Artificial Intelligence (AI) is changing retail. There are many ways that AI has been used to help retailers sell more. However a more recent phenomenon is that of using AI-aided clienteling where the sales associates can make decisions based on both their selling intuition and the insights provided by the AI.
First, let’s establish what is meant by AI in retail; AI is a set of algorithms that uses pre-collected data (or a constant ‘live’ stream of data) to come up with insights and predictions. Here’s our guide for retailers looking to optimise the performance of their sales associates using AI a) by being better targeters and b) by being better recommenders.
AI and Consumer Targeting in Retail
AI in retail can be a powerful tool for aiding consumer targeting. When profiling a consumer and their shopping habits, there can be a lot of data that needs to be understood so that a customer can be correctly profiled. Due to the sheer amount of variables to consider, it can be a mammoth task for a person (or even a group of people) to process the data. This is where AI steps in. Since most retailers have a clear understanding of their ‘elite’ customers (customers that spend a lot of money on a frequent basis), their data can be used to help train AI to identify other potentially ‘elite’ customers that can be targeted by sales associates. Once trained, the AI in retail can consistently identify customers with similar demographic attributes to the already identified ‘elite’ customers. With this identification process being automated, the sales associate only has to find a way to impress and influence these customers.
In order to find such customers, the system should classify all the customers using the data available to them (for the tech savvy reader, examples of classification algorithms are K-nearest neighbours, Naive Bayes and Random Forests). This can be further improved by incorporating some reinforcement learning into the mix so that the certain variables in the classification algorithms can continually be toggled for optimal consumer identification in a changing market landscape.
AI and recommendation systems
Associate Rule Learning
In order to effectively encourage customers to buy more, brands and retailers must introduce shoppers to products that they will want to buy. Of course it can be possible for fashion retailers to use stylists and associates to deduce the particular ‘style’ or taste of the targeted customer, however, that can be quite impractical for retail operations larger than a local boutique store. Luckily, AI in retail provides an alternative that is equally effective at understanding the taste of the client. Via association rule mining, it’s possible to find patterns between certain products and realise the possible ‘styles’ that encompass the various buying patterns that consumers show.
An example of this is the Eclat algorithm which puts groups of all products (that can potentially be brought together) into sets. Therefore, some sets could include every product in the catalog whilst others would include only one type of product. These sets are then sorted based on the amount of occurrences in the transactional data and their size (typically single-product sets are filtered out).
Using that, a sales associate can be presented with the recurring patterns and (by using their rapport with the customers) make an additional sale. Another benefit of having a human in the loop of the association rule mining is that it allows the sales associate to remove any information that they deem irrelevant. A smarter system would take the feedback of associates and use it to power a reinforcement learning algorithm that learns to make better associations with time and feedback.
Generative Adversarial Networks
A much more ambitious way to recommend clothes could be via GANs (Generative Adversarial Networks). A GAN consists of 2 neural networks and a sample of real world data: the generator which outputs potential samples and the discriminator which decides whether or not the output is real world data or the creation of the generator.
After each round, the generator and discriminator would try to outdo one another and in the process, the output could even be hard for humans to distinguish as real or created by the generator. Think of it as counterfeit art with a historian trying to distinguish whether it’s real or not. Interestingly, Nvidia used this technique to generate pictures of celebrities that never existed by using pictures of those that do exist and inputting them into a GAN. Similarly, retailers can input a set of curated outfits so that the GAN can output a set of AI-generated outfits. After thousands of rounds, the GAN could potentially output the outfits at a human level of competence.
How AI can affect your employees?
It should be noted that some of these algorithms (especially GANs) are currently at the cutting edge of AI technology. This means that to accomplish this project, companies would have to hire the best talent that the technology has to offer. Furthermore, machine learning algorithms require infrastructure to run upon and trying to host these programs in-house would not be cost effective, and hosting on a service such as AWS would require employing data engineers to determine the optimal data flow. All these problems can be avoided by outsourcing a lot of the implementations discussed above.
Most attempts fail not because of the inability to create the infrastructure needed to modernise, but due to lack of adoption of the tools needed to modernise. As tempting as it may be to implement many changes at once, if long term adoption of these changes is not achieved then all the prior efforts go to waste. Retailers must analyse their ability to handle change before determining the pace at which they are implemented. It is also important to pace the changes in a way which can be compared with how a long distance runner paces themselves during a race. If the pace is too fast, adoption could suffer and if the pace is too slow, they might fall behind. It is also important to curb any technophobia via training and continuous exposure.
Creating an in-house implementation of AI is a rather difficult and capital intensive affair. We, at Proximity Insight, are working to implement AI systems into the future of our clienteling app.