Viewpoints

Telecom networks are closer than ever to being able to fix their own data, and the implications are significant.

Submitted by: David Cottingham, CTO, IQGeo

Telecom operators run their businesses on a digital representation of the physical network that shows where every cable, splice, and customer connection should be. With an emphasis on “should,” because keeping that model accurate has been a losing battle. As the network changes constantly in the real world, the gap between what operators think they know about their network assets and reality grows wider. And until recently, operators had no feasible way to continuously find and fix those discrepancies at scale.

Advances in visual AI are starting to change that. Operators can now automatically compare what their records state should be in the field with what is actually there. This is the first step toward automated data quality at scale. The next is network models that not only detect these errors but update their own records accordingly. Telecom networks are close to being able to fix their own data, and the implications are significant: lower operating costs, faster service delivery, and a more reliable foundation for automation across the network.

The high cost of inaccurate network data

Everyone knows that poor network data comes at a high financial cost. Field crews get dispatched for unnecessary work or make repeat visits because the records mislead them. Revenue leaks out when services can’t be billed correctly or when unscrupulous contractors charge twice for the same connection work. Opportunities are lost when sales teams believe a new customer location is serviceable because the database says so, only to discover the infrastructure isn’t there.

Those costs become harder to control as networks grow through acquisition. Over years of growth and consolidation, many tier-one operators have inherited a patchwork of legacy networks and databases, with every merger and acquisition delivering a “black box” of network data that may be incomplete or out of date. The acquiring operator often discovers that the new assets are not adequately documented – if 20 companies were acquired in two decades, that’s 20 different record-keeping styles (and 20 sets of errors) now under one roof.

The UK provides a clear example of how these pressures are playing out. After years of rapid build, the “altnet” sector is now entering a phase of inevitable consolidation, with industry analysis showing collective losses of around £1.5 billion as smaller fiber players struggle to reach sustainable scale. In that environment, the quality of network records becomes a material factor in both integration cost and deal valuation, as acquirers discount prices to reflect the risk, time, and manual effort required to understand what has actually been built.

The financial impact does not stop at internal inefficiency. Operators without an accurate network model also face a growing risk of customer churn, because data accuracy directly underpins service reliability. When records are wrong, outages last longer, provisioning slows down, and avoidable errors reach the customer. This risk is becoming more acute as competition extends beyond traditional telecoms. Low-Earth orbit satellite providers are beginning to offer credible broadband alternatives in areas once served by a single terrestrial network. In this environment, network performance and service reliability become decisive.

Checking reality against records at scale

As these market pressures mount, the question for operators is how to achieve the scale of data clean up needed. Sending out crews to manually audit data is logistically unviable. And even if operators could find a way to deploy large enough teams to verify the state of their assets across their entire network, the cost would be astronomical.

Cue the growing interest in visual AI, which large operators are already using to both correct erroneous data and keep it correct over time. Today, visual AI analyses images taken by field technicians, scans for key details (like equipment types, connections, port usage) and cross-checks them against the system of record. If a discrepancy is found, say a fiber is connected to a different port than the records indicate, the system flags it and a human can either accept or reject the proposed correction. Visual AI is already deployed on parts of the network at major operators like Virgin Media O2, where visual AI is being used as a quality gate in the fiber build process. Field work is verified at the point of completion, before it is signed off or paid for, avoiding the need for large-scale, manual audits months or years after deployment.

Teaching networks not just to detect, but fix bad data

In most use cases today, visual AI systems act like inspectors, notifying operators of inconsistencies in their network data and suggesting corrections. Over time, humans may choose to hand over this decision-making responsibility to the network, essentially giving it the ability to maintain an accurate model of itself. We’ll move from today’s verify-every-suggestion model to a more autonomous mode where, for many routine discrepancies, the system just fixes them.

To reach that point, three things have to happen. First, visual AI models must be trained on telecoms-specific datasets to perform reliably across different network architectures, vendors, and regional build standards. Second, automated corrections must be governed by clear operational rules, with full auditability, so operators can trace what was changed, why it was changed, and what evidence was used. And third, these systems must be tightly integrated into the system of record, not operating as a bolt-on analytics layer but as part of the core workflow that updates network data in near real time.

The ability for network operators to correct bad data as part of normal business operations is hugely significant, both for smaller players hoping to be acquired, but also for tier-one incumbents with decades of data. As they look to deploy more AI technologies, the only way that these will add value is with a much higher degree of network model accuracy.

A new era of automation

The ability to keep network records in sync with the real world is the crucial first step toward fully autonomous networks that not only maintain their own data but start to manage their own operations. In the future, in addition to helping identify data discrepancies, visual AI will also allow operators to use images taken by field engineers to identify maintenance and monetization opportunities and create tickets for teams to address them.

Momentum is already building around visual AI in telecom operations because it delivers measurable returns where many AI initiatives fail to move beyond pilots. Data from MIT shows that only around 5 percent of organizations translate generative AI experimentation into real business impact. Visual AI avoids that trap by targeting a clearly defined operational problem. By automating tasks that are otherwise manual and error-prone, operators can realize benefits quickly, including fewer repeat site visits, reduced audit effort and faster job completion.

Network models that can fix their own bad data are the essential next step in extending that autonomy more broadly, because they establish the trustworthy foundation on which higher-level automation can reliably run. Without accurate, self-maintaining data, autonomy remains limited in scope. With it, the notion of dispatching audit teams to manually correct documentation errors will feel as outdated as paper maps in the age of GPS.

Get news like this in your inbox. Subscribe to the Broadband Communities newsletter!