Omnicell

The business is medication automation hardware and the models built inside it are additive to that. Omnicell One applies predictive and prescriptive analytics to pharmacy inventory, Omnicell Analytics scores diversion risk, and OmniSphere is positioned as the data platform unifying both. Remove all three and a very large working business remains, built on cabinets, robotics and compounding systems that dispense whether or not a model is running.

The Autonomous Pharmacy framing is a strategic vision for the category rather than a description of model driven decisions in the shipping product. One qualification belongs with the grade. In October 2025 Omnicell acquired ANiGENT, whose MAAP Analytics platform is genuinely machine learning native, and that brand is indexed separately here because it still trades under its own name with its own site and product.

The company therefore now owns model native capability that this record does not cover, and a reader comparing the two grades should treat them as describing different halves of the same corporate portfolio rather than a contradiction. The grade is not a judgement on the scale or quality of the automation, which is substantial.

The company has named its entire strategy after a claim it does not actually make. Autonomous Pharmacy is the brand and autonomous medication management is the phrase in its own filing, while the passage that follows commits to equipping pharmacists and nurses to focus on patient care rather than administrative tasks. That is a bounded and defensible position: the automation removes counting, stocking and reconciliation, and a licensed professional still owns the clinical decision.

The grade reflects that the boundary is inferable from the filings rather than stated as an oversight model, so what a nurse or pharmacist remains accountable for at each automated step is left to the customer to work out.

Predictive and prescriptive analytics is the full extent of the description offered, sustained across six years of product communication for Omnicell One and carried unchanged into the OmniSphere material. No method, model family, feature set, training population or performance measure is named for any of the intelligence products. A buyer evaluating the diversion analytics in particular has no published basis on which to ask how a score is produced or how often it is wrong.

The company acknowledges to shareholders that artificial intelligence enters its business both through its own products and through its vendors' offerings, which is a clear statement that third party models sit in the stack. No supplier is named anywhere a buyer can see it, and no line is drawn between capability the company built and capability it licenses. Investors are told the dependency exists and customers are not told what it is.

Adoption is verifiable at a scale almost nothing else in this index can match, through securities filings and an installed base spanning acute, retail and post acute settings. What is absent is measurement attributable to the intelligence layer. Inventory optimisation, diversion detection and waste reduction are described as outcomes the products drive, without published figures naming a customer, a baseline and a result. The distinction matters more than usual here: a buyer can confirm that thousands of institutions run the cabinets, and cannot confirm what the analytics changed at any one of them.

The published material treats security and model behaviour as separate subjects and addresses only the first. Whether customer medication and dispensing data trains models, whether models are tenant isolated or fitted across the installed base, and what retention applies to the data feeding the predictive services are all unaddressed.

This is where the HITRUST scope is least helpful, since it attests to controls over data at rest and in transit rather than to any commitment about the uses that data is put to once it is inside the platform.

The control environment around protected health information is externally attested through HITRUST certification covering both the medication management suite and the cloud platform that processes its data. The commitments themselves are unpublished. Business associate agreement terms, the position on data ownership, and the handling of dispensing records that identify patients are all matters a buyer resolves in contract rather than by reading anything on the site. Certification of controls is real assurance about how data is protected and a much weaker signal about what the vendor may do with it.

Two HITRUST certifications at the i1 level, one covering the medication management suite and one awarded to the OmniSphere cloud platform in June 2025, with the scope of the second enumerated product by product rather than left as a badge.

A published question and answer document goes further than most in this index: it states which systems were outside the scope at the time, explains why a single location appears on the certificate, gives the expiry, and provides a dedicated address for customer questions, with the certificate supplied on request. Naming the limits of your own certification is precisely the behaviour this axis exists to reward. Two things hold it below the top grade.

The i1 level is the moderate tier of the HITRUST scheme rather than its most rigorous, a distinction the company's own materials do not draw. And there is no self serve trust centre, so a security team seeking evidence enters an email exchange rather than downloading it.

Regulatory position is the least documented area of an otherwise well documented company. The product lines span dispensing automation, robotics and sterile compounding, categories where device registration and state pharmacy regulation both bear, and two passes across the company site and its filings surfaced no clearance, listing or classification statement.

The predictive services are marketed on inventory and diversion rather than on clinical determination, so clearance may well not apply to them, which makes the silence understandable without making it useful to a buyer who has to answer the question during procurement.

The only place this company discusses artificial intelligence risk is in the risk factors it files for shareholders, which name the incorporation of generative and agentic technologies into its products, services and processes as a source of uncertainty. Nothing equivalent is offered to buyers. There is no published governance position and no statement of what the models are tested for, which leaves the two places bias would bite here entirely unexamined.

An inventory model fitted to historical demand will under supply the units that were historically under served, and a diversion model scores individual named clinicians as suspects, which is an accusation carrying employment and licensure consequences.

Where responsibility sits when an automated recommendation turns out to be wrong is unaddressed in the published material. That lands harder here than in most records because of what these products touch. A stock out driven by a bad inventory prediction reaches a patient as an unavailable medication, and a false diversion score reaches a named clinician as an allegation of drug theft. Both failure modes have identifiable people on the receiving end, and neither has a published route to challenge the output or obtain a remedy.

Interoperability is a real engineering asset here rather than a marketing claim. The cabinets have integrated with hospital information systems for decades, OmniSphere is sold specifically on interoperability with the electronic health record and on unifying data across the automation estate, and the international product line describes a central intelligence layer giving a single view across an entire deployed fleet. The grade stops below the top band because named integrations with specific electronic health record vendors, and the depth of each, are asserted at platform level rather than documented.

A buyer can tell exactly which product runs where, which is rarer than it should be. OmniCenter is the on premise platform, OmniSphere is cloud native, and the company names Amazon Web Services as the host in its own certification scope rather than referring vaguely to the cloud. Both generations are in the field at once and the company is open about the transition between them. Residency below the level of the provider, meaning region, and any option for a customer to constrain it, is not addressed, which is what holds this below the top grade.

No price is published for any product line, and the company's own distribution partners list the cabinets as call for price. Third party listings circulate figures that are not merely unverified but mutually contradictory, running from roughly $18,000 to $65,000 for the same cabinet family, which is itself a signal that none is sourced from a rate card.

What the company does publish is investor facing: revenue mix, and a disclosed transition toward selling more products and services on a subscription basis. That tells a shareholder how the money arrives and tells a hospital nothing about what it will pay.

Coverage is broad and named rather than implied: acute hospitals and health systems, central pharmacy, perioperative and anaesthesia workstations, retail and specialty pharmacy, long term care, and intellectual and developmental disability residential settings, with separate product surfaces for international markets including the United Kingdom and the Middle East. It sits below the top grade because the breadth is documented as product availability by setting rather than as evidence of performance within any particular specialty.