare services.

Five AI use cases moving EBITDA right now.

Prior authorization agent that submits and tracks across payers.

Ambient clinical documentation with HIPAA-compliant transcription.

Patient no-show prediction with same-day rebook outreach.

Revenue cycle anomaly detection for denials and underpayments.

Clinician schedule optimizer balancing acuity and payer mix.

Five questions to ask before approving an AI purchase at a healthcare portco.

How does AI integrate with Epic, Athena, eClinicalWorks, or NextGen EHRs?

What's the HIPAA and BAA reality when your AI vendor trains on patient data?

What's the all-in TCO for AI clinical documentation including change management and integration partner fees?

What's the regulatory exposure on AI-assisted clinical decisions?

Why are we buying this AI tool instead of building it on the data team we already pay for?

The advisory across other PE sectors.

Bring an AI advisor into your next healthcare diligence call.