Banjo Health

The models are the engine of the core module, and the platform carries other value. CARE (Computer Assisted Review Engine) extracts information from faxed and digital requests, matches it against the plan's clinical criteria and recommends approve, deny or request more information, and Composer turns narrative clinical guidelines into decision trees.

Around them Banjo sells case management for prior authorization, appeals and grievances, a no code workflow and decision tree builder, deadline routing, a testing sandbox and a searchable decision history, which carry value without the models.

What the system may and may not do are both published, with the review path. CARE recommends a determination (approve, deny or request more information) with detailed evidence and an explanation; Banjo states the ultimate decision stays with human experts, and its October 2025 post describes the workflow: the model evaluates each request and recommends, then a human reviewer confirms or adjusts it. Every case therefore routes to a person, and the reviewer's adjustment is the override.

New or updated decision trees are tested in a sandbox before production. The residual risk is automation bias: a reviewer confirming recommendations at speed is the control. Banjo does not report how often reviewers change a recommendation, or how deny recommendations are reviewed.

Banjo names its approach and stops short of versions or accuracy. CARE extracts case data from faxed and digital requests and matches it against clinical criteria; Composer uses generative AI to turn narrative criteria into decision trees a plan can read; Banjo describes its models as proprietary, designed from the ground up for payer operations, and constantly learning. No accuracy figure, evaluation method or change log is published. Not all coverage policy reduces neatly to a tree: converting 'medically necessary considering the totality of the record' into a branch is an interpretive choice.

A structurally short chain, stated by the vendor. Banjo's October 2025 post says its AI is proprietary and designed from the ground up rather than built on an open source model, giving it control of every layer of the stack, which places the models in house. The hosting provider, any subprocessor that receives case data, and whether Composer's generative step calls an outside model are not published, and no subprocessor list is available.

Vendor reported percentages and a named customer, without method. Banjo reports 50 percent faster intake to decision, 70 percent less time drafting and updating decision trees and a 33 percent reduction in appeals, with no baseline, period or customer. Navitus, a pharmacy benefit manager, selected Banjo in January 2023; no measured outcome from the Navitus deployment is published on Banjo's site.

A categorical statement on where data goes, without the retention schedule behind it. Banjo's October 2025 post states its AI is proprietary, designed from the ground up, and not an open source model, giving it control over every layer of the technology stack, so case data is not described as reaching an outside model provider.

It also says the AI is constantly learning and adapting, which implies customer determinations improve the models, and nothing says whether that learning stays within one plan's boundary. Banjo reads prescriber notes from the EHR, so it holds clinical narrative written to support a medication request; no retention period is published for it. Safety engineering is not published beyond human review.

Compliance is claimed without the document behind it. Banjo states adherence to health privacy and CMS standards and holds HITRUST r2; its only privacy policy (effective January 2023) is written for the website and states no business associate status. No agreement or its terms are published. The platform reads prescriber notes from the EHR, so provider data reaches a vendor working for the payer, and nothing says under which agreements that content moves.

A recognized certification is named without the artifact. Banjo announced HITRUST risk based, two year (r2) certification, and its October 2025 post states it is HITRUST r2 certified. The company publishes no trust center, certificate or certification date, and since an r2 certification runs two years, neither the current certificate nor its scope can be checked from public material.

No device claim is made and the product is scoped accordingly: prior authorization decision support for health plans and pharmacy benefit managers sits outside software as a medical device. Banjo's stated position is that its AI recommends and human experts make the final determination, and that the platform supports CMS and state prior authorization requirements with a searchable decision history.

The live exposure is the growing set of state laws conditioning AI involvement in coverage determinations; Banjo writes about Washington's prior authorization law but states no product position on AI rules.

Responsible AI is committed to in policy language, with no evaluation behind it. Banjo's October 2025 post on securing AI describes human review of every recommendation as a core principle and speaks of blending automation with accountability; no governance framework, bias evaluation or subgroup results are published. Banjo also says its AI is constantly learning and adapting. A prior authorization model that learns from a plan's past determinations carries forward whatever patterns were in them, and in the pharmacy benefit a denied drug is felt at the counter.

Mechanisms let a reviewer challenge the output, and nothing stands behind it. Every AI recommendation carries detailed evidence and an explanation, Composer's decision trees let a plan read the logic applied to its members, new or changed trees can be tested in a separate sandbox, and the platform keeps a searchable history of every prior authorization decision. A reviewer confirms or adjusts each recommendation.

What is missing is anything standing behind the output: no accuracy figure, no commitment when a recommendation or a generated decision tree is wrong, and no route for a member or prescriber other than the plan's own appeal.

Integration is claimed with no system named. Banjo says it works with EHRs, eligibility verification tools, claims management systems, ePA vendors and API ready portals, ingests faxed and digital requests, and runs appeals through the same ePA vendor as the original request. Its partners page names no partner, and no EHR, ePA network or integration method is named on its product pages, including for reading prescriber notes, the hardest connection in the product.

A single hosted service with the location implied. Banjo's privacy policy (effective January 2023) says information it collects is stored in the United States, but it is written for the website and does not address the platform's case data. No hosting provider, region, tenancy model or subprocessor list is published.

Plans and pharmacy benefit managers competing for the same contracts run their criteria and determination history on a shared platform, and the company does not describe what separates one customer's environment, criteria library and determination data from another's, or whether any model spans them.

How Banjo prices is published, and what it charges is not. Its pricing page states pricing is set per customer by member population size, transaction volume, workflow complexity and integration requirements, through a consultative process. Enterprise agreements with health plans and pharmacy benefit managers.

CMS prior authorization rules are the stated adoption driver, so contracts may be negotiated under a compliance deadline, and the pricing page does not say whether pricing reflects the deadline driven implementation or steady state operation.

Coverage is named with validation behind part of it: payer side prior authorization, appeals, grievances and clinical criteria management for health plans and pharmacy benefit managers, across medical and pharmacy benefits. The pharmacy benefit manager setting is supported by a named customer (Navitus); no health plan customer is named on Banjo's site, and no validation is published for the medical benefit.