Part 1 ended on the real obstacle, and it has two heads. To move upstream and prevent denials, a revenue cycle team needs automation that learns. This is because payer rules shift, denial patterns drift, and a model that is accurate at go-live goes stale within months. Yet the thing leaders fear most about automation is "errors scaling faster than humans can catch." So the business ask sounds almost contradictory: I want my agents to improve, but I need to trust how they improve. Most AI tools force a choice between the two. Aivar's built to refuse it.
A word on shape, because it is the whole point. The revenue cycle is awash in AI tools right now. Most of them are point solutions that automate one task and ask the buyer to trust a black box. Aivar is built differently: not a product suite, but an operating layer. Three product accelerators —Convogent (conversational AI), Velogent (agentic process automation), and Kubogent (AIOps) — run on two foundations: ReVAct (trust for agents) for governance and Aiva (AI that compounds) for learning. Every agent on the platform sits on top of both, be it Aivar's or the customer's. That structure is what lets the system learn and stay trustworthy at the same time.
ReVAct is the trust layer. It governs every action an agent takes, in two modes. Before an action executes, it reasons, validates, and only then acts — synchronous gating that stops a bad decision before it touches a claim. After execution, it runs a full-trajectory analysis to catch drift, the slow and silent wandering that turns a once-reliable agent into a liability. That combination is the direct answer to "errors scaling faster than humans can catch." Most tools offer neither; the ones that come close offer logs after the fact, which is not the same as gating before. Pre-action gating and post-execution drift detection together are what make it safe to let automation run unattended where a single wrong code carries financial and compliance weight. It also reflects where the responsible-AI frameworks aired at AC26 were already pointing — continuous monitoring, accountability, and traceability. [6]
Aiva is the learning layer. Every execution leaves a decision trace. Every time a person steps in to correct an agent, that correction becomes ground-truth signal. Over time those traces build a context graph that compounds. The system grows sharper with every claim it processes, instead of staying frozen at its launch-day accuracy.In revenue cycle terms, that is the difference between a tool that knows last quarter's payer rules and one that keeps pace as those rules change. It is a learning layer, not a lookup table.
Together, ReVAct and Aiva answer both halves of that contradictory demand: the agents improve, and the way they improve is governed and visible. That pairing — not any single model — is the foundation everything else stands on.
On top of those foundations sit the accelerators that do the revenue cycle's work. Velogent handles the workflows — the rules-heavy, judgment-dense automation where most of the loss is created. Every decision it makes is gated by ReVAct, every step is auditable, and every execution feeds Aiva, so the workflow gets smarter over time.
Part 1 showed that roughly 90% of denials are preventable, with nearly half originating at the front end. [2] That is Velogent's target, and the mapping is direct:
• Denial prevention — check eligibility, benefits, authorization, and coding against payer requirements before submission, catching the breakdowns that become denials instead of appealing them later.
• Appeals, where prevention falls short — assemble and draft the appeal from the record. With close to 70% of denials, and more than 80% of appealed prior-authorization denials, eventually overturned [3][4], the recoverable money is real; the constraint has always been the labor.
• Prior authorization workflows — track requirements, prepare submissions, and surface the cases most at risk, exactly as CMS's new timelines and FHIR APIs reshape authorization through 2026 and 2027. [5]
• Coding and documentation integrity — flag the gaps between what was documented and what was coded before they cost a claim. And because Velogent's agents reason rather than follow brittle scripts, they do not break the moment a payer changes a form, which was the failure mode that has defined a decade of rules-based automation.
The other half of the revenue cycle are the conversations — with payers and patients — and that is Convogent's job i.e. production conversational AI across voice, chat, email, and in-app, in 25-plus languages, at sub-500-millisecond latency.
• The payer interface — orchestrate status checks, eligibility, and authorization across voice, portal, and API, routing each request to the fastest channel and keeping the phone call as the last resort, not the first.
• The patient financial experience — handle billing questions, payment options, and reminders in plain, multilingual conversation, the experience that increasingly determines whether patients trust and stay with a provider.
• Intelligent deflection — resolve the routine inquiries without a human, so staff time goes to the conversations that genuinely need judgment.
For a US provider, multilingual reach is not a nicety. It is the difference between serving a patient population and talking past part of it.
None of this matters if it cannot run safely in production. Kubogent is the Kubernetes-native platform underneath — training, deploying, and operating the AI at scale on EKS or self-hosted kubernetes natively, with audit trails and role-based access built in and GPU costs cut by 30 to 40%.
For healthcare, the deployment model is the headline. Convogent, Velogent, and the foundations run inside the customer's own cloud account. The data never leaves, and the intellectual property that gets built can be owned by the provider rather than rented back from a vendor. In a domain governed by HIPAA and by hard-won caution about where patient data lives, that is often the deciding factor.
Most health systems will blend all three, and Part 1's source survey found vendor relationships already turning "functional but increasingly complex." [1] The cleanest principle the leaders offered was to let the business objective decide, and to tie everything to one roadmap rather than a scatter of pilots. [1] A governed operating engine is built for exactly that: the speed and proven engineering of a partner, with the control, transparency, and ownership a team would otherwise have to build itself — and a single foundation that governs not only Aivar's agents but the customer's own.
The proof was on the AC26 stage. Intermountain Health described standing up a cross-functional team — analytics, coding, clinical documentation, data science, appeals, and digital technology working as one — around a custom denials tool, and reported multimillion-dollar recoveries across coding corrections and recurring denial patterns. Lake Regional, a rural system, showed the other end of the spectrum: roughly $4.2 million in front-end denial prevention in under six months, daily cash lifted from about $750,000 to more than $1.2 million within a year, and an operating margin that flipped from red to black — a 248% improvement year over year. (Both were presented at HFMA AC26; they are neither our case studies, nor guarantees.)
Neither result came from buying a tool. Both came from changing the operating model and putting governance and learning at the center — which is the whole argument of this series.
Prevention is the strategy. The hard part is doing it with automation that both improves and can be trusted to improve — and that is exactly the trade-off Aivar's architecture is built to resolve. ReVAct governs. Aiva learns. Velogent runs the workflows and Convogent runs the conversations on top of both, inside the provider's own cloud, on a platform — Kubogent — built to scale. For US revenue cycle teams ready to move upstream, that is a foundation built for the revenue cycle the data is already pointing toward.
To see how it maps to your denials, your payers, and your patients, get in touch for a demo.
1. HFMA — "The Revenue Cycle of the Future" report and "4 shifts that define the revenue cycle of the future" (April 2026; survey n=95). Source of the "errors scaling faster than humans can catch" concern, the "functional but increasingly complex" vendor finding, and the "let the business objective decide / build the right foundation before scaling" principle.https://www.hfma.org/revenue-cycle/ai-revenue-cycle-transformation/
2. HFMA / Conifer Health Solutions —"Preventing denials before they happen: How revenue intelligence is reshaping the revenue cycle" (June 2026).https://www.hfma.org/revenue-cycle/denials-management/preventing-denials-before-they-happen-how-revenue-intelligence-is-reshaping-the-revenue-cycle/
3. Premier, Inc. — "Claims Adjudication Costs Providers $25.7 Billion..." (February 2025); denial overturn rates.https://premierinc.com/newsroom/policy/claims-adjudication-costs-providers-257-billion-18-billion-is-potentially-unnecessary-expense
4. KFF — "Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024" (January 2026); 80.7% of appealed PA denials overturned.https://www.kff.org/medicare/medicare-advantage-insurers-made-nearly-53-million-prior-authorization-determinations-in-2024/
5. Centers for Medicare & Medicaid Services — "CMS Interoperability and Prior Authorization Final Rule(CMS-0057-F)" Fact Sheet (January 2024).https://www.cms.gov/newsroom/fact-sheets/cms-interoperability-and-prior-authorization-final-rule-cms-0057-f
6. Tetteh HA, Blagogee T, Robbins D —"Impact of AI on Access to Care: A Three-Pronged Approach for Enhancing Access to Surgical Care," The American Surgeon (2025). DOI:10.1177/00031348251329495. https://doi.org/10.1177/00031348251329495
Product architecture and capability details (the Governed Agentic OS, ReVAct, Aiva, Convogent, Velogent, Kubogent) are from Aivar's internal product documentation.Case studies referenced (Intermountain Health; Lake Regional) were presented atHFMA Annual Conference 2026 and are not independently published figures.
.svg)
.svg)
.svg)
.svg)
