Arize AI > Case Studies > Leveraging Machine Learning for Enhanced Telecommunications Services: A Case Study of Spark New Zealand

Leveraging Machine Learning for Enhanced Telecommunications Services: A Case Study of Spark New Zealand

Technology Category

Analytics & Modeling - Machine Learning
Platform as a Service (PaaS) - Application Development Platforms

Applicable Industries

Equipment & Machinery
Telecommunications

Applicable Functions

Procurement
Product Research & Development

Use Cases

Predictive Maintenance
Time Sensitive Networking

Services

Data Science Services
Training

About The Customer

Spark New Zealand is the country's largest telecommunications and digital services company. The company provides mobile, broadband, and digital services. Over the past decade, Spark New Zealand has launched several new digital services, including Spark Health, which supports the digital transformation of the health sector. The company's senior leaders set an ambitious goal to help all of New Zealand win big in a digital world. To achieve this, they aim to understand what New Zealanders want and what their needs are at a granular level. Machine learning plays a crucial role in this endeavor. The company has over 50 models in production and a team of over 20 data scientists and machine learning engineers.

The Challenge

Spark New Zealand, the country's largest telecommunications and digital services company, was faced with the challenge of understanding their customers' needs at a granular level to provide better services. The company aimed to expand the number of machine learning (ML) use cases across the organization to achieve this goal. They started their journey in machine learning by trying to predict churn and understand customer preferences. However, as the number of use cases and the size of the team expanded, they faced issues with model performance and monitoring. The dynamic nature of data and the need for continuous monitoring and troubleshooting of models posed significant challenges. Checking the performance of over 50 models every week was a tedious and time-consuming task. The company needed a solution that could help them monitor these changes more effectively and proactively approach the output of models.

The Solution

To address these challenges, Spark New Zealand adopted the Azure ML platform for end-to-end model development and productionization. They also implemented Arize, an ML observability platform, to monitor and troubleshoot model performance. Arize was selected for its ease of use, ability to calculate and visualize data drift, and its alerting capabilities. The platform also offered tools for fairness and bias, performance tracing, and explainability. With Arize, Spark New Zealand was able to monitor models and better understand and improve them. For example, they could analyze whether a customer was leaving because of the service or some other reason, and if the features captured that behavior. If not, they could do some feature engineering to focus their attention on ways of improving the model based on feedback. The company also used Arize to tie model prediction to business return on investment or other key performance indicators (KPIs).

Operational Impact

The implementation of machine learning and the adoption of the Arize platform have brought significant operational benefits to Spark New Zealand. The company has been able to move away from a reactive strategy to a more proactive approach in managing model performance. They can now monitor models and better understand and improve them, leading to enhanced customer understanding and service delivery. The company has also been able to tie model prediction to business return on investment or other key performance indicators (KPIs), ensuring that their machine learning initiatives are delivering tangible business value. Furthermore, the use of Arize has made it easier for the company to present technical material to a wider group of stakeholders, enhancing understanding and collaboration across the organization.

Quantitative Benefit

Improved marketing efficiency by almost sixteen percent year-on-year through in-house data capability.
Over 50 models in production running on a weekly or monthly basis.
Over 20 data scientists and machine learning engineers managing the models.