By: Frank Ploumen

Telecom has spent billions on AI and has little to show for it — not because the tools don’t work, but because the tools don’t talk to each other. Most service providers have run pilots, stood up proof-of-concept tools, or deployed a point solution bot or automation. What they haven’t done is build AI into the fabric of how they run their business. 

The pressure to move is real and understandable. But what’s often happening now is that urgency is driving decision-making without a long-term framework to support it. The result is a lot of activity and very little operational transformation. 

Understanding why — and what to do about it — is the difference between building a genuinely competitive business and accumulating what I call “AI orphans,” or isolated AI investments that work in a narrow context but can’t scale, can’t integrate, and can’t sustain the outcomes they were designed to create. 

Most AI Deployments Are Trapped Inside Silos 

The reason most AI isn’t truly operationalized isn’t a strategy or intent problem. It’s an architecture problem. 

Consider what this looks like in practice. A decade ago, everyone rushed to connect devices to the internet. Lights, thermostats, doorbells, smoke sensors, each with its own app, its own interface, its own logic. The result wasn’t a smart home. It was a connected home with the user in the middle, managing 27 apps that couldn’t coordinate with each other. The platform integrations that eventually made smart homes actually smart took longer to build, but they’re what survived. The point solutions got ripped out and replaced, and anyone who built a business around them was left scrambling. 

AI is following the same trajectory, only faster. 

Teams across telecom organizations — network operations, customer service, finance, marketing — are each experimenting with their own AI tools. Each team builds something that works within its own vertical, creating the illusion of progress. A chatbot that deflects password resets. A churn model that flags at-risk subscribers. A campaign tool that improves acquisition targeting. Each solution generates real results but doesn’t integrate or connect with the others. 

The output of one system becomes a report, a graph, or a PDF — something that a human carries across organizational boundaries and presents in a meeting, where other humans debate it and eventually make a decision. The AI stopped working the moment it reached the edge of its silo. Despite all the automation, there’s still a rate-limiting step: a PowerPoint deck or a conference room. 

This is the AI orphan problem in practice. Not a single dramatic failure, but a slow accumulation of point solutions that each solved a narrow problem and foreclosed on the possibility of solving a bigger one. If you deploy assets in your operational network that are one-offs, they become very hard to maintain. You end up slowing down your overall business in the long run because you have so much burden and so many touch points. 

Real Operationalization Starts With Outcomes, Not Tools 

The antidote to the AI orphan problem isn’t slowing down. It’s reorienting around outcomes before reaching for tools. 

In telecom, there are essentially four outcome categories that matter for service providers: acquiring new customers, maximizing revenue from existing subscribers, reducing churn, and optimizing the cost to deliver service. Every AI use case in the industry maps to one of these four buckets. So the question isn’t which tools are available to address issues in each bucket, it is which outcomes are most urgent, and what architecture is required to serve them at scale. 


That second question is harder to answer. It requires conversations across departments and demands an honest audit of which legacy systems can participate in an integrated workflow and which ones will become the weakest link in the chain.