By Przemek Tomczak, Director of Industry, Partnerships and Alliances at Ocient
Last week in Dallas, I had the chance to attend TM Forum Innovate Americas (Sept 10–11, 2025), and one line has been echoing in my head ever since:
“Data science is not a science project anymore — it’s about business.”
-Bert Lacher, Technology Solutions at Verizon
Or, as Mano Mannoochahr, Chief Data, Analytics and AI Office at Verizon put it: “When you call Verizon, AI is already at work before the call connects.”
Those statements perfectly captured the shift I saw across nearly every session at the conference: AI has moved beyond the lab. The conversations weren’t about pilots, proofs-of-concept, or abstract hype. They were about execution, scale, and measurable impact.
Take AT&T, for example, which reported more than 90% cost savings using fine-tuned small language models and a 20% boost in coding efficiency.
For those of us working at the intersection of data, AI, and telecom, the message was clear: the era of experimentation is giving way to the era of real business outcomes.
Start with the Business Problem
The most compelling stories didn’t begin with, “We trained a model.” They began with a business challenge:
-
Reducing customer wait times
-
Preventing network failures
-
Defending against new cyber risks
As Himanshu Polavarapu, Associate Vice President of Enterprise Architecture at Verizon, reminded us: “If you do experimentation in isolation, it’s not going to yield value… it becomes garbage in, garbage out.”
Too often, organizations get enamored with AI for AI’s sake. The projects that break through are the ones laser-focused on solving tangible problems. The technology only matters if it moves the needle on customer experience, operational resilience, or financial performance.
Why So Many Are Still “Stuck in the Gap”
Despite progress, many companies admitted they’re still stuck in what I’d call pilot purgatory — where concepts are proven but not industrialized.
The culprit? Weak foundations.
Three themes came up repeatedly as non-negotiables for scaling AI:
-
Data quality and trust. Clean, structured, well-governed data is the fuel. Without it, you’re just optimizing noise.
-
Composable, interoperable systems. Legacy stacks and siloed platforms can’t support the agility enterprises need. Modern architectures are table stakes.
-
Low latency. Particularly critical for real-time, agentic applications. If systems can’t respond quickly, the use case collapses.
Skip these, and you’ll never get from prototype to production.
Where AI Is Delivering Real Value Today
One of the highlights of Innovate Americas was hearing about concrete, already deployed use cases. These weren’t speculative — they’re delivering ROI today.
AT&T shared that its AskAT&T platform already serves 100,000+ users, processes ~5 billion tokens daily, and cut costs by more than 90% by moving from large to fine-tuned small models with RAG pipelines.
Examples included:
-
Customer Experience – Generative assistants that summarize calls, recommend next-best actions, and route customers faster. At Verizon, AI summarizes call reasons and suggests three likely actions before the agent even answers.
-
Networks – Self-healing systems, predictive SD-WAN health scoring, and anomaly detection preventing outages before they happen. AT&T already has 91 production RAG+FT use cases, with 250 more in development.
-
Enterprise Productivity – AI copilots for coding, HR assistants answering policy questions, and AI-powered SDLC platforms. AT&T has seen a 20% gain in coding efficiency and a 30% improvement in code acceptance rates, reducing manual changes.
-
Security – Defenses against synthetic phishing, prompt injection, and model poisoning. Some organizations are even red-teaming their own AI systems to harden resilience.
What stood out: these weren’t moonshots. They were practical improvements to operations, productivity, and resilience.
Standards and Governance: Enablers, Not Obstacles
Standards and governance were another recurring theme.
For interoperability, TM Forum’s Open Digital Architecture (ODA) is gaining traction. Shared schemas and ontologies are no longer academic — they’re the connective tissue enabling ecosystems of vendors and operators to scale without fragmentation.
And governance, often seen as a blocker, was reframed as an accelerator. This means:
-
Cataloging models
-
Embedding guardrails
-
Operationalizing oversight
As Martin Kessler, Vice President, Deputy CISO and Head of Enterprise Cybersecurity Services at Verizon, put it: “Operationalized governance isn’t a barrier — it’s how we go from pilot to production with confidence.”
These steps don’t slow innovation — they create the trust and consistency needed for innovation to move faster.
The Cultural Shift Ahead
Technology wasn’t the only focus in Dallas. Culture took center stage, too. One line from Verizon’s Shamik Basu summed it up: “If you hate the system, automate it.”
That mindset reflects where we’re heading. The next phase of AI adoption won’t just automate workflows — it will empower AI agents to interact dynamically with enterprise data by:
-
Issuing more queries
-
Running complex queries in parallel
-
Making decisions on the fly
To support that future, latency, governance, and data foundations will matter even more. Companies that don’t invest in these areas will fall behind as their peers unlock new levels of automation and intelligence.
Final Reflections
What struck me most leaving Innovate Americas was the maturity of the conversation. AI is no longer about “magic” — it’s about operational reality.
The playbook is becoming clear:
-
Start with outcomes, not models
-
Build the right foundations: data, systems, latency
-
Treat governance and standards as accelerators, not obstacles
-
Foster a culture where automation is embraced, not feared
For telcos and enterprises alike, that’s the path from pilots to production, from hype to outcomes.
To end, I’d like to share a final quote from AT&T.
“Agentic AI isn’t magic. It lets us reimagine processes we couldn’t touch before.”
-Andy Markus, Chief Data Officer at AT&T
That reimagination is already producing measurable results: AT&T has achieved a 2X improvement in free cash flow ROI per $1 invested, with expectations to scale to 4–6X over the lifecycle of its AI programs.
In side conversations, what stood out most was the challenge of network data — not just storing it, but making it actionable at scale and in real time. Many are still figuring out how to align cost, performance, and latency so AI can evolve beyond point solutions into truly agentic, cross-workflow intelligence.
The promise of AI is real. But the bottleneck isn’t just the models — it’s the ability to harness data at the speed and scale business demands.
The industry is ready to make that leap. And judging by the conversations in Dallas, the leaders who do will define the next era of telecom and enterprise transformation.