Ramya Ganti is the founder and CEO of Oprox, a VC-backed AI-native prior authorization platform for healthcare providers.
Prior authorization tools were built for a version of medicine that behavioral health does not resemble.
In procedural care, the request often centers on a fixed question: Does the imaging meet criteria? Has conservative therapy failed before a joint replacement? Does the diagnosis support the intervention? There is still friction, but the logic is usually tied to a code, a guideline and a narrow threshold.
Behavioral health is different. Medical necessity in applied behavior analysis, substance use disorder treatment, outpatient therapy and psychiatric medication management is longitudinal, subjective and context-dependent. It sits within symptom trajectories, functional impairment, risk assessments, recovery readiness and level-of-care judgments. A request can be clinically sound and still fail if the story is not told in the language a payer expects.
That is why prior authorization automation keeps breaking down in behavioral health. The problem is not simply that the forms are slow. The problem is that the decision itself is a judgment call.
Across my career in healthcare technology building automation that affects millions of dollars in provider revenue, one lesson has remained consistent: The hardest part of prior authorization is not moving information between systems; it is understanding how the decision gets made. In most of medicine, that understanding can be codified into criteria. In behavioral health, it stays stubbornly human, which is exactly why the tools that worked elsewhere stall here.
The stakes are clinical, not administrative. In the American Medical Association’s 2024 prior authorization survey, 94% of physicians said prior authorization delays access to necessary care. And when researchers ranked the toll by specialty, psychiatry, behavioral health and substance use treatment came out as the field physicians believed prior authorization harmed most. These are not delays in elective convenience. They are delays for severely depressed patients, children with autism and people in the brief window when they are ready to begin addiction treatment.
Each behavioral health modality also has its own authorization logic.
ABA for autism may be approved in blocks of therapy hours and reassessed periodically. Providers must justify not only whether treatment is needed, but why a particular intensity should continue. Substance use disorder treatment runs on a different clock: When someone is willing to enter treatment, delay is not neutral. Outpatient therapy can involve recurring reviews of whether continued sessions remain medically necessary. Psychiatric medication management adds formularies, step therapy, dosage rules and plan-specific documentation requirements.
One generic workflow cannot capture all these modalities, timelines and evidentiary standards. In behavioral health, the hard part is building the clinical argument. Does the documentation show why partial hospitalization is more appropriate than intensive outpatient treatment? Do symptom scores and functional examples support continued ABA hours? Does the chart clearly explain relapse risk, safety concerns, family instability or failed lower levels of care?
This is where horizontal prior authorization platforms can reach their limit. A generic tool may pull information from the chart, populate a form, submit it and track the status. That helps when retrieval is the bottleneck. In behavioral health, interpretation is often the bottleneck. A system that cannot distinguish an ABA reassessment from a substance use level-of-care review, an outpatient concurrent review or a medication exception may simply move the wrong information faster.
Technology, however, is only one part of the answer. Healthcare leaders should evaluate the full operating model around authorization.
First, improve documentation at the source. Clinicians should not be expected to write for payers, but organizations can create structured prompts, templates and training that capture the evidence reviewers repeatedly look for: functional impairment, treatment response, safety risk, failed alternatives and the rationale for the requested level of care. Better documentation reduces avoidable denials before automation is introduced.
Second, map workflows by modality and payer. Leaders should identify where requests stall, which information is repeatedly missing, how often peer-to-peer reviews occur and which denials are overturned on appeal. This creates a baseline for deciding whether the real problem is documentation, policy interpretation, staffing, handoffs or technology.
Third, build payer-specific operational intelligence. Provider organizations cannot assume payers will clarify inconsistent requirements or create a collaborative feedback loop. Instead, leaders should systematically analyze denial reasons, authorization outcomes, appeal decisions, policy updates and reviewer behavior across plans. The goal is to identify which documentation elements, clinical arguments and submission patterns consistently affect approval.
Fourth, choose technology and AI vendors with genuine behavioral health expertise. A vendor should understand that prior authorization is not a single submission event. The workflow can include an initial authorization, recurring concurrent reviews, requests for additional information, peer-to-peer discussions, denials and appeals. Each stage requires different evidence and a different clinical narrative.
Healthcare leaders should, therefore, evaluate whether a vendor can support the full authorization life cycle, interpret payer-specific requirements, identify weaknesses in the clinical record before submission and help teams construct stronger medical necessity arguments. The system should also preserve the reasoning behind its recommendations, maintain an auditable record and escalate ambiguous cases to experienced staff.
AI can assist by summarizing records, identifying documentation gaps and drafting authorization or appeal narratives. But it should not be treated as an independent clinical decision-maker. Its output is only as reliable as the documentation, payer policies and behavioral health expertise underlying it. Without appropriate oversight, AI can overlook context, overstate unsupported conclusions or apply criteria that are outdated or inappropriate for the patient’s level of care. Privacy, security, bias monitoring, clinical review and clear accountability must therefore be built into the implementation from the beginning.
In behavioral health, speed without judgment is not enough. A faster form filler that cannot reason through subjective, payer-specific medical necessity may only create a faster path to another denial.
The organizations that solve this problem will recognize the problem as a specialized clinical, operational and technology challenge spanning initial authorization, concurrent review and appeals. The real measure of success is not how quickly a request is submitted, but whether each request presents the evidence, context and clinical rationale needed to withstand payer review and protect timely access to care.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?