UnitedHealthcare
genAI companion for benefits navigation and care coordination
UnitedHealthcare launched Avery in March 2026, a generative AI companion open to 6.5 million members and targeting 20.5 million by year-end, with 90% of users resolving their request without a human advocate.
Objective
Make using health insurance simpler for the member, by answering 24/7 in natural language on coverage, reimbursements, and costs, and easing the load on human advocates for requests the member can resolve alone.
The deployment
Avery is a generative AI companion launched by UnitedHealthcare on March 26, 2026. It answers in natural language, 24/7, on coverage, reimbursement details, cost estimates, appointment-booking help, and care options, and can call network providers to book an appointment on the member's behalf. It is accessible through the UnitedHealthcare app and myuhc.com. At launch, Avery is open to about 6.5 million members on employer-sponsored plans and 160,000 Medicare Advantage members, with a target of 20.5 million commercial, Medicare, and Medicaid members by year-end. UnitedHealthcare says that 90% of the time, members who use Avery did not need an advocate.
Results Proof C
The launch and volumes are documented by the official UnitedHealth Group release and by established press (Fierce Healthcare). The launch scope (6.5 million) is in production; the 20.5 million target is prospective. Press naming the brand plus a primary source = C, not A because it is not a financial result.
How it works
Documented architectureThe stack in detail
- outil Avery Generative AI companion developed in-house by UnitedHealthcare: natural-language answers on benefits, reimbursements, and costs, and appointment booking by calling providers.
- llm LLM d'Avery (modele exact non publie) UnitedHealthcare does not name the underlying generative model; the sources describe a custom/in-house generative AI.
- plateforme Application UnitedHealthcare et myuhc.com Member's access surfaces to Avery, open to about 6.5 million members at launch.
- infra Systemes de benefices et de claims UnitedHealthcare Plan data, reimbursement history, and cost reference that feed the personalized answers; without up-to-date access, the assistant refers the member to a human.
How it runs, concretely
For ops teams-
1Member question Customer
The member queries Avery through the app or myuhc.com about their coverage, a cost, or an appointment.
-
2Personalized answer AI
Avery answers in natural language from the member's benefits and history.
-
3Action on the care journey AI
If needed, Avery calls a network provider to book an appointment on the member's behalf.
-
4Escalation Member advocate
Unresolved cases are passed to a human advocate (about 10% of cases).
-
5Learning Product team
The interactions feed the improvement of the answers and the self-service.
The member's own benefits, reimbursement, and cost data. Without up-to-date access to the plan and the claims history, the assistant cannot give a reliable personalized answer and refers to a human.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- benefits and plan data per member
- claims history
- a reference set of costs and the provider network
Org prerequisites
- a conversational product team
- advocates for escalation
- health compliance governance
Possible stack
- an LLM with RAG over the plan data
- conversational agent orchestration
- telephony integration for appointment booking
The plan, step by step
- Step 1Frame the data and compliance: map benefits, claims, and costs per member, set the health legal framework.Deliverable: An inventory of member data accessible in real time + a validated compliance dossier.
- Step 2Build the assistant on a narrow scope of frequent questions (coverage, reimbursements), in RAG over the plan data.Deliverable: An internal prototype tested by advocates on real cases.
- Step 3Connect human escalation and launch a closed pilot on a member segment.Deliverable: A pilot with self-service resolution rate and escalation rate measured.
- Step 4Add actions on the care journey (appointment booking, calling providers) with full logging.Deliverable: An assistant able to act, each action traced.
- Step 5Roll out progressively by line of contracts, tracking the use of advocates.Deliverable: A rollout with a self-service vs escalation dashboard by population.
First step: Reliably tie a member's benefits and claims data to a conversational assistant, with a clear human escalation.
Sources
- S1 UnitedHealthcare Introduces AI Companion Empowering People with Simpler Navigation, Personal Experience Primary archive pending
- S2 UnitedHealthcare launches Avery, a generative AI companion for members Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.