Evolent Health (EVH) AI Prior Authorization: Up to 20-Point Gain

Evolent Health (EVH) is using AI prior authorization review at scale. On its second-quarter earnings call on August 6, 2026, the company said its AI platform now evaluates more than one-third of the prior authorization volume that previously required manual clinical review, and that auto-approval rates have improved by up to 20 percentage points among customers where the models are deployed[1]. The 20-point figure is an upper bound for some customers, and the company has not disclosed a cost savings amount.

Where AI sits in Evolent's business

Evolent Health (EVH) manages specialty care for health insurers, including oncology, cardiology, musculoskeletal care and imaging. Its revenue comes mainly from two types of contracts. In one, it charges a service fee per member for technology and services such as authorization review. In the other, it takes on the risk of specialty medical costs within an agreed scope.

Before a physician orders a specialty test or treatment for a patient, Evolent reviews the request on the insurer's behalf to check whether it meets clinical criteria. This step is called prior authorization, and clinical reviewers used to handle it request by request. Auth Intelligence is the AI platform Evolent uses at this step. Its technology comes from Machinify, which Evolent acquired in 2024: an AI model evaluates the request first, and requests that qualify are approved automatically. Company policy is that AI is used only to speed up the process or to approve requests, and any recommendation that changes a treatment plan must be made by a clinician[1]. Prior authorization is Evolent's day-to-day core business, so the platform sits in core operations.

From test markets to scale in 18 months

The application took a year and a half to move from test markets to scale. On the February 20, 2025 call, the company said it had integrated the acquired Machinify authorization assets into its own platform, rebranded it Auth Intelligence and gone live in its first test markets. It gave no quantified results at that time[2]. In August 2025, the technology was integrated into a number of workflows and the company gave its first efficiency reading[3]. In February 2026, the yardstick shifted from review efficiency to auto-authorization rates, with figures drawn from focused pilot areas in the imaging business[4]. In May 2026, a new generation of AI models was rolling out faster in imaging[5].

On August 6, 2026, management described the second quarter as the turning point at which this work moved past pilots and reached scale. The platform's coverage had a reading for the first time, and management previewed a heavy deployment in the first quarter of 2027 alongside an important customer renewal[1].

What the latest auto-approval numbers measure

This disclosure was the first to answer both how large the improvement is and how much volume is covered. Among customers where the models are deployed, auto-approval rates improved by up to 20 percentage points, and management said clinical quality did not decline. Requests approved by AI are completed in minutes rather than days[1]. On coverage, more than one-third of the authorization volume that previously required manual clinical review is now evaluated by the platform[1].

Both figures carry limits. The 20 percentage points is an upper bound among some customers, and the company did not disclose the number, size or baseline of those customers. The denominator for "more than one-third" is the authorization volume that previously required manual review. The company's long-term goal of automatically approving 80% of authorization volume uses all authorization volume as its denominator, so the two cannot be compared directly[1].

Readings from earlier quarters also used different yardsticks: review efficiency improved by roughly 11%[3]; the real-time auto-authorization rate rose by over 11 points for chest CT scans and by 16 points for cervical spine MRI[4]; and the new models produced auto-approval increases in the high teens on the cases they evaluated[5]. These figures do not form a single quarter-by-quarter upward curve.

The financial link: adjusted EBITDA margin, not yet quantified

The financial metric this application is most likely to affect is adjusted EBITDA margin, through clinical review labor costs within cost of revenue. Auto-approved requests no longer take up clinical reviewers' time. Reviewers shift to complex cases that require a conversation with the treating physician, and the company does not have to add staff in proportion as volume grows.

On the investment side there is one explicit figure. In February 2025, management expected net implementation costs for its AI automation work to be a drag on 2025 adjusted EBITDA of approximately $10 million. That cost belongs to the whole automation program and is not borne by this platform alone[2]. On the return side, there are only qualitative statements so far: management said results are at the high end of its expectations and called the platform a key element in reaching its long-term margin targets[1]. The company has not attributed any revenue, cost or EBITDA amount to Auth Intelligence, so changes in margin or outlook cannot yet be attributed to this platform.

What is confirmed and what is still missing

What can be confirmed today is an operating change: AI has taken over a substantial share of the clinical review that people used to do, and the wait for approved requests is noticeably shorter. Whether the cost reduction has been realized still lacks one number, namely the review cost savings, or the change in cost per review, that this platform delivers on its own. The period after the heavy first-quarter 2027 deployment tied to the important customer renewal is when the company is most likely to give that reading.

Application assessment

• Authorization Review Copilot | Business position: core operations | Application stage: limited production | Coverage: single business unit | Value type: cost reduction

Sources

[1] Drillr · Evolent Health, Inc. (EVH) · 2026-08-06 · Earnings call

Quote: Among customers where these models have been deployed, we are seeing auto approval rate improvements of up to 20 percentage points.

[2] Drillr · Evolent Health, Inc. (EVH) · 2025-02-20 · Earnings call

Quote: While meaningfully accretive to 2026 and beyond, we do expect net implementation costs for this AI-based automation work to be a drag on 2025 adjusted EBITDA of approximately $10 million and that onetime investment is reflected in our outlook today.

[3] Drillr · Evolent Health, Inc. (EVH) · 2025-08-08 · Earnings call

Quote: We have now integrated this technology into a number of our workflows, improving review efficiency by roughly 11% in the last quarter since starting to roll it out.

[4] Drillr · Evolent Health, Inc. (EVH) · 2026-02-25 · Earnings call

Quote: For example, through this optimization, our real-time auto authorization rate for chest CT scans rose by over 11 points and cervical spine MRI rose by 16 points.

[5] Drillr · Evolent Health, Inc. (EVH) · 2026-05-07 · Earnings call

Quote: Our initial rollouts have shown auto-approval increases in the high teens on cases evaluated by these models, and in some cases up to 30%, all with minimal clinical value loss for our customers.