Diffusion, Not Adoption
The telecom industry has spent two years counting AI pilots. The more important question is how AI is changing what the network does, how customers experience it, and the role of commerce channels.
At Mobile Future Forward this month, Chetan Sharma reframed the AI conversation. The industry keeps measuring adoption: which companies have deployed tools, how many pilots are running, and what percentage of employees have access to a model. But adoption is just the beginning. The more meaningful question is diffusion: how has AI actually changed how value is created and captured?
The distinction matters because diffusion is harder to see and harder to fake. Innovation can be copied quickly; diffusion takes time and ecosystem-level coordination. That’s why Sharma makes the case that AI experiments are not the right measure on their own. More meaningful signals include revenue generated from computing and AI services, or the convergence of 5G/6G, sensors, compute, and AI creating what he called emergent intelligence: capabilities that arise from the network itself, not from any single node or application.
“Deployment is not diffusion. Measuring the wrong metrics leads to unintended consequences.”
—Chetan Sharma, Mobile Future Forward Keynote
That call to action resonates for me and my colleagues at iQmetrix, where we are actively implementing AI strategies that change what an operator can sell, how customers experience your brand, and what is happening at the moment of sale. The following takeaways from the event support the notion that the next wave of AI technology is driving material change to the consumer experience.
Physical AI
One of the overarching forward-looking points of discussion centered on AI-native radio access networks. The premise: AI in the RAN layer does not just optimize spectrum efficiency; it makes the network a platform for physical AI at the edge. NVIDIA and T-Mobile have been building toward this by integrating physical AI applications directly on AI-RAN-ready infrastructure. Robots, sensors, and dexterous systems are examples of intelligence that lives in the network rather than the cloud, operating with the low latency and local compute that real-world physical environments demand.
The direction is clear: the RAN layer is becoming a compute layer, and new services, including things we do not have names for yet, will be built on top of it. For consumers, physical AI will become embedded in day-to-day activities, from traffic management systems to robots that clean homes and need to know precisely where they are in real time.
AI diffusion in device management
Assurant’s session discussing diffusion, “enterprise beyond the POC,” was the most concrete proof of diffusion already happening inside the telecom supply chain. Three live AI applications are running at thousands of devices per day: robotic cosmetic grading that replaces subjective human inspection, a system making real-time routing decisions on whether a device should be repaired, recycled, or liquidated, and an ML-driven auction engine that auto-accepts, declines, and counters offers across tens of thousands daily device transactions. None of this is a pilot; it is the operating model. This kind of innovation not only exemplifies AI monetization in action, but also fundamentally impacts consumer choice and the new and used mobile handset economy.
Unlocking the BSS
The last panel discussion of the morning session with Chetan put a name to a problem that every operator in the room recognized: AI cannot help if the data is locked in legacy BSS and OSS systems. Telecoms must extract context into an ontology layer without touching the underlying platforms, allowing the lifecycle to retire legacy systems on its own timeline rather than forcing a rip-and-replace that nobody has the budget or appetite for.
The example that landed was order fallout caused by an inability to access inventory context. Fix the data flow, and you fix the customer experience. That improvement eventually shows up in store wait times, in how well-informed a care rep is before a conversation starts, and in whether a customer’s interaction ends with a solved problem or a transferred call. The data is the infrastructure. The agentic commerce layer is where relevant insights either shows up or does not. This point landed squarely with a strategy we have already put into practice: our Intelligent Commerce Operating System pairs with BSS vendors to unlock inventory visibility, advanced customer insight, and customized product and pricing recommendations that drive higher sales and customer satisfaction.
Measuring what actually matters
The right metrics for this moment are not only adoption metrics. They are diffusion metrics: what share of revenue comes from AI-enabled services, how many devices move through AI protocols versus human interventions, how much faster a customer gets through a transaction based on real-time AI-generated insights, and how much a care rep knows before the conversation starts. Those numbers will tell us whether the transformation is real.
The optimism at Mobile Future Forward was genuine. The use cases are real, the investment is serious, and the pace is faster than most of the industry expected. But Chetan’s framing is the right one to carry forward: innovation is the easy part. Diffusion, getting the whole ecosystem to move together, is where the real work happens. That’s the journey we’re on.
Get The Overview straight to your inbox for insights on what’s transforming telecom commerce.
