A Network Spanning 200,000 Cells Shouldn’t Require a Specialist to Answer a Simple Question
Telecom is an operationally intense business. Network performance directly affects customer experience, and customer experience directly affects retention, revenue, and brand. TELUS supports wireless connectivity for millions of Canadians across smartphones, home internet, connected vehicles, and municipal infrastructure. Understanding and acting on network performance data is a business necessity.
But TELUS faced a challenge that grows with every enterprise network: the larger and more complex the infrastructure, the harder it becomes to consistently answer basic operational questions, like “Are calls dropping?”
TELUS’s wireless network spans more than 200,000 cells operating across 3G, 4G, and 5G technologies, supplied by four different hardware vendors. Each cell reports performance data every 15 minutes, sometimes every 5 minutes, collecting thousands of metrics per cell. Every call attempt and millisecond of latency is a separate counter.
In practice, that meant teams working on customer issue resolution, capacity planning, proactive network optimization, and feature testing were starting from different numbers. A question like “Are calls dropping?” sounds simple. Answering it meant navigating thousands of vendor-specific counters across four hardware vendors and three network generations. Business teams couldn’t move without engineering.
When network experience is the product, that bottleneck is costly.
When the Data Exists but the Meaning Doesn’t
Three operational failures were compounding each other.
Inconsistent metrics across vendors made comparison challenging. With four hardware vendors using different definitions, teams could not make reliable cross-network performance comparisons. Capacity planning, optimization decisions, and customer issue resolution all depended on metrics that didn’t mean the same thing across the network.
Technical depth became a barrier to business insight. Without a centralized, standardized layer abstracting vendor complexity, non-technical teams couldn’t access the data they needed. Every insight required a detour through specialized engineering knowledge. Analytics work that should have flowed continuously became dependent on a small group of people who understood each vendor’s specific format.
Scale made manual approaches unsustainable. With thousands of performance metrics per vendor across four equipment providers, keeping data models current manually was not viable. The team needed a system that could keep pace with the network.
Legacy tooling created unsustainable modeling overhead. TELUS previously relied on a data observability platform that required a dedicated team to build and manage metric models within the platform. After an 18-month migration to BigQuery and Looker, the underlying data problem remained: how to expose governed metrics to non-technical users without recreating the same maintenance problem.
One Definition Everywhere
How do you create consistent, governed definitions of network performance across fundamentally different data sources, at a scale that can’t be managed manually?
Cornell Lee, who leads the data platform and engineering team at TELUS, argues: “Writing SQL for this is almost impossible. Even just knowing what table and what column to query is already a challenge.” AtScale’s semantic layer absorbed that complexity, translating business requirements into governed queries without requiring users to understand the vendor-specific data structures underneath.
AtScale also solved the Looker accessibility problem. Rather than requiring business users to understand BigQuery schemas or vendor-specific counter logic, AtScale sat between BigQuery and Looker as the semantic translation layer. The architecture operates in two tiers, designed to serve different audiences without compromising either.
At the technical level, low-level models captured vendor- and generation-specific data in granular detail, built for engineering users who needed precision. These models were automatically populated: Python scripts parsed vendor-supplied documentation and converted it into standardized measures, which were loaded directly into AtScale, eliminating the manual maintenance cycle.
Take VoLTE Drop Call Rate (DCR) as an example. In AtScale’s Semantic Modeling Language (SML), it’s defined simply as total drops divided by total calls, vendor-agnostic terms that stay consistent regardless of which equipment vendor a subscriber is on. Behind that single definition, each vendor’s underlying counters may differ significantly: one vendor may report a drop as a single counter, while another may combine several counters. SML abstracts all of that. The metric means the same thing everywhere.
“The KPI that an executive is looking at, this is the exact same as what an engineer would troubleshoot, and it is what we would detect anomalies on.”
— Cornell Lee, Data Management, Data Platforms & Engineering, TELUS
Consistent Numbers Across the Network
The AtScale implementation resolved TELUS’ data challenges and resulted in:
Consistent metrics, finally. Teams across the organization now access the same governed definitions. A call drop rate means the same thing regardless of which vendor’s equipment is being evaluated or which tool the analyst is using.
Complex data became broadly accessible. Technical complexity stopped being a prerequisite for insight. Business teams that previously waited on engineering specialists can now access trusted metrics directly in Looker or via Python without vendor-specific knowledge.
A foundation for continuous growth. The automated modeling approach means TELUS can onboard new data sources and new technologies without starting from scratch. The semantic layer scales with the network.
Teams that were siloed now share a single source of truth. Engineers and business analysts working on different aspects of network performance now operate from a common, governed foundation, enabling cross-functional analysis that wasn’t possible before.
“At TELUS, we say ‘Let’s make the future friendly.’ Our semantic layer is letting us scale up our analytics with each new use case we add. That makes our teams more effective, and makes for a better experience for our customers.”
— Adam Walker, Senior Design Specialist at TELUS
What Governed Analytics Actually Buys You
TELUS set out to answer a question: what is our users’ experience of our network? The semantic layer is what made that answerable consistently, at scale, for every team that needed to know.
Now every strategic decision and investment at TELUS is driven by trusted, quality data. The universal semantic layer gives the team one place where metrics are defined, and one answer that travels across every team and every tool that needs it.
“We’re starting to view this semantic layer as more than just a piece of infrastructure or architecture, but it’s really the way that we can create and serve data as a product across our entire organization.”
— Cornell Lee, Data Management, Data Platforms & Engineering, TELUS
