A company health in the mature and competitive Retail Industry rest on both vision and execution. The vision (or strategy in military terminology) consists in knowing one’s own strengths and weaknesses, choosing a customer segment and the needs to fulfil better and before the competition. Execution (or tactics) indicates instead the ability to define assortment and quality of products & services, stock levels and prices, as well as the store associate characteristics consistently with the strategy and effective in terms of profitability.
Discussing about precision trade and its evolution by means of Artificial Intelligence (AI) techniques, we will only deal with the execution level where daily battles in satisfying customers, beating the competition and especially in producing profits to stay on the market are won and lost. Going more in detail, by precision retail we mean the one in which the manager leverage KPI (Key Performance Indicator) to make daily decisions, while for intelligent retail we prefigure the one in which Artificial Intelligence algorithms propose specific actions. Anyone who knows the sector, especially in the base of the pyramid composed of small and medium-sized companies, knows how impractical the first option is. Making optimal decisions based on the data, assuming it is available, requires time and culture, even when a modern Business Intelligence is available.
Automatic replenishment proposals is a first area that can boost KPIs through a new generation of sales forecasting algorithms, aimed at providing a good level of service with the lowest investment in goods, also reducing the risks of obsolescence.
The forecast of the availability days of each product before a foreseeable break in stock and each supplier delivery time, allows timely and precise replenishment, automatically optimizing the profitability of the capital invested in goods, a KPI called GMROII from Gross Margin Return on Inventory Investments.
An intelligent algorithm learns from the past and can take into account future events such as weather conditions, recurrences, own or competitor’s promotions, but it could even take into account, as soon as it becomes known, unexpected events in the supply chain. It can also recommend transfers between stores or price changes.
To manage prices optimally it is necessary to know the demand elasticity for each individual product, i.e. the variation factor of the quantity sold against a change of the selling price. An intelligent algorithm is able to estimate the elasticity taking advantage of the price changes proposed by a different inventory management algorithm, decreasing it to dispose of the stocks that are proving excessive compared to similar products or, on the contrary, increasing the price for delaying an expected and inevitable out of stocks.
The dynamic price management, widely used by eCommerce leaders, is not popular in traditional retail, also because average information systems are not suited in this area. Another important fact, useful for optimizing not only the stock level but also the assortment, is the substitutability between products or how much sales shift to product B, C … and how much customers give up buying, when product A is missing. In this case, an intelligent algorithm can originate advantages from a negative fact such as the inevitable stock-out, estimating how sales have changed compared to forecasts in normal conditions.
Recommendation lists, widely used by eCommerce leaders such as Amazon, indicate the probability for customers who have purchased product A, that they will also buy products B, C … It is clearly a very powerful tool for assisted sales and for tailored promotions, presumably perceived as a useful service rather than a disturbance.
In order to create these lists one of the problems in traditional stores, but also in the chains, is that the Artificial Intelligence needs many data for the Machine Learning phase and these must come from customers who have given a profiling consent (a subset of the total). Even a medium chain has a fraction of the data compared to those available to eCommerce players. Fortunately, a further dose of intelligence can help to bridge this shortage, taking aggregate and therefore anonymous sales as an indicator (proxy) of the liking of each product in the store or in the chain. If available, market data can also be used. Although these are indirect data, they have the advantage of a limited variance, synonymous of good precision due to the large numbers, useful for integrating the relatively small numbers and high variance of data derived from profiling.
We saw a path in which AI proposes immediately useful actions (for example a price change) but each action also produces new data that, processed by other intelligent algorithms (for example to calculate the elasticity of the demand), indicate still better directions, in a path that is difficult to predict in the medium term, but which will be certainly explosive.
In the past, a quantitative leap (more of the same thing) has sometimes produced a qualitative one (new ways of doing and innovative mental models), but in this case we are faced with something like a chain reaction. Artificial Intelligence is a “transversal” technology able to improve practically any human activity, as happened for the great inventions of the past such as electricity, but in an even deeper way, because for the first time a machine is able to derive useful models to predict and decide with more precision and speed than an expert.
Currently this superiority is limited to specific sectors, such as recognizing images and data patterns or estimating the evolution of historical series, but it is so inexpensive that it can be used to better manage every single product in the assortment. The result, even if not singularly striking, will certainly be at an aggregate level, allowing even small companies to be managed optimally and to grow, if the strategic choices, still in the hands of humans, have been the right ones.