How Artificial Intelligence helps FMCG manufacturers and offline Retailers
Articles
The COVID-19 epidemic has temporarily changed our consumer habits. People began to save more; they reduced the number of visits to stores, and switched partially to online shopping. At the same time, the Internet channel share in FMCG sales has increased from a maximum of 2 to 3% due to the quarantine measures. This modest result is driven by the fact that the e-commerce growth factors are poorly expressed in mass-market goods. In the next 5 years, we should not expect an increase by order of magnitude in the share of online sales in retail.
With this market structure, competition for shelf space is constantly increasing among FMCG manufacturers. Since in this product category the potential of product properties has been exhausted for long ago, now the main factor is the price.
Before the introduction of restrictive measures this spring, the retail chain stores were massively attracting customers with all sorts of promotions. According to Nielsen’s estimation, the share of promos in sales of consumer goods in Russia has already exceeded 50%-line last year, both in terms of value and volume.
Planning to stay competitive, the management is looking closely at really new solutions for working offline. Yes, the market has successful cases of increasing revenue by increasing market share. But it is also possible to avoid the low margin trap by optimizing processes and increasing their efficiency. Modern scalable technologies and innovative approaches to sales management can improve business profitability even in adverse conditions.
“Locally and Precisely”
There is an obvious leader in the global market for photo recognition services — the Israeli vendor Trax. It started its operating activity more than a decade ago. In 2019, the company has finally confirmed its position through the takeover of its closest competitor, the European company Planorama. In addition, after the technology commercialization in 2014-2015, young companies with ambitions, including international ambitions, have started to appear on local markets.
While forming their business processes, the large and medium-sized manufacturers are using the Sales Force Automation (SFA) system everywhere. Its functionality allows them to optimize the collection and processing of orders, ensure the most effective management of field personnel and distribution network, set up a basic level of analytics for retail outlets, for secondary sales (Sell-Out), and for the effectiveness of the sales organization structure.
KPI structure for the Sell-Out sales management in FMCG
Some retailers are testing and implementing new technologies and artificial intelligence so far at the second level of the sales management system.
To reduce the burden on the payroll, all the market participants, including FMCG giants, are forced to abandon their own sales staff, and apply outstaffing in their practice. The advantages include cost saving and increased flexibility. However, the risks associated with the human factor are increasing: low loyalty, involvement, qualification, and motivation of external employees.
This has a particularly strong impact on collecting, processing, and auditing the field metrics. This is about the classic shelf KPIs of a merchandiser and sales representative: On Shelf Availability (OSA), Out of Stock (OOS), Share of Shelf, and many other indicators of product representation in retail outlets.
International corporations develop their own algorithms for achieving and controlling these indicators (“RED” for Coca Cola, “Perfect Store” for MARS, “ITOS” for InBev, etc.). But even in this case, collecting data from the shelf space, or auditing, is a “sticking point”. At the same time, second tier manufacturers and local players are trying to figure out the basic KPIs in order to manage their representation in retail for somehow. In both cases, innovative technologies have been serving the best catalyst for the process in recent years. In our case, it is a service of high precision photo recognition based on AI and aimed at control the range on shelves in offline stores, and improving the efficiency of merchandising.
Photo recognition by using the computer vision is not just a security issue. Today it is a commercial technology
Business development Director
Before and After the Rollout
The project is implemented in several stages. At the pilot stage, a basic level of accuracy is achieved for recognizing a new product base, in one or more categories. Field employees are trained how to work with the technology on the test site, they make visits according to the new rules and accumulate photo content. The neural network starts to better recognize products on shelves. The first results provide the customer with reliable information about the actual levels of implementation of the basic KPIs: OSA and OOS. It becomes possible to effectively manage the indicators in the pilot territory, and the confidence to the process is being formed.
In case of success, the project goes to the rollout stage: the number of the ecosystem participants increases to hundreds or thousands, and the photo database increases to hundreds of thousands, or millions of SKU images. Expanding the list of KPIs, customizing them, eliminating failures in the chains of personnel work, change management, technical support — all of this turns the product into a full-fledged SaaS service. Now services are provided by subscription, and the developer company receives recurring payments.
During this period, it is worth focusing on scaling, and not to go far beyond the basic metrics related to the distribution of product items, their availability on shelves, and share in category. Otherwise, the customer risks delaying the launch “in field”, which could lead to unnecessary losses for the business.
After the recognition technology is implemented everywhere, and the sales service trusts and relies on the data obtained through the technology, the third stages — extracting additional benefits — begin.
Its main task is to automate the more complex metrics and processes as a priority for a particular FMCG manufacturer. More often, we are talking about measuring the level of compliance with planograms (product layout rules), price monitoring, and competitor analysis. But there are also unique requests. For example, reading the menu or digitizing product categories in a store.
In practice, the planogram is decomposed into a set of specific rules for the display of goods. They are delivered to performers, i.e. merchandisers or sales representatives, and to auditor, i.e. image recognition system. This is how the process highest efficiency is achieved — the user understands clearly why the service has evaluated their performance at a particular level. Among the most common rules, there are: the number of product items; the width of display (both in faces and in centimeters); compliance of the shelf number; the brand layout in a single block; and the order of goods. With this, the system does not compare the layout photo with the target planogram pixel by pixel. In this case, the level of compliance would always tend to zero, due to many factors.
Usually, the focus is shifted to getting maximum benefits from process optimization at this stage, since the task of the technology payback has already been fulfilled before. By accumulating the everyday data, the user increases the depth and quality of the product range assessment which allows them to draw conclusions about the most effective merchandising strategies in real time, they can scale them, and test new hypotheses.
In the future, this kind of SaaS service may become one of the basic services of large retail FMCG networks, and not only of them. According to BCG estimation, the revenue could be increased by 7-9% by implementing AI solutions in sales, marketing, and production planning. Some mature market players in Western countries having more transparent and formalized terms of cooperation are following this path already.