Artificial intelligence and machine learning: the secret of Boticário’s demand planning solution
Challenge:
The beauty and cosmetics industry is marked by a number of complexities that challenge the predictive capacity of both human beings and traditional computational models. Its main characteristics include fashion, complex sales cycles, the lipstick effect, fast pace of innovation, and chain integration. And, to deal with such complexities, the most advanced machine learning and artificial intelligence tools have to be correctly applied.
Fashion, subjectivity and sales cycles that are not repeated
Marked by trend colors and different subjective evaluation criteria, such as fragrance and texture, beauty and cosmetic products restrict the traditional evaluation of the functional characteristics. In addition, non-repeating time patterns, different promotion and sales cycles every year, both in number and duration (non-stationary time series). A typical example is the Black Week, which has changed consumer behavior, introducing a new major sales period in the year and redistributing promotional sales cycles. These characteristics require flexibility and high learning capacity from the forecasting methods, exploring apparently non-existent patterns and quickly incorporating new sales behaviors in the forecasting.
Traditional purchasing relationships with macroeconomic indicators are variable. While in general an increase in unemployment and inflation tend to reduce consumer spending, an opposite effect – the lipstick effect – can boost purchases of beauty products. In addition, more expensive categories, such as perfumes, can have a migration of spending with imported goods. Then, the performance of the categories in the cosmetics market is a challenge for standard economic interpretation and requires the inclusion of a broad set of macroeconomic information.
Besides the lipstick effect and subjectivity, staying competitive in this market requires meeting trends with maximum speed. This implies products with short life cycles of usually less than two years. This portfolio dynamics limits the use of similar products in forecasting and traditional techniques that require large amounts of historical data, demanding from predictive models a quick identification and incorporation of emerging purchasing patterns.
With all such complexity, maintaining a high availability of products can result in high and unnecessary inventory throughout the supply chain. When combined, forecasting uncertainties and short life cycles often result in inventory loss by disposal or a drastic reduction of product margin due to product excess flow. Thus, a proper forecast of sales to the market (sell-out) is not enough. Reducing costs in the supply chain requires that both its lead-time and service structure and its efficient inventory operation parameters are incorporated into an accurate sell-in forecast. In a franchise system, this challenge is even greater, as it involves considering the behavior of different actors (franchisees) in different channels (direct sales, e-commerce and physical stores) in the forecast, without affecting the process efficiency with so many planners.
Solution:
Flexibility and efficiency of artificial intelligence
It is clear that traditional forecasting techniques are not able to respond to planning needs in the dynamic cosmetics market. For this reason, Boticário combined different techniques of artificial intelligence and machine learning, such as artificial neural networks, decision trees, and regression with support vector machines.
CTI Global was hired by Boticário to prototype and implement IBM Analytics solutions, including IBM Planning Analytics TM1®, a fast in-memory data analysis platform for financial and operational planning. To ensure forecasting with more precision, IBM SPSS® Modeler and the R language were used.
Results
- 20% increase in demand forecast precision when compared to the traditional method
- Reduction in inventory levels and reduced number of failure of desired products, leading to increased sales
- Constant learning at each forecasting cycle, with the ability to react quickly to the portfolio dynamics and market changes
A proper combination of these techniques and software allowed:
- Flexible planning, incorporating sales cycles and promotions that vary over time;
- Use of macroeconomic indicators that identify counterintuitive sales patterns, such as the lipstick effect;
- Resource efficiency, with most work performed with high-quality computer-based forecasting, with planners focused on the analysis of results;
- Chain integration, in which machine learning models do not only build sell-out forecast, but also include different service and inventory variables, generating a proper sell-in plan;
- Evolution flexibility, allowing quick testing and update of techniques, technologies and predictive models. For example, incorporation of cognitive intelligence (such as IBM Watson) to extend the predictive capacity with unstructured data (for example, consumer reviews, information from social media, etc.) is fast and easy.
History of results
Although the adoption of the new solution is still in its initial stages, Boticário has already seen impressive results, enabling the company to expand its market share.
“With IBM Analytics solutions, we’ve increased the precision of our forecasting by 20% when compared to traditional time series analysis,” explains Donald Neumann, Demand Planner at the Boticário Group. “By allowing us to align demand from end customers and franchises with higher precision, the software will help us minimize inventory issues without having to increase inventory levels across the supply chain. The result will be increased sales and reduced costs, that is, a situation that will benefit everyone and help us stay ahead of the competition. We can also send these insights automatically to our franchisees, with benefits to the entire supply chain,” he says.
The Boticário Group can achieve better results with marketing plans based on a better understanding of consumer demands.
“The project with IBM and CTI Global changed the relationship between our analysts and the marketing teams – now, they speak the same language. At the meetings, we’ve already noticed that our analysts started talking about the demand and behavior of customers, rather than the demand and behavior of franchisees – therefore, we are closer to the audience we want to talk to,” says Donald Neumann.
“With data-driven decision making enabled by IBM Analytics, we can design marketing plans that attract more people to stores to buy our products. Consequently, our marketing expenses will be used to obtain the maximum return,” he adds.
With a better understanding of consumer demand, which is constantly evolving, the Boticário Group has found the answer for long-term success: fulfilling the desires the consumers don’t know they have.
Donald Neumann adds: “The purchase of cosmetics and fragrances is a highly emotional activity. IBM Analytics solutions help us understand the various factors that influence consumer decision when choosing one product over another. By processing these insights, it is possible to ensure that we (and not our competitors) will be responsible for taking beauty into the lives of our customers.”
Solution developed:
- Cognitive Demand Forecasting Solution
Platforms used:
- IBM DB2®
- IBM SPSS® Modeler
Count on CTI Global to develop a customized solution for your company: