Telecom giant Vodafone is no stranger to the world of artificial intelligence (AI) and machine learning (ML), having used the technology for many years, with hundreds of data scientists who have built thousands of models.
While Vodafone has been able to deploy and take advantage of AI, it has increasingly faced a number of challenges over the past several years. Among the challenges was the problem of scaling its AI workloads in a standardized and repeatable manner. Vodafone also faced speed and security issues.
In a session at the Google Cloud Next 2022 event this week, Sebastian Mathalikunnel, head of AI strategy at Vodafone, detailed the issues his organization faced and what it had to do to overcome them.
[Follow VentureBeat’s ongoing Google Cloud Next 2022 coverage »]
“Vodafone is quite savvy in its data science journey,” said Mathalikunnel. “But looking back two years ago, it was actually this very issue of the size and scale of Vodafone’s data science operations that led us to believe we might have a problem on our hands.”
AI booster to the rescue
Mathalikunnel said that two years ago, he took several steps to create and run a production environment in Google Cloud for any Vodafone data scientist.
Not only were there multiple steps, but many of them were manual in nature, requiring time to set up. This situation also led to many default deployments where one data scientist’s deployment of Google Cloud AI was different from another.
He explained that Vodafone faces both vertical and horizontal challenges. Horizontal challenges were from trying to replicate workloads across markets, which was difficult because each environment was different. The vertical scaling issues were about the time and effort it took to get from a data science notebook to a proof of concept, and then into production as quickly as possible.
To that end, Vodafone developed a platform called AI Booster, which aims to help solve scaling challenges with a standardized set of tooling and processes. AI Booster relies on several Google Cloud components including Vertex AI, Cloud Build, Artifact Registry and BigQuery.
“We’re moving from a custom, coding-based approach to machine learning engineering, to an approach where everything is working based on standard components and pipelines that connect those components,” Matalikanal said. connect,” said Matalikanal.
Improving AI standardization with data contracts
Mathalikunnel noted that as Vodafone went through the process of building the AI Booster, it also identified areas where the process could be significantly improved.
For example, before AI Booster, he said that when Vodafone analyzed any ML workload it was running, about 30 to 35 percent of the code was related to data quality and data validation. was Vodafone now automates much of this work with a data contract approach.
Mathalikunnel explained that when the data is first digested by Vodafone, it triggers the analysis of the data in terms of its distribution and various characteristics, after which a contract is formed. What Vodafone then does is negotiate extensively with various stakeholders, such as data scientists and data owners, against this agreement. Once it is agreed that the data characteristics are what the stakeholders want, Vodafone puts the agreement back into the AI booster pipeline.
When the AI Booster pipeline runs, Mathalikunnel said it is able to automatically validate that the data meets the requirement it was signed off against.
One of the use cases where AI Booster has been used by Vodafone is with the company’s Net Promoter Score (NPS). NPS is a metric that aims to help predict customer satisfaction with Vodafone.
“What we’re trying to do with NPS is try to know or measure our customers’ happiness with our products,” Mathalikunnel said. “So as you can imagine, this is a very important use case for us.”