AI challenges in retail and manufacturing
Technological progress and the use of Big Data in business make AI-based solutions increasingly important, but implementing AI in retail and manufacturing is still perceived as walking the cutting edge of innovation. We discuss the main challenges with Ireneusz Prus, Director of Data Monetization at Maspex Group.
What convinced Maspex to implement AI solutions?
Maspex is one of the largest food producers in Central and Eastern Europe. We have been aware of the business potential hidden in our data and we also have known that only with advanced AI algorithms would we be able to fully harness its potential. Supporting our experts with the knowledge provided by artificial intelligence is another step forward in building our competitive advantage.
How did you start the process of implementing AI?
AI can enhance nearly every aspect of our business, so it was pretty challenging to choose one use case that could quickly deliver results and present real business value. As we didn’t have an internal Data Science team, finding an experienced technology partner was the first challenge. We decided to work with deepsense.ai, a leader in AI. Together, we conducted a data audit, which helped us to prepare for the ML implementation.
This led us to develop our first use case around our marketing activities and build a tool for simulating marketing campaigns and measuring their effectiveness. Maspex performs thousands of promotional activities spread across the whole supply chain. So far, promotional campaigns have been carried out solely on the basis of our experts’ knowledge. We suspected that there was a lot of room for optimization, but there was no tool that would identify these campaigns.
What did the cooperation between Maspex and deepsense.ai look like?
Implementing enterprise AI is very much a cross-departamental process. So, in the first stage we discussed – together with different business stakeholders – which external and internal data should be used to achieve our goals. The biggest challenge was to combine available data sources and to establish a uniform set of variables that would best allow us to predict high impact promotional activities.
Various data sources were integrated to establish the dataset: transaction data, data on promotions and their coexistence, data on competition activities and market trends, and other external factors such as seasonality of sales, weather conditions and variable restrictions related to the pandemic.
deepsense.ai then used that dataset to start training the predictive models. In parallel, models were built based on five different algorithms and ultimately the algorithm that gave the best results was selected. In addition to the predictions related to promotional campaigns, adjunctive algorithms were created to analyze the periods immediately before and after promotions and to evaluate the interactions between different types of promotions.
With deepsense.ai’s support we have developed a system that allows us to define promotion parameters using lists and sliders. We can define promotion, product and target audiences. The parameters are analyzed by the system taking into account the list of evaluated promotions, 12-week-ahead uplift and the impact on market share. As an output, users receive precise predictions of campaign effectiveness and are able to choose the most successful scenario.
What are the next steps?
We plan to further develop the AI implementation process and address the needs of various internal clients. As a market leader, we have multiple sources of data covering the whole retail value chain. It therefore made sense for us to develop our own Data Science team of experts specialized in advanced analytics, which we are currently doing with guidance from deepsense.ai. They will not only have a real impact on our operations but will also have an opportunity to create an innovative AI mindset around our company. The road ahead is full of exciting challenges.
