Australian retailers have progressed in their understanding and ability to realise value from advanced analytics – and particularly artificial intelligence as a subset of that – over time.

Back in 2015, only 36% of Australian retailers believed they were effective at actioning insights from data analytics.

But even then they had the makings of good analytics programs – namely, a flow of real-time data that could be used to train and refine advanced analytics models. The bottleneck was in retailers’ ability to effectively make use of it.

A lot has changed since then, notably a sustained investment by retailers in their digital and ecommerce operations, and a huge increase in the number of advanced analytics tools and models available for use.

The target use cases for advanced analytics have also been refined. While the last few years have seen retailers focus heavily on building out their digital capability and understanding how customers interact with these properties, much of the current work is around omnichannel, where digital complements the in-store experience and vice versa, and where moving between channels is seamless.

As KPMG Australia noted recently, there’s a specific focus in 2023 on analytics that can boost the bricks-and-mortar shopping experience for customers, through “the use of predictive analytics and real-time triggers to drive foot traffic and stimulate purchases”.

“Transactional, behavioural and motivational data points are already on many retailers’ radars, but to enable inbound personalisation, these measures need to be extended to real-time contextual data from apps, mobile devices and geo-location,” KPMG wrote. “Investment in deep analytical capabilities is also required, to support the coordination and curation of data in a decisioning engine.”

What’s notable about this is both the specificity and the sophistication of the use case. These are important factors that demonstrate growing maturity of advanced analytics thinking in the Australian retail space.

It also creates an opportunity to have a deeper discussion on what advanced analytics options are actually available to retailers today to achieve the desired outcomes.

Understanding the pros and cons of the various models is important to determining suitability. It also affects how IT infrastructure is best set up and configured to support a specific analytics use case.

Supporting the needs of different models

Within advanced analytics, artificial intelligence is probably the most discussed and experimented with. The mainstreaming of generative AI – popularised by chat-based interfaces like ChatGPT – has put the spotlight broadly on what additional value AI can bring to retailers.

AI itself is a broad umbrella term. Within AI, there are typically five types of models that are used today: machine learning, deep learning, recommender systems, natural language processing (NLP), and computer vision. Understanding each of these AI types and the types of data they interact with is important when looking at what data storage infrastructure would best support it.

Machine learning workloads typically utilise a tremendous amount of data, especially during training. More data usually means higher accuracy for the models. These models require a data storage platform in the background that can scale to meet the demands of the base data, while also growing as more data is collected and sampled. The storage also needs to be able to handle high input/output (IO) of data, since large amounts of data are passed from the storage to the model and back to enable it to function.

Deep learning is more intensive, and tries to mimic learning models of the human brain. It requires a tremendous amount of processing power and data to be effective, and the infrastructure behind it has to be able to continuously scale to meet the model’s demands.

Recommender systems will be familiar to retailers as they’re often used on ecommerce sites to recommend products and services to consumers based on analysis of past consumption patterns. These systems use millions of data points across the history of a user, and a culmination of data points across similar users, to predict future needs and purchase behaviour. They’re also unique because most of the decisions made are based on real-time information, and that requires powerful support systems in the background to enable them.

Natural Language Processing (NLP) will also be familiar to retailers as it’s used in chatbots and voice search apps like Alexa or Siri, which retailers may maintain channels on. These models listen and convert speech-to-text or text-to-speech to extract meaning and respond appropriately. From a storage perspective, such models are sensitive to IO; but depending on the use case, they’re also extremely latency sensitive. Microsecond or lower latency is important to ensure the responses to customer questions are fast.

Finally, computer vision is also seen in retail experiments. Woolworths, for example, uses it for in-store merchandising and for fresh produce recognition on scales and self-serve checkouts. Because these AI models use video to understand and interpret the world into something actionable, they are by far the most significant when it comes to the volume of data processed.

AI workloads vary in size and complexity. Their data, IO and throughput needs range from modest to massive, and their performance requirements cannot be underestimated. By making some strategic decisions around dataplatforms, Australian retailers will be best placed to get the most out of their AI investments.

Nathan Knight is vice president and managing director for Australia and New Zealand at Hitachi Vantara.