BLOG | Nobody woke up wanting eight billing systems

"No telecom has woken up one morning to say, I want eight billing systems. I want 16 ordering tools."

Nevash Pillay isn't joking. She's making a serious point about where the industry finds itself as it tries to scale AI across enterprises that weren’t designed with AI in mind. Carriers across the Americas are mostly organizations that have been operating for over a century — acquiring businesses, selling businesses, integrating systems that were never meant to talk to each other. The complexity isn't a failure of planning. It's the residue of history.

The problem is that history is now getting in the way.

TM Forum research calls this the legacy drag. 88% of CIOs across the Americas say migrating from closed legacy systems is difficult — well above the global average of 73%. 81% acknowledge dependence on legacy vendors and systems integrators. And 47% cite governance as their top barrier to AI adoption — higher than anywhere else in the world, and a specifically Americas challenge that skills investment alone won't fix.

Pillay sees this pattern playing out daily. She works with 80% of the world's telcos. The conversation is mostly the same.

The pilot trap

"Without trust in your data or your AI agents, it's impossible to move out of a pilot."

The argument is straightforward, but its implications are significant. AI agents don't just answer questions anymore. An agent in the network can be talking to an agent in billing, talking to another agent across the call center — together making decisions, 24 hours a day, seven days a week. At that scale, with that level of autonomy, the quality and trustworthiness of the underlying data isn't a nice-to-have. It's the whole game.

Getting the foundation right — the right semantics, the right ontology — gives AI agents the opportunity to work from a trusted source. Without that, hallucination follows. And at scale, an AI strategy built on untrusted data is practically impossible to execute.

What Lumen actually did

Idilio Moncivais leads financial intelligence and data architecture at Lumen Technologies — a company operating, as he confirmed, across 16 ordering systems and eight billing systems. The complexity is real. So is the progress.

At Lumen, trustworthiness is now part of change control and part of the software development lifecycle. That's not a principle or a policy aspiration. It's an embedded process. Data models are documented, tested, and published before they go anywhere near an AI system. Business logic is reviewed, guardrails are in place and human oversight remains part of the loop.

It sounds painstaking because it is. But it's also what makes the difference between a demo and a deployment.

Four things that actually work

Pillay outlined a practical framework that Databricks is applying across its telco customer base — four foundations that determine whether AI operates reliably or not.

The first is context: defining data in a way that carries meaning specific to the business. What a severity-one incident means at Lumen is not the same as what it means at AT&T. AI that doesn't understand that distinction will act on the wrong signal.

The second is ontology: linking all the data points that relate to a single customer — network experience, outages, call history, billing, complaints — so that when an agent acts, it has the full picture, not a fragment of it.

The third is control: full access management, lineage, and auditability. When a CEO asks whether the data can be trusted, "yes" is not enough. The question is: how do you know? The answer has to be evidenced, not asserted.

The fourth is choice: avoiding lock-in. The AI landscape is moving too fast for rigid architectural commitments.

What it delivers

The results at Lumen speak for themselves. Between June and October this year, the volume of AI-driven actions processed through their Databricks environment tripled. The foundation — trusted data, documented pipelines, governed architecture — is what made that scale possible.

That's what the boring work buys you.

The billing systems aren't going away overnight. Neither are the ordering tools, the legacy integrations, or the decades of accumulated complexity. And with 88% of Americas CIOs saying legacy migration is their hardest challenge, Lumen is far from alone. But the operators who treat that complexity as a reason to delay are making a choice — and it's the wrong one.

The foundation isn't glamorous. It's just what everything else depends on.

Watch the full conversation Nobody woke up wanting eight billing systems

Read the TM Forum research: Reinventing IT for the AI era - Americas Innovate Americas 2026: Legacy IT is blocking Americas telcos from turning AI into growth | TM Forum