From Dumb Pipe to Intelligence Platform -
The Telco Transformation Playbook
Every operator has run the same arithmetic and arrived at the same uncomfortable answer. Traffic is compounding. Revenue is not. PwC's latest outlook puts global telecom service revenue on a path from US$1.15 trillion in 2024 to roughly $1.32 trillion in 2029, a CAGR of about 2.8 percent, while global monthly mobile ARPU is expected to tick down to $6.20 in 2029 from $6.32 in 2024. Usage soars; willingness to pay flattens. Two decades of capital expenditure have produced a business whose unit economics drift gently downward.
The industry's standard response has been to treat this as a pricing problem – bundle harder, tier more finely, chase adjacent services. It is more usefully understood as an inference problem. The network already knows an enormous amount about demand, intent, and timing. Very little of that knowledge is currently converted into anything an operator can invoice. What has changed in the last few years is not the data; it is that the analytics and machine learning stack needed to act on it has become operationally realistic.
The asset is the inference, not the data.
It is worth being precise about what the asset actually is, because the loose phrase “monetizing subscriber data” has caused a decade of confusion and no small amount of regulatory grief.
Raw network data has two awkward properties: its market value is close to zero, and its liability profile is close to unbounded. Nobody buys a table of location pings. What buyers want is an answer – who is likely to churn this month, which household is in-market for a device upgrade, which neighborhood justifies a new store. That answer is a modeled output, not an extract.
Two properties make network data unusually good raw material for producing those answers. It is consented, through terms of service the subscriber has actually agreed to, and it is network-verified, which means it cannot be spoofed or fabricated in the way that cookie- and SDK-derived signals routinely are. In a digital advertising market with a chronic fraud problem, verified-by-infrastructure is a genuine differentiator rather than a marketing adjective.
The operators furthest along have understood this as a shift from selling data to selling analysis. Vodafone's Analytics business packages anonymized location intelligence for retailers, event organizers and city planners. Orange Business went broader still, building out big data consulting and AI-based solutions for enterprise clients in finance, healthcare and logistics. In both cases, the deliverable is an insight product with a service wrapper – something that can be versioned, supported and renewed.
What operators are actually modeling
Strip away the vocabulary and the models that earn their keep fall into a small number of families.
The first is retention. Churn propensity is the oldest use of operator analytics and still the most reliably profitable, because the operator is acting on its own P&L: a subscriber saved is revenue that does not have to be re-acquired. What has improved is not the concept but the resolution – daily scoring against behavioral drift rather than quarterly scoring against demographics, and next-best-action models that choose the intervention as well as the target.
The second is commercial timing. Device upgrade cycles, plan migration, roaming, and add-on propensity are all questions about when rather than who, and the network's high-frequency signal is unusually good at when. A model that identifies the two-week window in which a subscriber is receptive is worth considerably more than one that identifies a broad segment with no timing attached to it.
The third family sits outside the operator's own P&L, and it is where third-party revenue lives: audience qualification for advertisers, footfall and catchment analysis for retailers, mobility patterns for transport and municipal planning, and – in markets with thin credit bureaus – alternative scoring inputs that widen financial inclusion. These are analytics products sold to someone else, which is precisely why they demand the governance discipline discussed below.
The bottleneck is plumbing, not modeling.
The strategic case is easy. The engineering is not, and this is where most data monetization programs quietly stall.
A modern operator's signal is scattered across Call Detail Records, OSS/BSS platforms, deep packet inspection systems, location infrastructure, and subscriber management – streams that are structurally complex and, in most estates, thoroughly siloed. Each was built for a different purpose, on a different cadence, under a different owner. Joining them into something a model can learn from is a multi-year, capital-intensive undertaking with very little to demo in year one.
