By Raphaël Bertholet, VP R&D, Retail Software Division, Symphony RetailAI.
Historically, demand forecasting using traditional, manual methods has proven laborious and complex.
First, the data needs to be aggregated and evaluated. For well-established products, this is extremely time-consuming due to the sheer volume of data which must be captured and analysed. And for new-to-market products there is a complete lack of historical data of any kind, which means forecasting demand for these items is based purely on conjecture rather than facts.
Data quality has also, in the past been an issue for retailers where product data from multiple sources, is gathered either directly from suppliers or extracted at the point of sale via various consumer touch points, digitally or in store. Collecting data in this way results in data arriving in different, non-compatible formats – including Excel spreadsheets – so retailers must allocate significant time to sort and clean it before it can be analysed. This involves manual intervention which creates opportunity for human error.
AI isn’t just able to handle volumes of data that are too large for humans – it can also see what is hidden in the figures, spotting anything unusual and making correlations that allow us to understand the subtleties of consumer trends.
In retail, the amount of data generated each day is enough to make your head spin. Among the major brands this can represent more than 5 billion data lines or around 2 terabytes each day. With every week that passes, over 10 more terabytes of data are produced. In these circumstances, the challenge is how to carry out analysis before the data becomes obsolete.
One of the biggest advantages of artificial intelligence is precisely its ability to analyse vast amounts of data continuously. In the world of retail every sale brings new information, so it is no longer possible to limit yourself to looking at and analysing data once a week, nor even once a day. Nowadays, retailers must monitor curves and trends almost in real time to stay ahead.
The unusual: demand forecasting’s glass ceiling
The tools of demand forecasting – statistics and Business Intelligence – allow us to analyse what we want, and especially what we know. A buyer or a manager of a store usually focuses on the products they know well, and supplies products which are known to sell. Forecasts and decisions will be helped – and intuition reinforced – by standard tools. However, the latter quickly reach their limits when an unusual situation arises.
It’s during these unusual circumstances when the most complex forecasting errors occur: for example, when a product isn’t selling “as it should”, or conversely when a product is selling a lot more than previously (and suddenly is out of stock.) It is sometimes difficult to see the correlations which explain the disparity between the forecast and the reality.
There can be multiple correlations. For example, sales of certain products can fall due to a special offer of a similar product from the same category (this is referred to as cannibalisation). Or even the sales of a product which dramatically decrease before being on special offer because consumers hold off on their purchase. Another typical example is when sales of certain products take off because of a special offer on a complimentary product (such as a special offer on paper plates that makes sales of plastic cutlery increase).
The launch of new products is another type of unusual circumstance. In the absence of existing data, the forecasting of supply and demand is often based on intuition, with the whole margin of error that this brings.
In the great majority of these cases, demand forecasting collides with a glass ceiling and remains restricted by a margin of error which is very difficult to reduce. AI changes this.
Understanding the hidden subtleties of demand
Because AI can understand hidden counter-intuitive correlation, it is able to identify what’s unusual. Or, at the very least, it presents correlations and gives recommendations that will help retailers to understand why.
AI means that forecasters and heads of purchasing can look where they should be looking – including at external factors and contextual data such as the weather or at socio-demographic information.
In this way, some consumer behaviour, which would otherwise have been judged inexplicable or irrational, can be explained. And this is what makes it possible to improve forecasting accuracy by several percentage points. These few points are the most difficult to gain because, with traditional tools, it would require an impossible level of time and resources.
It is these few points which will from now on give you the competitive edge in the world of retail.