AI Value-Creation Plays for PE Healthcare Services

Provider orgs are administrative-cost businesses wearing a clinical costume

A physician group or RCM company makes money by collecting on care delivered. The leak is everywhere between the visit and the deposit. Initial claim denials hit 11.8% in 2024, and hospitals lose an average of 4.8% of net revenue to denials (HFMA/TechTarget). A single manual prior authorization eats up to 24 minutes of staff time and costs roughly $3.41 versus $0.05 fully electronic — CAQH pegs industry-wide administrative waste at $20 billion (CAQH 2024 Index).

The PE read is simple. You’re not buying clinical edge — you’re buying a labor-arbitrage and leakage-recovery business. AI’s job is to shrink cost-to-collect and stop revenue from walking out the door.

Revenue cycle is the clearest near-term play — the smart money already moved

This isn’t a thesis you have to talk sponsors into; they’ve voted with checks. New Mountain Capital combined three portfolio companies into Smarter Technologies, an AI-RCM platform expected to clear $800M+ in annual revenue (Healthcare Dive). Blackstone is acquiring RCM firm AGS Health from EQT for ~$1.1–1.3B — a company doing ~$60M EBITDA that EQT bought for $320M in 2019 (Becker’s). The value-creation math: buy at 4–7x EBITDA, bolt on AI to cut cost-to-collect, exit at 10–15x (onhealthcare.tech — treat that arbitrage as the sell-side narrative it is).

Autonomous coding is the one play with real, audited numbers

Most AI healthcare claims are mush. Medical coding is the exception — it’s measurable, and the numbers are good. Fathom reports a deployment hitting 95.5% automation at 98.3% accuracy across all service lines, and an average 42.3% reduction in cost-to-code (Business Wire). Call these vendor-reported, but coding has a built-in honesty test: claims either get paid or denied, and auditors check the codes. If a portfolio company codes high-volume, repetitive encounters (radiology, path, EM), this is the first place to look.

Prior auth: real provider ROI, but know which side of the table you’re on

For a provider org, automating PA is a clean cost play. Cohere Health reports ~85% real-time PA determinations, and Geisinger saw a 63% reduction in PA denials after deploying it (IntuitionLabs).

Here’s the Zavient warning. The PA-AI story has two faces. Payer-side denial AI is a regulatory minefield — UnitedHealth’s nH Predict faced allegations of a ~90% error rate, and a 2026 Stanford/Health Affairs study found denied patients win on appeal over 80% of the time but only 0.2% ever appeal (Health Affairs). CMS rules now require explicit reasoning for AI-assisted denials, and several states ban AI as the sole basis for denial. If your portfolio company touches the payer side of PA, the AI is a liability, not an asset. Provider-side automation is the safe, EBITDA-positive play.

Ambient documentation: buy it for retention, not a fake productivity story

Ambient AI scribes (Abridge, Nuance DAX) are the most-hyped play and the one most likely to fool an operating partner. The evidence is mixed: a DAX study showed a 20.4% drop in note time, but a longitudinal NEJM AI study of 112 clinicians found DAX did not make them more efficient as a group (NEJM AI). The real money isn’t time saved — it’s revenue capture: Riverside Health saw an 11% rise in physician wRVUs and a 14% increase in documented HCC diagnoses (PMC review). Buy ambient for retention and accurate acuity capture, not “see more patients” — and watch the upcoding/audit risk.

The unglamorous wins: scheduling, no-shows, and call deflection

The least sexy plays often have the cleanest ROI because the baseline is so bad. AI-driven reminders and no-show prediction cut missed appointments 20–38%; El Rio Health, an Arizona FQHC, reported a 32% no-show reduction and ~$100K/month revenue increase (Healthcare IT News). For behavioral health and home health, every recovered slot is near-pure margin on an already-staffed clinician.

What kills these initiatives

• Buying ambient AI for a productivity story the evidence doesn’t support. NEJM-grade studies show no group efficiency gain.
• Confusing payer-side and provider-side PA AI. Provider-side = savings. Payer-side denial AI = lawsuits, CMS mandates, state bans.
• No baseline metrics before deployment. If you can’t state denial rate, cost-to-collect, and no-show rate today, you can’t prove the AI moved anything at exit.
• EHR integration debt. Tools that don’t sit natively in the EHR die in the workflow.
• Upcoding exposure. Ambient + autonomous coding can inflate acuity — audit and clawback risk, not durable EBITDA.
• HIPAA / data-governance hand-waving. PHI flowing to a vendor model without a BAA is a deal-killer in diligence.