By Alex Goncharuk, VP Global Retail at Intellias
The emergence of effective Generative AI (GenAI) solutions over the past year has fired the starting gun in the race for AI dominance. In terms of retail, it enables Dynamic Pricing, Product Customisation, and many other consumer-friendly innovations. However, organisations that are only just considering integrating GenAI now are already 1-3 years behind. They must understand that data is the most fundamentally important basis of any successful GenAI deployment. AI needs meaningful data to generate beneficial output – known as ‘garbage in/garbage out’.
The first step is to prepare for data collection and quality control and how to maintain oversight of the output. No matter what, ‘garbage data’ will always find a way to creep in, so gatekeeping is essential. Constant monitoring will enable businesses to eliminate poor data as it is inputted into the system.
One of the significant challenges with this approach is that most retailers have multiple customer touchpoints, which are often siloed, meaning data cannot be easily shared. For GenAI to work effectively, this data must be in a collaborative network, with each touchpoint connected point-to-point in a comprehensive web of communication. If organisations don’t link all these data points, they could lose a substantial amount of significant data.
Building the right ML model
Data won’t go anywhere unless directed, and once it is in the system, it needs to be maintained and controlled to provide optimal value. For GenAI, that typically means introducing Machine Learning (ML) and MLOps, a core function focused on streamlining the process of taking ML models to production and then maintaining and monitoring them. Retailers should also consider infrastructure on demand to scale their operations up quickly and affordably to handle more data.
Once companies have the data model and the infrastructure in place, gathering the data can pose additional problems, particularly around compliance with GDPR/CCPR regulations. Some organisations, including in the retail industry, accidentally collect too much data because they are unaware of what is required, leading to privacy issues. Companies should also consider healthcare in this context because retail environments, such as pharmacies, deal with much more sensitive data and know much more about their customers than typical high street stores.
Dynamic pricing is the real prize for retailers in an AI-enabled environment. However, it is only achievable if organisations fully understand what customers will willingly pay and how much the item can sustainably be sold for. GenAI is the pole star in this respect, as it can utilise the data to learn customer behaviour and adapt prices and promotions accordingly, considering individual preferences and external factors, such as seasonality.
AI can also help customise product descriptions and contracts to make them more appealing to consumers. By having access to search history, what has been added or removed from the basket, and which pages users spend the most time on, AI can build a remarkably accurate picture of each customer and what they might want.
Putting data first – carefully
Retail organisations that are only beginning their ‘Data First’ journey now will only bear witness to the benefits in 2-3 years. However, those organisations with data-first strategies that have adopted AI and ML will be well on their way. AI and ML have existed longer than you might think, with the most significant difference being that data storage is now much less expensive, meaning organisations can store, sort, and organise much more data much more quickly. That means more customisation and more insight into customer behaviour.
Before retailers proceed, however, they should reflect on the case of US behemoth Target, which, around ten years ago, used all the data it collected on loyalty cards to build up detailed profiles of each customer. So successful was the company that it could identify pregnant customers simply by what they were purchasing. However, this wasn’t just nappies or formula milk; in some cases, Target could tell a woman was pregnant before she knew herself because her pre-pregnancy purchases matched with other women who then became pregnant. This level of accuracy can be off-putting and potentially dangerous, particularly when such sensitive information is at stake.
Ultimately, there is no getting away from the fact that big data is now the central platform for growing retail businesses, so if a retailer is yet to get on board, it is time for them to make up for lost time. But they must remember to handle that data with extreme care.