Cox Communications on scaling Claude across largest privately-held telecom company in US

Cox Communications is a family-owned company and the third-largest cable provider in the United States, serving six million customers over a fiber-powered network that reaches roughly 12 million homes and businesses. Mark Greatrex, the company's president, and Eric Pace, who leads its Center of Excellence for AI, sat down with Anthropic to talk about rolling out AI to 15,000 colleagues, why leadership and values shaped the path, and what they are building next.

Anthropic: Cox has been around for a long time, serving millions of residential and commercial customers across the country. How has your approach to AI evolved to get you here?

Mark Greatrex, President, Cox Communications: When I arrived as chief marketing and sales officer, we had a big focus on data, marketing and sales science, and on ROI. We then applied machine learning to network management, to real-time decisioning for revenue growth, and to retention, which built the capability to embrace AI early. We created our Center of Excellence for AI in 2024, and I asked for three things: AI to accelerate profitable revenue growth, which is our number one priority; AI to step-change customer experience; and AI to take real cost out of operations.

What's the biggest barrier to getting 15,000 people to actually adopt AI?

Eric Pace, Head of AI, Cox Communications: The first barrier to adoption is literacy, and the second is leadership support. If your leaders aren't talking about it, using it, and demonstrating publicly that these things are okay, and that they're not here to take jobs but to amplify the work people do, you won't see the adoption you need. Without Mark standing up and showing we're behind it, it would be very hard to sell our value proposition to the rest of the company.

With so many possible places to apply AI, how did you decide where to focus?

Pace: Mark created a space where he said: revenue growth, experience, and expense are our focus, but you have carte blanche to find the best ways to get the benefits. When you give a team that kind of freedom inside a clear framework, you can run with it in really innovative ways. Most companies we talk to are going after efficiency AI, which everyone needs. But Mark pushed us to pay attention to top-line growth and experience too. The governance we had in place let us say no to a lot of ideas and focus only on the ones we believed would drive real benefit.

Greatrex: I trust our teams: We set a high bar, create a framework, and in return Eric and the team gave us security, governance, and the discipline to build the business cases. They've earned the credibility to decide which are the best tools.

How much do the values of an AI partner factor into which AI companies Cox works with?

Greatrex: As a business leader, I've become much more discerning about the values of the LLM partners we work with. When I look at who's in the consideration set, I ask who is going to shape and steer a path to the responsible use of AI. Anthropic comes to the fore. That matters enormously to us. We're a family-owned business focused on doing good. The responsible use of AI is huge, and I lean toward the partners where I see that values alignment.

Pace: That alignment of company values made it easy. Our default guidance to teams is: if you're going to use an LLM in the cloud, use Anthropic through Amazon Bedrock unless there's a specific reason to deviate. A lot of what we need in order to adhere to our responsible AI principles comes out of the box. With other options, we'd have to go double down on some of that. There's a trust layer with Anthropic that makes the alignment to those values very easy for us. When teams run their own comparisons, the results back it up. I've never seen anybody say, "I was using Claude and I found this other model was better."

What has your team actually built with Claude so far?

Pace: We've built a set of multi-agent systems, all backed by Claude through Bedrock, against our B2B marketing and sales funnel: lead validation and enrichment, campaign creation and brief execution, and seller enablement. Those are running in production now and starting to scale. On top of them, we're building our most ambitious piece: an agentic system that can take a whole sales segment from the first signal to a closed order.

Claude Code is one of your biggest use cases. How is it being used, and has it spread beyond engineering?

Pace: We did an enterprise rollout of Claude Code before we ever gave anyone access to Chat or Cowork. So many employees were building their own agents on other platforms that when we told them we also had Claude Code, they rebuilt those workflows in it and started handing the hardest tasks off to richer environments Claude could help them build.

We're on our way to 2,500 Claude Code users and climbing, more than half of them non-development engineers, business folks building their own agents to see what happens in the "software for me" era. We give them guardrails and say, so long as you stay in this frame, happy hunting. Because we proxy all the calls through our gateway, we have visibility into everything they're doing. The way we put it: You have an everything maker and an anything doer, and you just have to be clear about what you want. If you're comfortable in an IDE, use Claude Code; if not, Cowork will get you there.

Are the two of you using these tools in your own work?

Pace: The last 12 months have completely inverted how much I do with AI versus without, just because of how comfortable I've gotten in the CLI and the IDE. I use Claude Code for pretty much everything. I've also moved back upstream and started using Cowork for long-running research tasks. I built a personal-brand engine the other day in Cowork. After a couple of months, there's infinite power there for the everyday person.

Greatrex: I try to model behavior, and first and foremost that means being curious. I use Claude all the time now as my thought partner. It broadens my field of view, makes me a lot more creative, and a heck of a lot more productive. 

What results have convinced the business to keep investing?

Greatrex: We now know we got a 7x ROI on our first year of investment. And we've kept up the pace of AI investment even as a lot of our other discretionary development has been trimmed. The belief in AI has remained because of the results.

How did you bring skeptical leaders along?

Greatrex: We’re pioneering a new frontier so it’s natural to have some healthy skepticism. But it’s important to stay curious and open your mind to what’s possible.  Eric and his team took building the business cases and reading back the ROI very seriously. I spent enough time with Eric to see the vision and the value, and I knew there was a real there there.  

Pace: For example, we have a project we call The Cube. The Cube is a six-layer set of data and AI-generated inferences about customer behavior, built on top of the classic machine learning and data science we already had. What Claude adds is the layer on top. It's one thing for a classic machine-learning model to give you a churn prediction. It's another to extend that prediction, make it situational and context-aware, and then lay out all the situations in which we need to treat that churn, because churn isn't treated equally in every scenario. When you tell people you're going to extend ML models, they ask what you're even talking about. So you tell the same story a thousand times, you get Claude to help you rewrite it, and you have the conversation once more. Once you see it click in the room, the ideas start flowing, and that's when you go build something together.

With 15,000 colleagues full of ideas, how do you keep the work from fragmenting into disconnected projects?

Pace: We're a company of 15,000 colleagues full of ideas, so we put them through a funnel: latitude to build, but in the construct we provide, where we can gate the use cases. I don't need a thousand flowers to bloom, I need 10 that produce value. We bring people together and say, before you build the thing, go talk to the team that already built it, and you can both use it. 

What are you most excited to build next?

Greatrex: One of the toughest challenges is finding new sources of profitable growth. If we can restart growth in our core business—whether residential broadband, wireless and video, or commercial services—that would be huge. 

The one I want to crack is the revenue side. My expectation is that 70 to 80% of the value will come through effectiveness, not efficiency. We sell high-margin services, so if we can drive volume and revenue growth, that will dwarf any cost savings. I'm looking for demand generation and sales conversion.

Pace: The moment that proves the investment paid off is the prospect-to-order agentic function we're building for B2B marketing, where we go full lifecycle against an entire segment. Not with humans at the helm, but with humans auditing the right parts of the flow. We want the next three months to show us what the next three years could look like. Once we prove the pattern, we have a template we can take to any department in the company. I'm over the moon excited about it.