Amperos

AI agents perform the main collection tasks, combining payer conversations, portal navigation, document retrieval, appeals, and corrected claim submission. The service is designed to carry information between those steps rather than automate only a phone call. Revenue cycle specialists and customer staff contribute judgment on exceptions, while configured procedures guide routine work. This division of labor makes AI essential to the software's central capacity and workflow proposition.

The conventional reporting and implementation services support that proposition, but removing the agents would remove the primary automated claim handling function offered to healthcare organizations.

Customers can limit the service to chosen claim categories and supply operating procedures and payer priorities. Named deployments route exceptions to experienced customer staff, while revenue cycle specialists support complex work. These are meaningful boundaries on the assigned workflow. The current recovery page nevertheless promises that nothing slows collections, an unqualified reliability statement that exceeds the more limited operational controls described.

Detailed uncertainty thresholds, prohibited claim actions, and required approvals remain incomplete. The oversight model therefore includes human escalation while leaving the advertised limits of autonomous completion and error handling insufficiently reconciled.

The platform uses large language models across voice and computer interaction to carry out collection workflows. Its processor list names OpenAI, Anthropic, Deepgram, and ElevenLabs, providing visibility into possible language and speech services. The current product also describes how payer conversations, portals, appeals, and resubmissions are combined.

Specific production model versions, task assignments among suppliers, training provenance, evaluation thresholds, and change approval practices remain unspecified. This establishes a recognizable technical approach and several suppliers, while leaving the relationship between a deployed release and its measured performance incomplete.

The September 2026 processor list names OpenAI, Anthropic, Deepgram, and ElevenLabs alongside AWS, Azure, communications providers, and other service vendors. This identifies several organizations potentially involved in language inference, speech, hosting, and collection support. The list does not map every provider to a particular workflow or state whether all are active for each customer.

Supplier specific training restrictions, retention periods, processing regions, and downstream contractual obligations remain unspecified. It therefore provides useful partial visibility into the AI supply chain without establishing a complete, current account of how claim and conversation data pass through it.

The September 2026 Providence Anesthesia case reports a 69% lower cost per claim worked relative to an internal biller handling the same volume. Assigned volume increased 50% within two months of the expanded service going live. The case also identifies $12.8 million of accounts receivable worked, which is an inventory measure rather than cash recovered. This is a named, dated operational comparison with a defined cost metric.

Independent evaluation, detailed claim denominators, and comparable current outcomes across customers remain unspecified. The findings support the particular workflow and economic comparison described, without proving general recovery performance.

The privacy policy allows recording and transcription of conversations and treats those materials as nonpersonal once identifying information is removed. It then permits broad use and disclosure of that nonpersonal information. The policy does not define a healthcare deidentification method or establish a specific model training restriction.

Retention limits, deletion of downstream copies, technical safeguards against fabricated claim information, and AI safety incident procedures remain unspecified. Human specialists and customer escalation paths provide operational review, but they do not resolve the data reuse and lifecycle terms for recordings, transcripts, or generated claim documentation.

The company claims HIPAA compliance, and its September 2026 privacy policy applies to both the website and back-office services. That policy names several processing providers and describes encryption and other safeguards. However, much of its personal information language concerns account details, communications, and general service use.

An accessible customer BAA, PHI specific contractual terms, subcontractor business associate obligations, and coverage of individual product or trial arrangements remain unspecified. The compliance claim and processor list therefore provide relevant context without establishing the complete contractual treatment of patient information used in collection work.

The company's own site carries an AICPA SOC assurance badge and HIPAA compliance monitoring language associated with Drata. These indicate an assurance claim, alongside privacy-policy descriptions of encryption, firewalls, and account protections. The audit type, auditor, reporting period, covered systems, and a current retrievable report remain unspecified.

The available claim therefore supports recognition of a security program without establishing the precise scope of its independent assessment. Compliance monitoring and technical controls also serve different purposes from an audit artifact and do not independently verify coverage of each AI provider.

The current product works on insurance collections, denials, payer follow-up, appeals, and corrected claim submissions. Its outputs concern reimbursement administration and the provider's financial workflow rather than diagnosis, treatment selection, or direct patient monitoring. Clinical records may supply information for a claim without making the service a clinical decision tool. No FDA clearance is claimed for these administrative functions. The relevant product scope is therefore revenue cycle work, and the available descriptions do not establish a separate diagnostic or therapeutic intended use for the agents.

Deployment teams translate customer procedures and payer priorities into operating rules, and specialists guide work that requires judgment. Reporting identifies collection trends and denial causes, giving customers a practical way to review workflow performance. These are operational controls rather than a defined framework for governing model behavior.

A current AI oversight structure, formal model release criteria, subgroup performance comparisons, and review materials for customer governance committees remain unspecified. Earlier customer advisory activity does not establish a current model governance process or evidence that outcomes are comparable across claim or patient populations.

The Standard Support Program is incorporated into customer subscription agreements and expressly covers errors and unexpected product behavior. It offers email and established shared-workspace channels, with target initial responses ranging from four business hours for critical service issues to two business days for ordinary issues. This is a concrete customer route for reporting and addressing problems. The targets concern response rather than guaranteed resolution or claim recovery.

Output specific indemnity, compensation for an incorrect action, and reimbursement for lost revenue remain unspecified. Website terms alone do not establish the allocation of liability in a customer's subscription agreement.

The Providence Anesthesia deployment describes production work inside a named practice management workflow. Agents take assigned claims, call payers, navigate portals, retrieve explanations of benefits, submit appeals, and resubmit corrected claims. Actions write back into Imagine, while exceptions route to Providence's team. This links a named customer, named system, and actual operational read and write activity.

The Boulder case adds another production writeback example but leaves its system unnamed. General compatibility references elsewhere are therefore secondary to the concrete Imagine deployment; they do not independently establish the same depth for every listed platform.

The service is delivered through a hosted processing stack, with AWS and Azure among the providers authorized to receive personal information. Customer systems connect to the claim workflow, and implementation teams configure procedures and integrations. These details identify hosting suppliers and operating setup without defining where each type of data is processed.

Exclusive residency, model inference regions, tenant isolation design, backup locations, and private deployment options remain unspecified. A US customer base or named cloud vendor alone does not establish a commitment that claim records and conversation data remain in a particular country.

Purchasing follows a demonstration and commercial agreement. The ROI calculator uses staffing costs, claims volume, average charges, backlog, and current collection rate to estimate potential benefits. Its displayed labor cost and collection-rate projection are inputs or estimates, not a subscription price. Current service fees, any percentage of collections, minimum commitments, and implementation charges remain unspecified. Customer savings stories explain possible value, but they do not establish the amount a buyer would owe for a particular assignment of claims.

Providence Anesthesia and Boulder Care provide named current deployments in anesthesia billing and substance use treatment. The former describes complex reimbursement during a practice management migration, while the latter uses the service for denials and follow-up in a telehealth organization. Additional named customers and specialty offerings span dermatology, eye care, orthopedics, dental care, and other provider groups. These references support several operational settings.

Comparable results by specialty, payer mix, or claim complexity remain unspecified, so one customer's recovery economics cannot establish expected performance for another group's billing rules and reimbursement conditions.