GEICO
self-service virtual assistant on the customer account
GEICO runs a self-service virtual assistant (Kate in the app, an AI assistant on the site) that handles high-volume account questions; at its launch, PYMNTS reported an increase of about 10% in app engagement.
Objective
Replace static FAQs, voice menus, and forms with a conversation that knows what to ask, to resolve routine requests in self-service and relieve the call center.
The deployment
GEICO runs a self-service virtual assistant on its digital channels. In the app, the Kate assistant, launched in January 2017, answers, in natural language, high-volume, low-stakes account questions: balance, next payment date, retrieving the insurance card, coverage summary, document downloads, roadside assistance requests. On the site, GEICO now exposes an AI assistant that displays coverages, cards, and documents, and answers policy questions, disclosing on every exchange that the user is talking to an AI and offering to hand off to a human agent when the request exceeds it. At its launch, PYMNTS reported an increase of about 10% in app engagement attributed to Kate. The assistant is still in service and has evolved into a generative layer on the site, in a logic of a conversation that knows what to ask rather than pointing to a help page.
Results Proof C
Assistant documented by GEICO's official page (proof of current service) and by trade press reporting the launch and a quantified engagement increase. The engagement figure dates from launch (2017) and is self-reported; the existence of the service is verifiable live on the site.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- outil Assistant virtuel GEICO (Kate) In-house conversational assistant launched in 2017 in the app, since evolved into a generative layer on the site; GEICO does not publish the underlying technologies
- llm Couche de comprehension du langage naturel et genAI NLU to identify the request, then generative AI on the current site assistant; exact model not named publicly
- infra Systemes de compte et de police GEICO Authenticated access to policy data, payments, and documents, the condition for self-service (balance, due date, card, downloads)
How it runs, concretely
For ops teams-
1Identifying the request AI
The assistant understands the customer's question in natural language and discloses that it is an AI.
-
2Self-service response AI
It displays the requested account information (balance, due date, card, document) or performs the routine action.
-
3Handoff to an agent AI and customer service
When the request exceeds self-service, the assistant offers to move to a human advisor.
Access to the customer's account data (policy, payments, documents). Without authentication and up-to-date data, the assistant cannot answer account questions and is limited to generalities.
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
- Authenticated access to account data (policy, payments, documents)
- Catalog of self-service intents and actions
- Session logging
Org prerequisites
- AI transparency policy and human handoff
- Account access security
- Quality loop on the responses
Possible stack
- Custom conversational assistant (GEICO's path)
- Market chatbot/virtual agent platform connected to the customer system
- Framed LLM + connectors to the policy systems
The plan, step by step
- Step 1Extract the five most frequent account questions from the call center logs and frame the rules: AI transparency, human handoff, access security.Deliverable: Catalog of prioritized intents and validated governance rules
- Step 2Connect the assistant to the account data in authenticated access (policy, payments, documents) through read connectors.Deliverable: Secure account access tested end to end
- Step 3Launch a pilot on the app or the site with the five intents, AI disclosure on every exchange, and a handoff to an agent.Deliverable: Assistant in pilot with session logging
- Step 4Measure self-service resolution and call deflection, correct faulty responses, then widen the scope of intents.Deliverable: Quantified deflection / resolution review and extension roadmap
First step: Target the five most frequent account questions and handle them in authenticated self-service before widening the scope.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 Get Help Fast with Our GEICO AI Virtual Assistant Primary archive pending
- S2 Neither Gecko Nor Caveman, GEICO Adds Virtual Assistant Kate Established press archive pending
- S3 Geico introduces virtual assistant 'Kate' 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.