Using AI to automate business processes and create revenue streams

One of the enduring features of the Covid-19 pandemic has been the acceleration we have seen in machine learning in enterprises.

According to CCS Insight’s latest Senior leadership IT investment survey, fielded in July, more than 80% of firms are now trialling artificial intelligence (AI) or have put it into production – this figure’s up considerably from the 55% reported in 2019.

AI is no longer viewed as an experimental, longer-term source of innovation for companies. Rather, it is a technology that can deliver quick transformational and business value, particularly in helping companies automate processes and create new sources of revenue.

A great example is Australia’s leading energy company, AGL. The utility represents around 30% of the total energy capacity in Australia’s national electricity market, and for the past three years has been using machine learning in a variety of innovative ways to boost automation in its operations.

Many of its 3.7 million customers use solar power and connected batteries for their household energy, and the company has developed a “virtual power plant” product to enable them to give back energy to the grid.

AGL has built thousands of machine learning models that help to remotely manage, collect and analyse metadata on energy use from each battery to better understand and forecast capacity across its network. Machine learning also automates the process of collecting, feeding and trading the spare capacity as an asset on the national wholesale energy market, generating additional revenue for the company.

It is a highly distributed environment, with each battery a rich source of metadata, but the stochastic nature of solar energy data requires machine learning on a large scale to make it work. AGL uses Microsoft Azure Machine Learning service for training and inferencing, along with other Kubernetes-based and analytical software, to enable a standardised environment for code management, automated machine learning, MLOps, and real-time performance monitoring and model retraining.

AGL’s virtual power plant has not only won several awards in furthering energy sustainability in Australia, it has also reshaped demand for power in the energy market, essentially rewarding customers for supporting the grid. This promises to improve grid reliability and help customers save on energy bills. AGL claims the underlying architecture has enabled it to train thousands of machine learning models in one 20th of the time normally required.

What is most fascinating is the level of automation at play, particularly in the potential for trading spare energy on the open market. This aspect would have been impossible on a large scale without machine learning.

Recently speaking with David Broeren, AGL’s general manager of integrated energy technology, he was bullish about the future. He highlighted that the provider was adding more data sets, such as snow levels and cloud cover, to improve its forecasting with machine learning.

He also hinted at the many new opportunities in expanding the virtual power concept deeper into homes and industries by connecting assets beyond solar batteries to the grid, such as electric vehicles, back-up generators and data centres, for example.

AGL provides a good example of the level of automation that machine learning can power. I expect many more similar examples to emerge at scale in the future.

Nicholas McQuire is a senior vice-president and head of enterprise and AI research at CCS Insight.