Plenful

Language models do the defining work in one product, inside a platform that carries value without them. The Referral Agent uses LLMs to read clinical notes and other unstructured record data to find referrals that never appear as coded orders, a capability that does not exist without the models.

Around it sit 340B audit logic across every eligible claim, rebate reporting, Maximum Fair Price reconciliation and intake and prior authorization workflows, built on configurable rules, crosswalks and data connectors, much of which is orchestration rather than inference.

The review structure is described for one product, while the company's marketing promises completeness. The Referral Agent page describes shifting 340B teams from manual search to targeted review of claims the model flags with high confidence, with referral documentation generated and sent back to the referring provider, so a person reviews what the model surfaces.

Against that, the home page says "Plenful's AI identifies all your revenue opportunities among your messy data", and the solutions page offers to "Identify 100% of missed savings and referral opportunities", while the same home page reports a 4 percent missed-opportunity rate. A promise of completeness undercuts any review step: a reviewer checks what the model surfaced, not what it missed, and staff told every opportunity is found have no reason to look further. The same promise cuts the other way in 340B, where a claim surfaced as eligible that is not is an audit finding.

Named only as advanced LLMs within a rules platform. The Referral Agent uses language models to read structured and unstructured record data, reconstruct referral intent and match it to claims; the wider platform combines unlimited configurable rules and crosswalks with AI tailored to each customer's data, which is the right structure where some 340B criteria are bright-line rules and others need a note read. No model, supplier, version, evaluation or precision figure is published, and where rules end and models begin is not stated.

The architecture is described and no provider is named. The Referral Agent uses what the company calls advanced and best-in-class LLMs to read structured and unstructured record data, reconstruct referral intent and match it to downstream claims, and the wider platform combines configurable rules and crosswalks with AI tailored to each customer's data.

No model provider, hosting provider or subprocessor list is named; the privacy policy mentions third-party analytics providers for usage data, and the agreement allows customers to connect third-party services.

Named customers with attributed figures, and no baseline method. Case studies and quotes name Samaritan Health Services (96 percent time saving on 340B reporting), Mercy Med (auditing 100 percent of 340B claims), Renown Health and BioMatrix Specialty Infusion Pharmacy (four times capacity), and the company reports more than 100 organizations on the platform.

Headline figures add a 33 times audit capacity increase, a 4 percent missed-opportunity rate, $1,770 and $1,850 average savings per claim, 75 percent less manual work per prior authorization and a 98 times referral review capacity increase from early customer deployments. None of them comes with a baseline, volume, period or audit outcome.

A categorical training commitment is published in the customer agreement, without a retention schedule. The agreement says Plenful may use service data, including protected health information, to generate outputs, classifications and workflow actions for the customer, and "will not use PHI to train models for the benefit of other Plenful customers".

The privacy policy adds that service data may be used in anonymized form to train machine learning models supporting features, and that personal information is kept as long as necessary for its purpose. So a customer's PHI may tune models used for that customer, and anonymized data may train shared models. The Referral Agent reads clinical notes about named patients to establish a financial eligibility fact, and nothing states what is kept from a note after the determination.

No HIPAA status or business associate agreement is published, and a substantive document reaches the service. The customer agreement governs service data including protected health information: Plenful may use PHI to generate outputs and workflow actions for the customer and will not use it to train models for other customers, and the privacy policy makes the customer responsible for having the authority to share records with Plenful. Neither says HIPAA or business associate.

Business associate status is structurally certain, since the platform reads claims and clinical notes for covered entities, and the chain is layered: 340B data moves between the covered entity, contract pharmacies, child sites and third-party administrators, several of them separate legal entities.

Controls are asserted with nothing independent behind them on the company's own pages. The privacy policy (August 2026) says Plenful takes reasonable, industry-standard measures and restricts access to employees who need it, then states it makes no representation as to the reasonableness, efficacy or appropriateness of those measures; the customer agreement warrants only that it accurately describes its safeguards.

No SOC 2, HITRUST or ISO 27001 attestation, security page or trust center appears on the home, Referral Agent, privacy or terms pages. An AI-generated third-party software profile lists HIPAA and SOC 2 Type 2 for the company, which is not the company's own claim. The platform ingests claims, pharmacy records and clinical notes from more than 100 organizations.

No FDA pathway applies and none is claimed; the product automates 340B, rebate, Maximum Fair Price and prior authorization administration. The regime that governs is the 340B program, administered by HRSA, with covered entities audited for duplicate discounts, diversion and patient definition, plus Maximum Fair Price reconciliation under Medicare drug price negotiation. Working inside that regime does not amount to a regulatory pathway.

The allocation of risk is explicit in the customer agreement, which says Plenful has no liability for the customer's compliance programs: the covered entity answers to HRSA for eligibility the platform surfaces.

Nothing is published on how the models are governed or tested, and every published figure measures capture in one direction. The home and Referral Agent pages report a 4 percent missed-opportunity rate, $1,770 and $1,850 average savings per claim, a 33 times audit capacity increase and a 98 times increase in referral review capacity.

None measures the opposite error: how often a claim the model surfaces as 340B eligible would fail an HRSA audit, or how often captured claims are later reversed. The customer agreement states that Plenful has no liability for the customer's compliance programs. Because configuration encodes the customer's own reading of eligibility, a more expansive reading raises capture and audit exposure together.

Mechanisms exist to check an output, and the contract stands behind none of it. The Referral Agent produces closed-loop referral documentation for each captured claim, described as a defensible, HRSA-ready evidence record, so a reviewer or auditor can trace why a claim was captured.

The customer agreement warrants only that the services perform materially in accordance with their documentation, with termination as the exclusive remedy; disclaims that the services are error-free; states that Plenful has no liability for the customer's compliance programs; and indemnifies only for intellectual property claims. The 4 percent missed-opportunity rate measures misses, not wrongly captured claims. The consequence of a wrong eligibility determination is a 340B audit finding the covered entity carries.

Integration is described by format with no system named. The home page says the platform connects disparate systems through PDFs, flat files, APIs, HL7, SFTP and databases, with unlimited rules and crosswalks, and the Referral Agent sends referral documentation back to the referring provider "via major EHRs". It names no record system, third party administrator, pharmacy management system or ePA network with a method or a live site. Reaching contract pharmacy claims and matching them to referral evidence in the record is the hard part of 340B, and it is claimed rather than shown.

Nothing is published about where the platform runs or where data rests. No hosting provider, region, tenancy model, residency commitment or subprocessor list appears on the home, Referral Agent, privacy or terms pages. The question that matters most is where language model inference runs: the Referral Agent sends structured and unstructured record content, including clinical notes, to what the company calls best-in-class LLMs, and if that inference happens at an outside model provider, narrative documentation about named patients leaves the covered entity on every review.

Sales run through enterprise agreements with health systems and pharmacy organizations, with no public pricing and no rate card.

Coverage is named, with validation behind part of it. The company names three buyer groups: 340B covered entities (audit and optimization, referral capture, rebate management, Maximum Fair Price reconciliation), specialty pharmacies (benefit investigation and prior authorization) and infusion teams (order intake and prior authorization), across health systems, academic medical centers, specialty and retail pharmacies and provider groups.

The 340B work is supported by named customers and case studies (Samaritan Health Services, Mercy Med, Renown Health); the specialty pharmacy and infusion lines carry headline figures with no named reference behind them. No clinical scope is claimed.