As part of a multiyear effort to consolidate these various legacy networks into a single architecture, Lumen tapped AI to simplify the network, modernize workflows, and streamline the disparate data that lives across its ecosystem.

“If there was an equipment that was made in the last 40 years, the chance of it being in our network is fairly high, extremely high,” says Alex Mercier-Dalphond, senior vice president of infrastructure modernization and operations at Lumen. “Trying to transform our network to be more simplified, more agile to support the AI workloads that we’re facing now is not a novel idea. AI just enabled us to deliver on that vision,” he says.

Lumen executives say this effort has delivered measurable business impact in a multitude of ways, including fewer customer outages and reducing the time it takes to answer vital customer questions from months to mere minutes. All together, Lumen says the aggregate impact of the initiative delivered $350 million in annualized cost savings in 2025, with a clear path to $1 billion in savings by 2027.

While consolidating systems and updating near-obsolete equipment might seem like an obvious way to get savings, strategy and execution are critical, as is the ability to understand precisely how and where the value is being delivered. Lumen attributes its success to the formation of its Lumen Transformation Office (LTO), standardized processes, and obsessive setting of KPIs from the outset of each transformation project.

“There’s a lot of governance and kind of a strict, maniacal focus, if you will, on how we track and measure those things internally so that we’re attributing them to the right activity,” said Kina Corcoran, chief data officer at Lumen. 

The transformation

When embarking on this transformation, the first priority was to create a semantic abstraction layer of all of the company’s data to gain an understanding of what data represents customers, what represents network elements, and all the other data objects. Lumen’s tech team then implemented an AI agent to sit on top of the more than 500 data systems, tools, and applications so the team could query everything in one prompt, which enabled them to start figuring out where to streamline. 

The results were revealing. As they began to merge inventory systems, for example, they discovered that the same asset would routinely be present in multiple systems, often with conflicting information. AI enabled them to rebuild those data models—millions or potentially even billions of records, according to Mercier-Dalphond—into one coherent system.

With the data now in one system, everything got easier. One impact with clear measurable success has been the reduction in time it takes to make sense of the complex factors relevant for each customer account, which is crucial in helping customers modernize themselves. For example, looking at a customer who has a legacy product, they might need to ask what specific region they’re in, what would be a viable go-forward product for them that’s low impact to change based on the infrastructure they currently have, and what’s the price differential for them to upgrade?

“That complex question could sometimes take us months to answer with multiple people. Now we get an output in a matter of seconds,” said Corcoran, adding that the value for both employees and customers “can’t be overstated.”

For Mercier-Dalphond, who oversees the hands-on with the equipment, time-savings have also had a remarkable impact. For example, for a planner on his team to consolidate equipment for an enterprise customer, they previously would have had to manually go into the 17 enterprise inventory systems and find which have the relevant assets. It would then take around two days to create the plan and another two to execute it, with a lot of handoffs within the company required. That planner would also have to alert the network operating center to update the inventory. All of this would set off an alarm lasting two or three days, which they’d also have to tell other teams to ignore.

Now with a single AI-powered workflow, Mercier-Dalphond said, it takes planners 15 minutes to make the plan and do all that traffic consolidation, and another 15 minutes to execute. And no one has to call the network operating center to alert them to the work.

“It’s all automated. It sequences everything right up to the boots on the ground,” Mercier-Dalphond said. His field team also has access to the tool. So if the reality in the field turns out to be different than what the planner came up with, it can quickly recalculate and remake the plan without the field technician ever having to leave the site or go back to the planner. 

“AI really sharpens our ability to overcome that complexity, so it really made that cost of not fixing it just impossible to ignore,” said Corcoran.

Measuring impact

The $350 million in annualized cost-savings Lumen is claiming for 2025 is an aggregate representing measurable impact across a wide variety of different use cases and work streams. The real priority for the company when it comes to actually measuring the business impact of this AI transformation, however, has been to approach it on a far more granular level.

Whenever company leaders are looking at use cases to reimagine or a potential AI initiative, they set KPIs specific to the use case from the outset. 

“Every time we’re going to invest a dollar in AI, we need to have a clear return that is actually measurable. So, for us, if we were able to say that it’s going to improve our customer satisfaction, it’s tied to the churn metric, and the churn is tied to dollars at the end. If it’s in legacy simplification, it’s tied to ticket reduction,” said Mercier-Dalphond.

To standardize this with an official process and expectations, the company set up the LTO, which Mercier-Dalphond believes has been “the secret sauce.” Working with the LTO, every business unit uses the same business case template, which is used to define the KPIs that will be used to track the specific initiative.

“It never became a science project because we were always anchored on, ‘What’s the problem?’ ‘Is it moving the needle?’” said ​​Mercier-Dalphond. “And if sometimes the solution is simple automation, we’re just going to do simple automation.”