Machinify’s $670 Million Bet: Fix America’s Medical Bills Before They’re Paid

The most important moment in a medical bill is not when an auditor finds an error. It is the instant before a health plan pays it. After that, a neat software problem turns into a scavenger hunt: reopen the claim, justify the finding, contact the provider, recover the money and absorb the irritation. The industry has a suitably bloodless phrase for this - post-pay recovery. In ordinary English, it is pay first, investigate later.

Machinify wants to reverse the verbs. Its software ingests claims, medical records, contracts, eligibility files and payment policies; models look for coding mistakes, pricing anomalies, overlapping insurance and suspicious billing patterns; clinicians, coders and lawyers handle the cases where judgment matters. The aim is to make a defensible decision while the claim is still moving, not months after the cash has gone.

The company got interesting when it got specific

When founder Prasanna Ganesan described Machinify in 2018, the pitch was horizontal: enterprises had oceans of data and very little ability to turn it into useful decisions. Customers would supply raw information and an objective; the platform would learn from past decisions and recommend what to do. Battery Ventures led a $10 million Series A, with GV and Matrix Partners joining.

Healthcare claims were an early example, then became the company. That progression offers a more honest “what changed their mind” than a cinematic pivot story. The public record does not show a single revelation. It shows gravity. Claims had enormous data sets, repetitive decisions, measurable dollars and enough rules to make generic machine learning sweat. In 2023, Machinify sold its prior-authorization application to Evolent. By 2025, payment integrity was the whole billboard.

“AI where it works. Experts where it matters.”Machinify’s operating shorthand

That division of labor is more than a reassuring slogan. A coding edit may be deterministic. A chart review can require clinical interpretation. A subrogation case - deciding whether an auto insurer, employer or other party should pay - may need legal muscle. The combined operation includes clinicians, coders and more than 30 in-house subrogation attorneys. Machinify is not selling a robot that makes medicine disappear into a black box. It is selling a queue in which software handles repeatable work and routes the arguable work to people who can defend it.

That last word matters. Healthcare payment software does not merely need to produce a score; it needs to produce a reason. A provider may challenge a reduced payment. A regulator may ask how a Medicaid determination was reached. A health plan’s own medical director may disagree with the model. The useful output is therefore not “computer says no.” It is the claim, the relevant contract or clinical rule, the evidence in the chart and a traceable route to the decision. Machinify describes its determinations as transparent and auditable. Any buyer should test that claim with an ugly edge case, not a polished demo.

Five products follow the money

Coordinate looks for other coverage and determines which plan is primary. Subrogate identifies claims where someone else may be responsible. Audit reviews charts, itemized bills and diagnosis-related group coding. Pay edits and reprices claims in real time, while also mining paid claims for leakage. Pharmacy handles benefit coordination, compliance and overpayment review for drugs. In 2026, genetic-testing expertise joined the platform, pairing a specialized knowledge base with each payer’s medical policy.

The clever connective tissue is the loop. A post-pay recovery is not merely collected; its pattern can strengthen pre-pay logic. An analyst’s correction becomes training material. A missed opportunity becomes a new detection concept. That is the difference Machinify claims over the traditional stack of fixed edits, separate audit firms and heroic spreadsheets: the parts are supposed to teach one another.

Picture one claim trying to escape

Imagine an inpatient bill that arrives with a diagnosis-related group code implying a more complicated, expensive stay. A basic rules engine can check whether the code exists and whether the fields are complete. The harder question is whether the medical record supports the billed severity. Machinify Audit can flag the claim, retrieve the associated chart and put the relevant evidence in front of a clinical reviewer. If the code is unsupported, the payer can correct it before payment. If the evidence is ambiguous, a person makes the call. The reviewed outcome can then inform the next similar claim.

Now change the problem. The bill is perfectly coded, but the patient had another plan that should have paid first. That is Coordinate. Or the injury came from a car accident and an auto policy may be responsible. That is Subrogate. Or the claim price conflicts with a contract. That is Pay. The products are distinct because the legal and clinical questions are distinct; the platform thesis is that they should share context. One member, one provider and one pattern can appear in several places.

This is where “real time” earns or loses its meaning. A fast false positive is not an improvement. It can delay a legitimate payment, trigger an appeal and send a provider’s billing staff into another portal. The win is high-confidence automation paired with an efficient path for exceptions. Machinify says its models are trained on billions of claims and improve through feedback. Buyers still need to measure precision by claim type, watch overturn rates and ask whether the system reduces total administrative work on both sides of the bill.

A startup became the name on a rollup

Machinify’s scale did not arrive one contract at a time. New Mountain Capital first assembled The Rawlings Group, Apixio’s payment-integrity business and VARIS in 2024. It acquired Machinify in early 2025, put the startup’s name on the combined company and installed its AI platform as the technological center. The announcement described more than 2,000 employees, over 60 health plans and more than $500 million in revenue. David Pierre became chief executive; Ganesan moved from founder-CEO to a senior product role and is now chief technology officer.

Then came Performant Healthcare. Machinify agreed to pay $7.75 a share in cash, roughly $670 million, a 139 percent premium to Performant’s 90-day volume-weighted average. The deal closed in October 2025, adding government-program relationships and more audit and recovery capability. By July 2026, Machinify said it served more than 85 plans representing over 270 million lives.

This is also what could fail first. Five operational histories do not become one product because the website has five tidy verbs. Data models must agree. Client teams need consistent reporting. A payer buying Coordinate should feel the learning arrive in Pay. In a 2025 KLAS First Look, customers praised Machinify’s expertise and responsiveness, and all 14 who answered the repurchase question said they would buy again. Some still worried that leadership changes and a larger company could disturb the intimate vendor relationship they valued. Integration is not back-office housekeeping here. It is the product promise.

The early customer evidence is encouraging but bounded. KLAS interviewed 19 organizations from a company-supplied list. Ten of the 16 respondents with an overall score were highly satisfied, five were satisfied and one was dissatisfied. Respondents pointed to financial savings, lower outside medical expense and less provider abrasion. They also revealed how uneven adoption remained: 77 percent of relevant respondents used the subrogation team and 69 percent used advanced benefit coordination, while only a third used the unified AI decisioning platform. The merger created breadth on paper faster than customers could absorb it. That gap is opportunity and warning in equal measure.

Who pays Machinify - and what it costs

The customer is the payer: national insurers, regional plans, Blues organizations, managed-care plans and government programs. The internal buyer usually runs payment integrity, claims operations, finance or cost containment. Machinify can license software, take contingency fees based on recovered or avoided dollars, provide managed services, or blend the three. Public list prices do not exist. That makes sense for an enterprise system whose fee depends on claim volume, modules, implementation and risk-sharing, but it also means the $670 million headline should never be confused with what a client pays.

Contingency pricing can lower the hurdle because the vendor gets paid when it finds money. It can also reward recoveries after payment, even when prevention would be less disruptive. SaaS licensing reverses the incentive but asks the buyer to believe the savings forecast. A blended contract can reconcile the two if both sides agree on a baseline: which dollars count as avoided, which would have been caught anyway, how overturned findings are treated and who pays for the human review. In this market, measurement design is part of product design.

The alternatives are familiar names such as Cotiviti, Optum, EXL, Gainwell, Lyric and Zelis. The less obvious competitor is inertia: a core claims engine with rules, an internal review team, several recovery vendors and a spreadsheet connecting them. Machinify’s pitch is breadth without black-box outsourcing - use its staff, use your own, or mix both. Its strongest condition is a large payer with enough clean history to learn from, enough leakage to justify integration and enough governance to supervise automated decisions.

A sensible rollout would start narrower than the sales deck. Pick one claim family with a known error pattern, establish a retrospective baseline, run the model beside the existing workflow and review every disagreement. Measure net savings after fees, review expense and appeals. Track how quickly findings become deployable pre-pay logic. Expand only when the reasons survive clinical and provider scrutiny. Machinify offers modular deployment and says outcomes often appear within six months; the valuable part of that claim is modular, not six months. A plan should earn its way from one decision to the next.

The real test is less drama

Payment integrity vendors like to advertise dollars found. The healthier metric may be dollars that never need recovering. Machinify says it supports hundreds of millions of claims a year and generates billions in savings. It has also published narrower examples: more than $70 million identified in one Medicaid genetic-testing program and over $47 million in missed Medicare Advantage premium restored across three plans. Those figures come from the company, and the useful question for any buyer is how the opportunity was measured, validated and sustained.

The company’s latest language shifts from “payment integrity” to “payment intelligence.” Marketing aside, the distinction is sensible. Integrity is an audit after the event. Intelligence is a continuous decision system: read the claim, understand the policy, compare the history, explain the finding, learn from the resolution. If it works, the visible outcome is deliberately boring - fewer corrections, fewer letters, fewer calls, fewer dollars taking the scenic route through American healthcare.

That is a more demanding promise than finding a pile of mistakes. It asks Machinify to make its acquired businesses behave like one system, keep experts genuinely in the loop and prove that earlier intervention creates accuracy rather than indiscriminate denial. The company has bought the reach. Now it has to make the machinery disappear.