Big Data and AI driven Promotion Management
Bi4 Group contributes to the world’s leading AI powered Revenue Management Solution for the Consumer Packaged Goods industry.
Bi4 Group is the strategic partner for Visualfabriq to develop one of the most innovative AI solution in the Fast Moving Consumer Goods market that automates forecasting of baselines, promotions and total revenue with high accuracy, using historical data and cloud-based machine learning tools.
Because vendors of Fast Moving Consumer Goods (FMCG) face fierce competition everywhere, optimal revenue management is of tremendous importance. Visualfabriq is a fast-growing SaaS company that helps CPG companies improve their revenue and margins, using big data technology, Artificial Intelligence (AI) and data workflows to integrate data, turn it into insights and increase profitability.
The company’s flagship product is called Trade Promotion Master (TPM), a system used for promotion management. It aggregates and manipulates product information and produces different sorts of reports, dashboards and graphics, so that companies that sell consumer packaged goods (CPG) can obtain the most relevant information on promotion performance. The solution was co-developed by Bi4 Group.
Whereas traditional revenue management software has trouble aggregating data from different sources, TPM proves to be a valuable alternative by using a more agile approach.
TPM is a high-level data portal that shows how promotions are performing at any moment. For instance by using TPM, you can offset actual profit compared to predicted profit.
Through complex calculations, TPM takes into account thousands of variables that define the final price of each promotion. These are used for selling CPG’s to a supermarket with the maximum possible impact for the intended promotion goal (whether it’s volume, revenue, profit, penetration or repeats). A clever AI algorithm suggests most part of the promotion entry, from automation of the promotion entry to optimization of the promotion performance.
The solution is used by several tier 1 companies in the Americas, Europe and Africa, with smart regional intelligence for the specific market demands.
One powerful component in achieving this is the application of Bayesian time series statistics for predicting the baseline sales volumes of a product. This machine learning tool takes into account certain factors, such as special dates like Black Friday and Christmas holidays, and accurately interprets how sales have evolved during a five-year period. This analysis serves then as the basis for future forecasts. Models are created “on the fly” by the system, trained and presented to the user. Because the model uses machine learning techniques to become self-predictive, there’s no need to contract a data scientist for development of a baseline forecasting model and training it procedurally.
Because the tool has been used thoroughly by many large companies, it has been able to refine itself substantially. For AI, the more data, the better it works. Documented comparisons between realized and predicted sales show that the tool is very precise and able to select the most important factors that influence promotion pricing mechanisms. There are tens of thousands of factors that are relevant apart from price, so it makes sense to use a tool to construct the most appropriate model with a subset of truly relevant predictive factors.
It uses a mix of leading open source Python machine learning libraries, augmented by own developed optimizers and automation approaches. This gives the users access to the world’s leading classification, regression and clustering algorithms. In this case, gradient boosting analysis was used for prediction and forecasting promotions, as well as investigating the relationships between sales variables and Bayesian time series for the baseline forecasting.
The power of AI
Without data, AI is useless! In order to be able to handle large data volumes, Visualfabriq uses a AWS multi-server infrastructure using multiple scalable machines. All data used is stored in the AWS cloud, inside a cluster machine and a MongoDB database. A powerful reporting module uses tables with columnar storage that enables optimal positioning and compression of information on disk, so that reporting is created instantly.
Developing the cloud-based promotions management solution for Visualfabriq came with a number of challenges. Because large data volumes were aggregated, displaying the information quickly while maintaining the stability wasn’t easy. The same goes for maintaining the stability of the platform while more clients are added. Finally, applying the right algorithms and transforming them into reliable calculations proved to be another challenge for the developer team, as well as building trust in the end product and resulting predictions. This is a consequence of adopting new, advanced technology.
However, there’s no doubt that AI technology will become an essential component for revenue management everywhere. As current tools for revenue management are insufficient for handling large amounts of data, big data and machine learning tools can make a difference in making fast and accurate predictions. Together, they enable automation of workflows that will replace human intervention. More and more companies are discovering what AI solutions can offer for revenue management. Those that use it will have a headstart on the companies that don’t.