NASHVILLE — November 5, 2026 — As "decision intelligence" is applied to everything from reporting dashboards to general-purpose AI, Lyric today released Defining Healthcare Decision Intelligence, a paper that sets the standard for what the term should mean in healthcare, proposing three conditions any system must meet to qualify: specialized healthcare expertise built into the software; recommendations that can be traced to the policy and data behind them; and human accountability for the final determination.
Written by Chief Technology Officer Yevgeny German, Chief Product Officer Brian Berkowitz, and SVP, Engineering Alex Ulyanov, the paper argues that healthcare decision intelligence requires more than sophisticated technology and is defined by healthcare depth: clinical, payment, and policy expertise encoded in the software and maintained as evidence and regulation change. That depth, which the authors argue general-purpose technology cannot build quickly, is what allows AI to augment human decision-making, taking on the work around a decision while a person remains accountable for the determination.
The stakes are concrete. In fiscal 2025, CMS estimated that $28.83 billion in Medicare fee-for-service payments were paid improperly, a figure CMS notes is not a measure of fraud. Unlike a credit decision, the paper notes, a wrong healthcare decision reaches an individual directly: a patient waiting on care, a clinician acting on incomplete information, or a provider relationship built over years.
What is healthcare decision intelligence?
The paper defines decision intelligence as the discipline of engineering how decisions get made, then measuring what they produce. Systems built this way combine analytics, business rules, and AI to support decisions inside the workflow where they occur, then feed outcomes back into the logic that produced them.
Its defining property is timing. Work performed before a decision is analysis. Work performed after is reporting and recovery. Decision intelligence operates while the decision is still open and can still be changed. Healthcare decision intelligence applies that discipline to high-stakes healthcare decisions, where policy is complex, accountability cannot transfer to a model, and decisions must remain traceable to the policy in effect when they were made.
Why did the authors write the paper?
The authors wrote the paper because the term has outrun its meaning. When a category lacks boundaries, health plans cannot compare what they are being offered, organizations cannot tell whether an investment moved them forward, and the conversation stalls at capability claims. In a field where decisions affect patients, providers, and plans directly, the authors argue the bar should be higher, and it should be shared.
"Healthcare holds clinicians, coding, and payment policy to rigorous standards. The systems now shaping those decisions deserve the same," said German. "We wrote this paper to raise that bar. A demanding, shared definition protects health plans, providers, and patients, and it gives everyone building in this space something common and real to be measured against."
A standard built around three conditions
The paper offers health plans a pragmatic way to evaluate any system that claims the healthcare decision intelligence label:
- The expertise is in the software. Does the system encode clinical, payment, and policy expertise, and keep it current as evidence and requirements change?
- Traceable at decision speed. Does the system deliver each recommendation while a decision can still be changed, and trace it to the policy, data, and logic in effect when it was made?
- Human judgment. Does the organization operating the system configure its rules and thresholds, with a person responsible for decisions about care, coverage, and payment?
The authors call that traceable line, from a recommendation back to the policy, data, and reasoning that produced it, the decision's lineage, and argue it has to travel with the recommendation rather than sit in a document somewhere.
Applied honestly, the authors write, the standard excludes much of what is currently described as decision intelligence. Dashboards offered as the decision mechanism fall short on timing. Automation that cannot identify which policy version applied falls short on traceability. General-purpose AI without encoded healthcare expertise falls short on expertise.
Automate the work. Keep people accountable for decisions.
The authors draw a distinction between the work surrounding a healthcare decision and the determination itself.
Finding records, assembling evidence, checking which policy applies, routing cases, and identifying discrepancies are often time-consuming tasks well suited to automation. Deciding what those findings mean for an individual's care, coverage, or payment is different.
"Much of an expert's day is spent gathering and assembling information, not deciding," said Ulyanov. "That's where AI creates the most value, bringing thousands of unstructured elements into well-defined structured data that can be presented for expert review. The determination itself belongs with someone who can be accountable for it, keeping the human in the loop in all of this."
Under the proposed standard, AI can help experts build and maintain decision logic, track policy changes, summarize complex records, and prioritize cases that need attention. The logic governing an outcome remains deterministic, tested, and versioned before deployment. Human oversight is designed into the workflow, not added as a final approval step after an automated process has already framed the decision.
Because probabilistic models introduce a kind of error that deterministic rules do not, the authors argue that error is tolerable in the work around a decision and not in the determination itself.
In a complex review, for example, AI can summarize the record and pinpoint where a chart and a claim imply different diagnosis groupings. The reviewer still decides whether the grouping is wrong. The model shortens the path to that judgment without taking it.
The line the authors draw tracks with regulators' direction. In February 2024, CMS told Medicare Advantage plans that an algorithm may assist a coverage determination but cannot replace an assessment of the individual patient's circumstances, and since January 2025, California has required a licensed physician or qualified healthcare provider to decide any denial or modification of care based on medical necessity.
"Oversight placed at the end of an automated process is oversight in name only," said Berkowitz. "By the time a case reaches a person, the framing is already set. Designing for human judgment means cases that need judgment arrive with the evidence and rationale assembled."
Built on decades of healthcare expertise
The standard draws on more than 35 years of Lyric's clinical, payment, and policy expertise. More than 100 health plan clients, including nine of the top 10 U.S. health plans, use Lyric's healthcare decision intelligence technology to validate more than 10 billion claims annually. Lyric's technology processes 2 million transactions per hour with 99.9%+ uptime. Those clients have realized more than $20 billion in annual savings, and 60% have worked with Lyric for a median of 24 years. Lyric42, Lyric's healthcare decision intelligence platform, is designed around the principles the paper describes, augmenting human decision-making rather than replacing it: health plans configure their own rules and thresholds, see the decision lineage behind each recommendation, and make the final determination. The platform combines policy, data, deterministic rules, responsible AI, workflow intelligence, and clinical and payment expertise.
"In healthcare, AI is only as good as the expertise behind it," said Halsey Wise, Lyric CEO. "Our clients answer for their decisions, so each one needs to come with its decision lineage: the policy, the evidence, and the reasoning behind it. That depth is what more than 100 health plan clients rely on, and it's what our technology team builds into Lyric42 every day."
What can healthcare learn from the industries that went first?
The paper traces decision intelligence from Britain's prewar air-defense research through its maturation in credit and fraud, supply chain planning, and insurance underwriting, and draws lessons for healthcare:
Depth beats breadth. None of the early adopters succeeded by applying general-purpose technology to a specialized problem. They built depth in one domain and maintained it as the domain evolved.
Timing is the defining property. Decision intelligence operates while the decision is still open, not before it as analysis or after it as reporting and recovery.
The rules are falling behind the medicine. Clinical evidence moves faster than the policy and coding systems that govern care and payment. Securing a new Category I CPT code can take 18 to 24 months after an application is filed.
Governance accelerates. Organizations that treat governance as a constraint move slowly, because every expansion reopens unsettled questions. Organizations that embed it move faster, and the authors argue they will define the category.
The paper identifies payment accuracy as healthcare's first proving ground for decision intelligence, given its high transaction volumes, standardized data, codified reimbursement policies, and measurable outcomes. Over time, the authors expect the same principles to extend across the healthcare ecosystem, to decisions made before, during, and after care, as the technology and underlying data mature.
What to watch in 2027
The authors expect four shifts as healthcare decision intelligence matures:
- AI agents take on the work, not the decision. Expect agentic AI to handle more of the retrieval, assembly, and routing behind healthcare decisions, while people make the final determinations.
- Traceability becomes a requirement. As regulators and buyers raise expectations for AI governance, organizations will need to show how decisions were made and what they were based on.
- Decisions move earlier. Intelligence shifts from catching problems after the fact to acting while decisions are still open, and eventually to anticipating them.
- The standard spreads across healthcare. Payment accuracy is proving out first, with decisions made before, during, and after care to follow.
The full paper, Defining Healthcare Decision Intelligence, includes the history of the discipline, case studies from credit, supply chain, and insurance, the regulatory landscape, and the complete standard. The State of Healthcare Decision Intelligence in Payment Accuracy applies the standard function by function across payment accuracy.
About the authors
Yevgeny German is Chief Technology Officer at Lyric.
Brian Berkowitz is Chief Product Officer at Lyric.
Alex Ulyanov is SVP, Engineering at Lyric, where he leads the engineering organization behind Lyric42 and Studio42.
About Lyric
Lyric is the trusted leader in healthcare decision intelligence. Built on more than 35 years of proven results, Lyric combines responsible AI with clinical, payment, regulatory and policy expertise to deliver transparent, auditable insights in milliseconds across real-time claims workflows. Lyric augments human decision-making so health plans can make faster, more accurate payments while maintaining full control. Today, Lyric supports 200 million lives, with nine of the top 10 U.S. health plans relying on its platform to reduce waste, improve efficiency, and support accurate payments. To learn more, visit lyric.ai.
Media Contact
Neeha Curtis, Chief Corporate Affairs Executive
