Verizon
Verizon says sales through its 28,000 service agents are up nearly 40% after a Gemini assistant
Verizon built an AI assistant for its 28,000-person customer service team using Google's Gemini model and about 15,000 internal documents. The consumer group's CEO says sales through the service team are up nearly 40% since deployment, as agents shifted from handling problems to selling.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
up nearly 40%
Increase in sales through the service team since deployment, per Verizon's consumer group CEO1
nearly 15,000
Internal documents fed into the model1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Verizon runs a very large customer service operation. Its consumer group wanted agents to answer questions faster and spend more of each contact on selling products, not only fixing problems.
What they built
An AI assistant that sits on the screen of customer service agents and helps them find the right answers to customers' questions. Verizon built it by feeding nearly 15,000 internal documents into a version of Google's Gemini, a large language model, which is an AI system that reads and writes text. The tool serves a 28,000-person service team.
How they did it
Verizon began deploying the AI features in July 2024 and ramped up to full scale in January 2025. The company works mainly with two cloud providers: Amazon for application deployment and computing, and Google Cloud for analytics and AI. It chose Google because it could deploy the AI to tens of thousands of people, according to the consumer group CEO. He said there are much easier ways to cut costs than replacing staff, and described the change as reskilling care agents into selling agents.
Results
Reuters reports that the assistant cut call times and freed agents to sell. Verizon's consumer group CEO says sales through the service team are up nearly 40% since deployment. Google Cloud's CEO described the scale as enormous compared with what others are doing. Verizon announced the results at Google Cloud's annual conference.
What to copy
Pair an agent-assist tool with a change in what agents are asked to do. The time it saves only becomes revenue if the role changes too.
Lessons for your project
Aim the assistant at the agent's screen, not the customer. Verizon's tool gives service agents answers while they talk to customers.
Use time saved on calls to change the job. Verizon moved agents from solving problems towards selling and calls this reskilling in real time.
Ground the model in your own material. Verizon built the tool by feeding it nearly 15,000 internal documents.
Roll out in stages. Verizon began in July 2024 and reached full scale in January 2025.
Pick a platform that can serve tens of thousands of users. Verizon says that was why it chose Google over rivals.
Caveats
- The sales figure is a statement by a Verizon executive, reported by Reuters at Google Cloud's annual conference. Neither Reuters nor the article tests it.
- The report does not say how sales were measured, over what period, or what share is due to the AI assistant rather than other changes.
- Reuters says only that the model is a version of Google's Gemini. The exact version is not named.
- The article is syndicated by AOL and the original Reuters page was not opened. Reuters reports the call-time reduction without a figure.
- Deployment dates, July 2024 start and full scale in January, are as reported by Reuters, without a day.
- Google Cloud's press release of the same day, a vendor source, confirms the Gemini-based assistant but states no sales figure. It says the tool is deployed across 28,000 care reps and retail stores.
Sources
- Verizon says Google AI for customer service agents has led to sales jump
Reuters (via AOL) · 2025-04-09 · News · aol.com
- Google Cloud and Verizon Drive Customer Experience Improvements for Verizon Customers with Gemini Integration
Google Cloud · 2025-04-09 · Press release · googlecloudpresscorner.com
Summarized in our own words from the sources above. Spotted an error? See corrections.