BT Group’s Digital unit has announced the deployment of a new internal Machine Learning Operations (ML-Ops) platform called AI Accelerator.
The new platform orchestrates, accelerates, and monitors AI model deployments developed by BT Group’s data community, assessing their efficacy and behaviour to drive value from the business’ 29 petabyte data estate. AI models are the core building blocks needed to make use of AI in practice, they deliver value by classifying and making predictions from data in ways that humans can’t.
“A core driving principle for BT Group is to find a way to safely accelerate the time to value, pound per petabyte, of data,” says Adrian Joseph, Managing Director, Data & AI, for BT Group. “AI Accelerator gives us a path to more rapid value with clear oversight of AI use case parameters and performance.” Digital’s goal is to underpin over £500m of internal value from data & AI in aggregate*, through improved customer experience and enhanced efficiency within each of the Group’s business units, over the next five years, and help power the transformation required to support BT Group’s long-term success.
Through the use of common frameworks and templates, AI Accelerator will drastically reduce the administrative and technical processes involved with taking a new AI use case from prototype to production. In keeping with BT Group’s Responsible Tech principles, AI Accelerator has built-in triggers to make sure that new AI use cases are properly assessed in line with the company’s data privacy, security and ethics principles, ensuring the safe and responsible use of AI across the business.
BT Group’s unique combination of tech talent, partnerships and the technology contained within AI Accelerator will reduce the core of this process from 6 months in April 2022, to 6 days by March 2023. This acceleration is a result of the standardisation of tools and data handling, as well as streamlined software update and monitoring, drawn from the enhanced automation and templated best practices provided by AI Accelerator, and frees up months of time for every data scientist and analyst tasked with bringing new use cases to life across the Group.
Coupled with the rise of modular, re-usable data products from within BT’s digital brain, the AI-powered intelligence layer on the Group’s data estate, it represents a major step forward in forging BT Group’s path to becoming an AI-led company.
The new ML Ops platform was built in partnership between BT Group’s growing data & AI teams within its Digital unit, and Datatonic, a leading data & AI consultancy and Google Cloud partner, and adds to the partnership BT Group has for its data with Google Cloud.
Over time, AI Accelerator will provide ongoing monitoring of AI models in production across BT Group’s estate, flagging any ‘drift’ from baseline norms in the way the AI is consuming or deriving insight or outcomes from data. In this way, it will act as a scanner for the Group’s Digital Brain, continually assessing each model’s health and prompting expert data scientists where necessary to refine and optimise them. As a result, data scientists will have more time to focus on future data science projects, driving further innovation and accelerating use of AI across the Group, instead of constantly maintaining each AI model.
“Thanks to the work the team has done accelerating our migration into Google Cloud, we’ve a healthy testbed for AI Accelerator and are seeing it catalyse AI acceleration across the Group,” says Dr Zoe Webster, Artificial Intelligence Director in BT Group’s Data & AI team. “As we progress, we’ll be able to track the health and the performance of our AI use cases in real time, ensuring consistent, safe, ethical delivery of value from our phenomenal data resources.”
AI Accelerator will allow the Group to accelerate its build, measure, learn process for new AI use cases, including, for example, churn/propensity models, or to test models that personalise our apps and services for customers, delivering better customer experiences and/or value within our business units. Many models are already live, generating value both from the identification of revenue opportunities for the Group (e.g. customer propensity models in BT Group’s Global business) and through driving efficiencies (e.g. network planning and fibre and ethernet build optimisation models for Openreach and BT Group’s Enterprise business).