Adonis

Detection and execution are both model driven: predicting which claims will be denied before submission from historical denial patterns and payer behaviour, clustering denials to find systemic causes, and then autonomously navigating payer portals, drafting appeal letters from unstructured clinical notes, and progressing claims.

The company describes moving past descriptive analytics dashboards to a continuous loop of signal detection, prioritization, and agentic execution, which is the substantive difference from reporting tools in this category.

This is the most autonomous posture among the provider side revenue cycle vendors indexed here, and the disclosure does not match it. Agents are described as autonomously progressing claims to resolution, logging into payer portals, and drafting, packaging, and filing appeal letters, with more than 70 percent of standard claim statusing and appeals routines fully automated.

An appeal letter is a formal representation to a payer assembled from a patient's clinical record, and no review step, confidence threshold, or escalation criterion was retrieved for the filing decision. The platform states appeals are auditable, which helps after the fact, but auditability is not oversight. Buyers should establish in contracting what is filed without human sign off.

Nothing substantive is published about the models. No architecture, no base model, no accuracy or precision figures, no confidence threshold, and no evaluation methodology.

The available descriptions are positioning rather than disclosure. A proprietary Intelligence platform, advanced agentic AI, and agents that leverage the right context and capabilities describe intent, not mechanism.

The published numbers are operational outcomes: more than 70 percent of standard claim statusing and appeals tracking automated, and a reported 20 to 30 percent reduction in long term revenue cycle operating spend. Customer results are specific and creditable, including nearly $200,000 in denials recouped at one orthopaedics practice and close to $500,000 in underpayments surfaced at a gastroenterology group within five months. Those measure money found, not how often the system was right, and a system can recover real money while still erring often in either direction.

For a platform whose agents act without human intervention, the absent number that matters most is how often an agent's action is subsequently corrected or reversed.

Controls are named on a public privacy and security page rather than behind a sales conversation, which is more than several peers offer: encryption in transit and at rest, role based access control, regular audits, vulnerability assessments and penetration testing, and mandatory employee security and privacy training, with a privacy policy stating how protected information is collected and used.

The data reach is deeper than the category norm and worth naming, because agents extract unstructured clinical notes and patient charts directly from the record system, which is full clinical documentation rather than claims metadata, and carry it into a vendor environment for synthesis. Three gaps hold the grade.

No retention or deletion schedule is published for that extracted clinical material, which has no ongoing purpose once a claim resolves and is therefore the clearest candidate in this segment for a defined expiry. No sub processor list. And no statement on whether customer data trains or improves the models, which for a platform learning payer behaviour patterns across many provider customers is a question a buyer should not have to ask, since the cross customer learning is the product.

One claim is unverifiable as written: the company states that throughout its assessment history zero high risks have been identified, and without the assessor, the scope or the number of assessments that has no denominator. Ask for retention on extracted clinical material, the sub processor list, and the training position.

Named references are strong for a company founded in 2022: Mount Sinai Health System uses the platform to identify and prioritize billing exceptions, with its system vice president of revenue cycle quoted directly, and Fox Valley Orthopedics reported recovering nearly $200,000 in denials. Commercial metrics are unusually concrete, with more than 4x revenue growth in 2025 and net retention above 130 percent, the latter being a genuine signal since it means existing customers expanded. Held back from A because the 20 to 30 percent operating spend reduction and 70 percent automation figures lack published baselines.

The data reach is deep. Agents extract unstructured clinical notes and patient charts directly from the electronic health record, which is full clinical documentation rather than claims metadata, and they carry it into a vendor environment for synthesis.

Controls are named on a public privacy and security page, which is more than several peers offer: encryption in transit and at rest, role based access control restricting access to authorised personnel, regular audits, vulnerability assessments and penetration testing, and mandatory employee security and HIPAA training covering PHI protocols. A privacy policy states how protected health information is collected and used.

Three gaps hold the grade at C. No retention or deletion schedule is published for the clinical material extracted, which has no ongoing purpose once a claim resolves. No subprocessor list. No statement on whether customer data is used to train or improve the models, which for a platform learning payer behaviour patterns across many provider customers is a question a buyer should not have to ask.

One claim is unverifiable as written. The company states that throughout its assessment history zero high risks have been identified. Without the assessor, the scope or the number of assessments, that has no denominator.

Adonis is a business associate. It processes clinical documentation and claims data on behalf of provider organisations, so the status follows from the work.

What supports it publicly is thin. A privacy policy addressing how protected health information is collected and used, mandatory employee HIPAA training, and a statement that the company follows HIPAA compliance measures. No externally validated framework sits underneath any of it, which distinguishes this record from peers whose HIPAA position is backed by a HITRUST certification mapping onto the Security Rule.

No business associate agreement or its terms are published, the company does not state its role in those words, and no review or evaluation cadence is given, so none of the three routes to a higher grade in this index is taken.

One architectural point deserves attention in the agreement rather than on a web page. Where agents authenticate into payer portals on the provider's behalf, the flow of protected health information runs outward through channels the provider's own business associate agreement may not have contemplated. Establish what the agreement says about automated third party portal access and about the credentials used to perform it.

Real controls are described and no certification is actually claimed, which is an unusual combination and the reason for the grade.

The privacy and security page names substantive practice: encryption in transit and at rest, role based access control, regular audits, vulnerability assessments and penetration testing, internal risk assessment and mitigation, and recurring employee security and HIPAA training. That is more operational detail than some certified vendors publish.

But the compliance statement is the weakest formulation this index has encountered. The company says it upholds standards by following SOC2, HIPAA and CCPA compliance measures. Following compliance measures is not holding a report. No type is given, no auditor is named, no report is offered under non disclosure, and there is no trust centre. No HITRUST or ISO 27001 is claimed.

The contrast with the company's own published material is difficult to miss. Adonis publishes buyer guidance for health systems evaluating revenue cycle technology which explains SOC 2 Type 2 specifically, describes it as the framework enabling third party auditors to review a vendor's environment, and advises prioritising vendors that meet it. Its own page does not state that it holds one.

Absence of a claim is not proof that no report exists, and a buyer should simply ask for it. But a vendor that wrote the guide should be the easiest in the category to answer this question about.

The FDA has no jurisdiction here and, unlike credentialing or payer side utilization management, no accreditor or regulator examines provider side revenue cycle vendors at all. The regulatory weight falls on the customer, since the health system signs and submits the claim and carries False Claims Act exposure for what it contains.

This vendor carries a second exposure most of its peers do not, and it comes directly from the architecture. Adonis AI agents log into third party payer portals and, in the company's own description, mimic manual human clicks to run medical necessity and authorization checks. Automated access of that kind is commonly restricted or prohibited by the portal operator's own terms of use, and the credentials being used belong to the provider rather than to Adonis.

No certification addresses this. A buyer should establish, in writing, which payer portals are accessed by automation, whether that access is permitted under each payer's terms, whose credentials are used, and who bears the consequence if a payer treats it as a violation and suspends access.

The agents also draft and file appeal letters, which are governed by payer contracts and, for Medicare Advantage, by CMS appeal rules. Those obligations sit with the provider too.

The autonomy claim is strong and the oversight description is thin, which is the wrong way round for this workflow.

The company states that its agents monitor, alert, recommend and deploy fixes without any human intervention, and reports automating more than 70 percent of standard claim statusing and appeals tracking. Appeal letters are described as compliant and auditable, and an audit trail is claimed.

What is not published is any review step, escalation threshold, or confidence rule governing when an agent acts alone and when a person is brought in.

The specific risk is worth stating plainly. An appeal letter is a representation to a payer about a patient's clinical circumstances, assembled here by extracting unstructured clinical notes and synthesising them into a document that is then filed on the provider's behalf. If the extraction misreads the record or the synthesis overstates it, an inaccurate clinical claim has been submitted to a payer under the provider's name. Auditable means the action can be reconstructed afterwards; it does not mean anything checked it beforehand.

No bias evaluation, accuracy measurement, or error rate is published.

Nothing substantive is published about the models: no architecture, base model, accuracy or precision figures, confidence threshold, evaluation methodology or warranty, indemnity or remediation commitment. The available descriptions are positioning rather than disclosure, since a proprietary intelligence platform, advanced agentic capability, and agents that leverage the right context describe intent rather than mechanism.

The published numbers are operational outcomes, covering the share of claim statusing and appeals tracking automated and a reported reduction in revenue cycle operating spend, alongside specific and creditable customer results describing sums recouped at named practice types.

Those measure money found, not how often the system was right, and the distinction is not pedantic: a system can recover real money while erring often in either direction, because a recovered underpayment is visible and counted while an appeal pursued on a wrong basis, a claim statused incorrectly, or a recoverable amount never identified generate no artefact at all.

For a platform whose agents act without human intervention, the absent number that matters most is how often an agent's action is subsequently corrected or reversed, because that is the only measure that captures the errors the money figures cannot. Ask for the reversal and correction rate on autonomous actions, the confidence threshold governing when an agent acts rather than escalates, and what an agent does when a payer response is ambiguous.

Positioned deliberately as an overlay rather than a replacement system of record, available through the Epic Connection Hub, and reaching in both directions: extracting unstructured clinical notes and charts from the EHR, and logging into third party payer portals to execute checks. Reaching into payer portals is the harder half, since those interfaces are outside the customer's control and change without notice.

Cloud platform positioned as an intelligence overlay on top of existing systems of record rather than a replacement, which lowers the integration burden and is the right architecture for the problem. API access is authenticated through OAuth 2.0 and OpenID Connect. Availability through the Epic Connection Hub reduces integration friction for health systems already on that platform.

Beyond that the axis is unanswered. No hosting provider is named, no region or data residency statement is published, and there is no subprocessor list.

The gap matters more here than the equivalent gap would at a pure analytics vendor. This platform extracts unstructured clinical notes and patient charts out of the electronic health record and moves them into a vendor environment to synthesise appeal documents, so the question of where that material comes to rest, and for how long, is central rather than incidental.

No published pricing and no published pricing mechanism. Contracts are negotiated privately.

What circulates publicly is speculation rather than information, and it is worth naming because it looks authoritative. One widely shared investment analysis constructs an entire contract value model for Adonis, then states plainly that the company likely operates a hybrid of base platform fee plus success based pricing on recovered revenue. Likely is the operative word. The same piece asserts revenue and margin figures no outside party could know about a private company. A competitor's comparison page describes custom enterprise pricing with per agent packaging and outcome based contracts, which comes from an interested party and is not corroborated.

None of that is vendor disclosure and none belongs in a budget.

Two questions are worth putting directly, because the answers change the risk profile. If any portion is success based on recovered revenue, establish the rate and what counts as recovered. If any portion is priced per agent, establish what constitutes an agent and what happens to the price as the vendor consolidates work into fewer, more capable ones.

The absence stands out against the company's own publishing habit. Adonis produces an annual revenue cycle benchmark report drawing on responses from more than 120 healthcare leaders, and it publishes buyer guidance on evaluating revenue cycle technology. It discloses a good deal about the market and very little about itself.

Clearly bounded to provider side revenue cycle with an explicit orientation toward payer dynamics: denials, underpayments, aged accounts receivable, prior authorization status, and appeals. The founder's stated background building technology for health insurers informs that positioning, which the company frames as levelling the playing field with payers.