Intelli-tail: Artificial Intelligence for Retail



The success of eCommerce world leaders, such as Amazon, is due not only to low prices, high quality and convenience of service, but also depends on a special ingredient: Artificial Intelligence (AI). With these techniques, for example, the history of visits to the website and previous purchases of each customer are analyzed to provide useful recommendations and attractive incentives. In warehouse management, AI is used to get the best service at the lowest cost (not running low on stock, while minimizing investment) through sales forecasts made by analyzing not only past history, but also many external factors.

No traditional algorithm would be able to discriminate with the same effectiveness any significant correlations down to the details of the individual customer and product, or evolve all by itself. In fact, Machine Learning (ML) is the AI subset that deals with creating "algorithms that learn from data" in order to predict, recognize, understand and classify. Once the mathematical model is automatically created, the answer to any question becomes very fast.

An example of the level reached today by these techniques is the new physical store format that Amazon is experimenting with. For more information

To the objection that AI is too sophisticated for small chains and independent stores, we anticipate that a fast democratization of these techniques is in progress, along with the smart transformation of many daily management tools, at an affordable cost.


Why now?

Artificial Intelligence is a vast scientific discipline that brings together mathematics, statistics, computer science and neurology, with ethical, philosophical, and even political implications for the impact on employment and regulations. AI began about 50 years ago, but in these last few years it has undergone an astounding acceleration. ML (Machine Learning) is the foundation, for example, for voice recognition and understanding used by any smartphones through tools like Siri, Cortana and others, or image recognition, up to the automation of many knowledge-intensive human activities such as, for example, medicine and accounting.

The reason for this "explosion" is the convergence of the natural progress of science with Cloud Computing, that provides the resources for handling the so-called Big Data (data which due to its volume, speed and variety, cannot be managed effectively with traditional techniques) and the vast computational resources needed for algorithm learning, at very competitive costs and without any initial investment. Lastly, because real Cloud Computing is much more than the self-service of machines that must be managed with traditional methods, only in recent years have several cognitive services become available that are ready for use by ordinary programmers and sometimes even by advanced end-users) and not necessarily by "data scientists".

For more information


Why does ML change the rules of the game?

For the first time in history, ML automates tasks that hitherto could not be dealt with by a computer, reaching a level of precision and speed that can exceed human capacities. Self-driving cars that will reduce the accident rate, or automatic medical diagnosis, are just two examples.

Returning to the topic of Intelligent Retail, let us consider the example of recommendation lists (whoever buys product x, usually also buys product y) or grouping customers into homogeneous clusters in order to put into practice the most effective promotions for each one. For a software program this is no easy task, but much simpler than recognizing the sex, age and even emotional state of a customer, simply by analyzing his image, as is already possible today with ML. Although recommendation algorithms can be programmed directly based on knowledge of the product category and statistics, today it is better to rely on ML in order to automatically extract rules from large volumes of data. With the evolving buying behaviors and/or the availability of new data, such as the weather or the popularity of products on social media, all that is needed is a new "training" operation in order to obtain an updated and enhanced algorithm. Lastly, choosing within a few seconds the most effective coupon, taking into account not only historical data, but also products that the customer is buying at that precise moment, is far beyond any human capability.

In terms of customer service, ML already allows questions to be answered and handles useful conversations with customers through natural language, spoken or written in a messaging App.

With regard to product management and inventory, the main problem is the attention and expertise of staff in analyzing data and making decisions. Data visualization techniques associated with Business Intelligence (BI) and Data Warehouse (DWH) greatly facilitate the task of extracting insights, but they become insufficient at individual customer and product level. ML can take into account many different factors such as, for example, the impact on balance sheets, the satisfaction of a major customer or outstanding deviations from the mean, and report to the decision-maker the most important facts, acting as an experienced consultant, thereby helping not only when time is tight, but also in terms of expertise.

This is an example where AI does not replace people, but helps them to work better, in an environment such as Retail where time and skills are in short supply caused by ever-shrinking margins. Another paradigm shift is due to the fact that models created by ML look at what will happen in the future, while traditional analysis and data visualization tools look at the past.


Why is Intelligent Retail only realistic in the Cloud?

ML requires large volumes of data and huge computing power to create mathematical models that can respond quickly to requests from stores. Such resources would be very expensive to purchase and manage in-house and furthermore greatly underutilized. Even more decisive is the fact that the resources and skills needed to deliver ready to use ML cognitive services can only be found at reasonable costs in the Cloud.

This suggests that the store management software and ML services should be in the same Cloud because data is subject to a sort of inertia, even though it consists of electrons traveling at the speed of light and not atoms subject to the force of gravity. Indeed, to move large data volumes takes time, even with good bandwidths available in the stores and at the chain HQ. In addition, the delay between question and answer (called latency) is not negligible even for small exchanges of information.

As demonstrated by the initial projects, a Store Management SaaS such as aKite, specifically designed to leverage a modern Cloud platform, is in the best position to take advantage of this new revolution. The main reason is the architecture, scalable and open by design to cooperation with other Cloud services, plus the direct and immediate availability of historical data to ML services.

For more information