Verizon
genAI agent assist: an LLM connected to the internal document base and customer context feeds the advisor the relevant answer and offer in real time
Verizon equipped 28,000 care advisors and stores with a Gemini assistant (Vertex AI) connected to 15,000 internal documents, reaching 95% answerability and a nearly 40% rise in sales through care in Q1 2025.
Key points
- genAI agent assist that feeds the advisor the relevant answer and offer in real time.
- Gemini on Vertex AI, RAG over about 15,000 internal documents.
- 95% of requests covered, sales through care up nearly 40%.
- Evidence B, confirmed status, 28,000 advisors and stores equipped.
Objective
Give each advisor a reliable answer in real time to shorten calls, resolve better, and turn service into a sales channel while the base loses subscribers.
The deployment
Verizon connected Gemini to its care platforms via Vertex AI. The advisor stays in control, but a conversational assistant called Personal Research Assistant looks up the answer across about 15,000 internal documents and the customer's context, then surfaces it in real time instead of leaving the agent to dig through a knowledge base. A second agent, Problem Solver, guides troubleshooting. The system covers 28,000 advisors and stores. Verizon reports 95 percent of requests the assistant can answer. Over the quarter, sales made through the care team jumped nearly 40 percent: Verizon reskilled its care agents into agents able to sell, while the group was losing postpaid subscribers after price increases. The rollout is part of a five-year collaboration with Google Cloud.
Results Proof B
Deployment figures confirmed by the official Google Cloud release (28,000 advisors, 95 percent answerability) and a nearly 40 percent sales rise reported around Verizon's Q1 2025 results. Vendor release plus press pickup: solid, but answerability is an internal measure and the +40 percent does not isolate the share strictly attributable to the AI.
How it works
Documented architectureThe stack in detail
- llm Google Gemini LLM that generates the responses of the conversational assistant (Personal Research Assistant) and the troubleshooting guide (Problem Solver).
- plateforme Google Cloud Vertex AI Hosting and orchestration platform for the conversational agents connected to the document base.
- outil Agent Assist Panel / Customer Engagement Suite Panel integrated into the workstation of the 28,000 advisors, which surfaces the relevant answer and offer in real time.
- infra Corpus documentaire interne Verizon About 15,000 documents (procedures, offers, pricing, troubleshooting) indexed in RAG and reindexed continuously.
- integrateur Google Cloud Technology partner for the rollout, as part of a five-year collaboration with Verizon.
How it runs, concretely
For ops teams-
1Index the internal knowledge Data / platform team
About 15,000 documents (procedures, offers, pricing, troubleshooting) are made available to the LLM in RAG. Without this clean base, the assistant hallucinates.
-
2Assist the advisor in real time AI / Gemini
During the call or chat, the assistant reads the intent and customer context, surfaces the relevant answer and offer in the Agent Assist Panel.
-
3Keep the human as decision-maker Care / sales advisor
The advisor validates, rephrases, and decides the offer. The assistant proposes, it does not close the sale in their place.
-
4Measure answerability and sales Data team + care management
Tracking of the share of requests covered, call duration, and sales through care to decide the extension.
The quality and freshness of the indexed document base. If a document is stale or missing, the assistant answers wrong with confidence and the advisor propagates the error.
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
- A clean, up-to-date internal document base (procedures, offers, pricing, troubleshooting)
- Access to the customer context in real time at the advisor's desk
- GDPR governance over the customer data exposed to the LLM
Org prerequisites
- A care/retail team trained to work with an assistant
- A process to maintain the document base (who keeps it up to date)
Possible stack
- Vertex AI + Gemini, or Azure OpenAI, or an in-house RAG on a market LLM
- An agent-assist component integrated into the CRM / advisor desk
- A vector search engine over the internal docs
The plan, step by step
- Step 1Choose a high-volume care domain and audit the associated documentation (freshness, owners, gaps).Deliverable: A documentation inventory with owners and update status.
- Step 2Index the documentation in RAG, connect the LLM, and test answerability internally on a set of real questions.Deliverable: An assistant in sandbox with an answer rate measured question by question.
- Step 3Launch the pilot on a group of advisors against a control group.Deliverable: A reading of call duration, answerability, and satisfaction, pilot vs control.
- Step 4Integrate the customer context and offers into the advisor panel, with the human keeping the decision.Deliverable: An assistant that proposes the relevant offer at the workstation.
- Step 5Set up the document-base maintenance process and plan the extension to the other domains.Deliverable: A documentation-update ritual + a costed rollout plan.
First step: Choose a high-volume care domain, index the corresponding docs in RAG, and measure answerability and call duration on a pilot group of advisors against a control group.
The same pattern, proven in other industries
Amazon
genAI shopping assistant
Booking.com
genAI trip planning assistant
Burberry
genAI copilot for consultants (search, image recognition, complete-the-look)
Sources
- S1 Google Cloud and Verizon Drive Customer Experience Improvements with Gemini Integration Interested party archive pending
- S2 Verizon Teams Up with Google to Give 28,000 Customer Support Reps a Personal Research Assistant Established press archive pending
- S3 Verizon AI boosts sales amid customer churn (reprenant Reuters, resultats T1 2025) Secondary 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.