The Short Answer
When a private equity firm puts AI agents into a healthcare portfolio company, the data comes first: Partners Group's published AI result was built on a clean proprietary dataset, and KPMG lists fragmented, unstructured data among the main bottlenecks. On ibl.ai you own all the code and the data, so the agents and the cleaned data stay assets of the company at exit.
This is a practical note for operating partners, portfolio CFOs and healthcare CIOs. It uses the deals and figures sponsors have actually published through October 2026, and it is explicit about which numbers come from healthcare and which do not.
What have private equity firms actually done with AI agents in healthcare companies?
The visible activity is concentrated in revenue cycle management, where the work is high-volume, rule-bound and measured in dollars. Two sponsor-backed companies show the pattern.
R1. Funds affiliated with TowerBrook Capital Partners and Clayton, Dubilier & Rice completed their acquisition of R1 on November 19, 2024, at $14.30 per share in cash and a value of about $8.9 billion.
On October 14, 2025, R1 announced it would acquire Phare Health, which builds AI-native inpatient coding and clinical documentation improvement, and it completed the deal on October 22.
At the announcement, R1 said its R37 lab already had agentic applications in production with select clients.
R1 reported autonomous coding accuracy as high as 97% for service lines like emergency room and physician office visits. That figure is R1's own claim about its best service lines, not an independent audit.
Smarter Technologies. On May 19, 2025, New Mountain Capital, a firm with more than $55 billion under management, combined three of its portfolio companies into one revenue cycle platform built around virtual agents.
The three were Access Healthcare, an operations company founded in 2011, SmarterDx, a clinical AI company for revenue integrity, and Thoughtful.ai, a revenue cycle automation platform.
At formation, the combined company served more than 200 clients, including more than 60 hospitals and health systems.
| Company | Sponsor | Move | What it pairs the AI with |
|---|---|---|---|
| R1 | TowerBrook, CD&R (closed Nov 2024, ~$8.9B) | Acquired Phare Health, Oct 2025 | An existing revenue cycle operation and its client data |
| Smarter Technologies | New Mountain Capital (formed May 2025) | Merged three portfolio companies | An operations business founded in 2011, plus clinical data AI |
| Foundation Risk Partners (insurance, not healthcare) | Partners Group (result published Aug 2026) | Paired with Version 1, another portfolio company | A clean, proprietary policy-level dataset |
The common thread is in the last column. Neither healthcare sponsor deployed AI on its own. Each put the AI next to an operation that already held the workflow and the data it runs on.
How much EBITDA do AI agents actually add at a PE portfolio company?
The most specific sponsor-published result we found is modest, credible and from insurance rather than healthcare. Read it as what one well-run program disclosed, not as an industry benchmark.
On August 3, 2026, Partners Group reported a 120 bps EBITDA margin uplift at Foundation Risk Partners, a US insurance brokerage with over 3,000 employees across 68 locations, worth USD 10 million of EBITDA impact.
The release says one agent cut the average new-client policy processing cycle by 94%, which led to a doubling in close rates. The doubling is close rates. EBITDA moved 120 basis points.
The program ran with a seven-person team augmented by Version 1, and the first use case went from concept to production in 14 weeks.
KPMG's September 2026 keynote interview in Private Equity International estimates that AI-driven efficiencies could deliver a five to 15 percent EBITA improvement, and upwards of 20 percent on direct headcount-related costs.
KPMG adds that those gains can become savings or fuel for growth, and calls the work a two-year journey rather than a sprint. Note the metric: EBITA, before amortization, not EBITDA.
Why does the data have to come before the AI agents in healthcare?
Because the agents can only be as good as what they read. Partners Group's own release puts the dataset first: FRP "built a clean, proprietary policy-level dataset that laid the foundation," which Version 1's agents were then built on.
Head of Private Equity Wolf Scheider credited the pairing with Version 1 and described the result as having "turned a clean dataset into USD 10 million EBITDA uplift."
KPMG's James Gardner says it directly: applications are limited by how good the data is. Paul Pan lists a data bottleneck around fragmented and unstructured data sets among the implementation challenges, alongside security and governance and a shortage of AI-fluent talent.
Healthcare makes every one of those harder. The clinical note is unstructured, payer rules differ by contract, and the records are protected health information, whose use and disclosure HIPAA restricts, and which a vendor can only process under a business associate agreement.
Phare Health's own description is instructive: it builds a view of the patient journey from unstructured and structured data so coding decisions carry a full evidence trail.
The evidence trail is a data product before it is a model output.
The clinical-record version of this problem is in Healthcare AI's Bottleneck Was Never the Model.
The same gap shows up outside healthcare. Kaufman Rossin's mid-market survey found 94% of companies using generative AI and 2% running it at scale, which we covered in 94% of the Mid-Market Uses GenAI. 2% Have Scaled It.
Who owns the AI agents and the cleaned data when a PE firm exits?
Whoever the contracts say, and that is decided at deployment, not at exit. KPMG notes that leading firms are now shaping AI exit stories so the business earns a value premium at sale.
An exit story is only as good as what transfers. A cleaned dataset, the agent definitions, the evaluation sets and the prompts that encode payer rules are real assets if the portfolio company holds them.
If they live in a vendor's tenancy under a subscription, the buyer inherits a contract and a renewal date. That is an operating expense in the data room, not intellectual property.
Pricing shape matters for the same reason. Per-seat licensing scales with headcount, which is the one line an AI program is meant to decouple from growth, so it is the wrong shape for this work at portfolio scale.
At FRP's 3,000 employees, Microsoft 365 Copilot at its published $30.00 per user/month, paid yearly is $1.08 million a year before a single agent touches a policy. Usage-based or owned infrastructure bills for the work done instead.
What should an operating partner ask before putting AI agents into a healthcare portfolio company?
Five questions settle most of the risk, and each one maps to a published finding above rather than a vendor's pitch.
| Question | Why it matters |
|---|---|
| Where will the cleaned dataset live? | It is the input Partners Group credits for the $10M result |
| Who processes the PHI, under what agreement? | HIPAA decides who may touch the data before any model does |
| Can we change the model without rebuilding? | A two-year program will outlive the model it started on |
| Is the target margin in bps, not pilots launched? | KPMG describes the shift from 20 board-facing pilots to EBITDA discipline |
| Does it transfer at exit? | Owned code and data are assets; a subscription is a renewal date |
There is a sixth question for any agent that writes into a billing or clinical system: who approves the write. Single-turn accuracy does not answer it, as covered in Hospital AI Aces Single-Turn Tests. Grade the Actions.
The payer side is already pushing back: the Blue Cross Blue Shield Association estimated about $942 million in added coding-intensity costs from 2023 to 2025, without proving AI caused it, the subject of Who Audits the Coder?.
Where does ibl.ai fit in a PE-backed healthcare company?
On ibl.ai you own all the code and the data. The platform, Agentic OS, runs inside the portfolio company's own perimeter, model-agnostic across any LLM, with no per-seat pricing, so you can deploy anywhere from a private cloud to a fully air-gapped network.
That is the shape the published results point to. The dataset, the agent definitions and the evaluation sets stay with the company, transfer at exit, and keep working when the model underneath changes.
Our forward-deployed engineers do the data-first work with your team: the data work, the payer rules and the review gates, not just the model wiring. Healthcare deployments are described on Medical and Healthcare.
1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.
Want an AI program the portfolio company owns at exit?
We deploy agents as source code the company keeps, with the data and evaluations in its own systems, in its cloud, on-premise, or fully air-gapped.
Book a 30-minute demo or talk to the ibl.ai team. ibl.ai is family-owned and operated from New York, NY.
Sources: R1 take-private terms from R1's November 19, 2024 completion release filed with the SEC; Phare Health, the R37 lab and the 97% coding accuracy figure from HIT Consultant's report of R1's October 14, 2025 announcement; Smarter Technologies from Healthcare Innovation, May 19, 2025; Foundation Risk Partners figures and quotes from Partners Group's August 3, 2026 release; KPMG's ranges and quotes from Making the AI roll-out count, Private Equity International, September 2026; Copilot pricing from Microsoft.