By Manuel Davy, Executive President and Founder of Vekia
With a design based on advanced automatic calculations, Machine Learning has become a fundamental tool for companies using Big Data in their decision-making processes. The growing number of connected objects used in everyday life touches on most activities and is now at a decisive stage of maturity. Retail is just one sector that is substantially impacted by this challenge and the digital revolution is now affecting all operational systems of production.
A classical algorithm, however powerful and precise, executes only what it was programmed for and is limited to instructions previously set.
One of the primary differences between classical programming and Machine Learning-based programming is that the execution of tasks is always automated in classical programming. At no point does the program possess the capacity to enhance its own data modelling or retrieve new data. As a result, there is no room for innovation in the program process unless modified by a user in the forms of updates or new versions.
The notion of automatic learning, specific to Machine Learning, is essential. Machine Learning is the set of processes used to develop or modify behavioural models influenced by activity and experience.
Automatic learning and Artificial Intelligence (AI), are not new concepts. There are a few factors that aide a more effective implementation and revolutionizing its use. Two of these factors are:
- Big Data. By definition, Machine Learning feeds on data and needs this data permanently and in high quantity. The multiplication of information, its diversity and availability vastly increase the analysis and learning capabilities. The computing power of today’s computer systems enable us to process these data quickly.
- Interaction of connected objects. Computers, tablets, smartphones, watches, handheld terminals and other everyday objects are now connected. These items have become permanent collectors and transmitters of consumer data, habits, preferences and behaviours.
More than ever, professionals and consumers alike are connected and they participate in the constant feeding of systems that analyse our behaviours and buying habits.
When paired with the raw material provided by AI, there is no longer the need to manually feed the data as it is automatic and instant.
Applying Machine Learning to the World of Retail
In the current context of innovation, faster time-to-market and product customization forces retailers to rethink their supply chain and inventory management to not penalize their growth.
With Machine Learning, the data is vital to contribute in the creation of the statistical model to be applied. This contrasts with conventional algorithms since the model is defined before being applied to the data.
This difference is significant since it transforms an empirical scheme to an evolved system, autonomously adapting the solution and the model to be applied. The primary objective is to constantly aim for highest economic performer in the short and long term. This is a revolution for calculation and forecasting models.
In the world of Retail, these assets are fundamental and drive operational updates, resulting in meeting consumer expectations.
This approach has enabled Amazon to provide individual recommendations for books or Netflix movies. It’s relevant, scalable and constantly adapting to the behaviour of web users. Although unable to make predictions what we are going to buy on the next shopping trip, they can assume it with a very weak error margin.
In the field of inventory management and replenishment, Machine Learning brings important advantages. With the example of sales forecasts, retailers traditionally consult historical sales to make decisions on replenishment volumes while keeping external constraints that generate peaks and troughs (such as holidays, sports events and heat waves) in mind.
Unfortunately, even with the most accurate calculations, these forecasts are likely to be distorted. For example, if the initial stock is insufficient, consumer demand will not be reflected in sales due to stock-outs. These same breaks will also increase sales of similar products still in stock, by carry-forward. Symmetrically, a product sold with a large discount sales and reduces the quantities sold of similar products. For all these subjects, Machine Learning will reveal how many sales would have been made with an infinite theoretical stock and model the impact of promotions. Unlike conventional algorithms, the process is automatically executed.
With the treatment of anomalies, an item that does not sell in a specific store of a retailer, may sell very well in all other stores. This happens frequently and can happen for a variety of reasons. For a DIY store, it could be a bucket of a specific colour of paint placed in the wrong aisle.
Customers who search for the specific colour of paint will not find it where it’s supposed to be, thus a reduction in the volume of sales of this reference in this store location. As a result, there will be a long-standing stock error (the article will not be sold or the store will not be replenished). Machine Learning identifies this type of anomaly, called suspect stock, where conventional algorithms would not recognize this difference and would treat these anomalies as “normal” inventory.
Because Retail’s economic performance is closely linked to the ability to collect and analyse large and automated data volumes from various sources, Machine Learning is the next era. Profound operational and retail changes must be put in place alongside advanced industry and customer knowledge to improve the critical functions of the company.