Computerworld Interview – Unleash the full potential of the algorithm – process optimization through the use of AI

Today, Artificial Intelligence (AI) is an integral part of business transformation. It is changing the way organizations and companies operate in the marketplace. Numerous Machine Learning algorithms and models predict consumer trends, optimize manufacturing processes and offer unparalleled opportunities to improve business operations on many levels. 

In an interview with our expert Tomasz Mirowski, CTO at 3Soft, we will discuss an example of the implementation of a solution based on Artificial Intelligence, which not only brought significant benefits in the form of optimization of logistics processes, but also resulted in a significant reduction in incurred losses.

The implementation of Artificial Intelligence in a company often involves the transformation of numerous processes within the organization. Let’s look at a specific case of implementing an AI solution at one of your clients, a large FMCG company – what did the first stages of cooperation look like?

In 2018, a client approached us with a seemingly simple problem – they wanted to minimize waste in his stores due to the increasing amount of expired products on the shelves. They wanted to plan deliveries in such a way that it would be possible to sell specific goods on an ongoing basis without creating overstock of specific products.

The client provided us with all the necessary data, which was the starting point in working on the design of a platform for predictive analytics, based on Machine Learning and tailored to his specific needs.

We started by creating a PoC (Proof of Concept) for a single store. At this stage of the project, there were no complex data sets, so-called Big Data, yet. There was one Data Analyst and one Data Science Specialist working on the project. It was a very quick solution to see if we would be able to do something of value to the customer to deliver the expected results at full scale – and we’re talking about numerous points of sale. The first model was based on checking products with a short shelf life. The results turned out to be really promising.

How did the further phases of work on this project develop?

We leaned towards products with a long shelf life. Again, the results came out very well. The prospect of further work on the project had great potential. Therefore, the decision was made: we put the project into practice.

We gradually increased the scale of operations. We expanded the model to more stores and countries. We switched to cloud solutions to increase scale and enable automation.

We started with a single CSV file forecasting one product in a selected store. The current solution includes sales value forecasting for thousands of products in multiple markets. It’s fair to say that we started with ten megabytes or so of data, eventually moving to petabytes in real time to provide daily information on what the next day’s sales will look like.

What were the main challenges during the development of this project?

One of the main challenges was scaling up – we had to move from the PoC development phase to developing a full-fledged solution that covered multiple stores and countries. This required standardizing data formats and overcoming technical hurdles. In order to build the final solution, we had to build a team of qualified specialists with diverse competencies and skills over several years. We also had to determine with the client how to plan the entire process so that we could use the solution as soon as possible. We created a universal MVP (Minimum Viable Product) for all countries and stores. We assumed that we would create it within one year, and then gradually improve it as we used it. So we had to build with the client a kind of standard for exchanging data between the different systems: the national system, the client’s system and the client’s central platform. That was our biggest challenge at the beginning of the work.

How have you been able to address new products and evolving data requirements?

For new products, we have developed algorithms based on historical data to forecast their sales value. We have an algorithm to build a product family based on the sales history of different products, which enables us to forecast the value of a new product. In addition, we have incorporated additional data, such as the location of products in stores, to further improve our predictive models and optimize supply chain operations.

How did you address the need for flexibility and scalability in the solution?

These were key issues. We leveraged cloud native services combined with the Cloudera platform to achieve a hybrid solution that could dynamically adapt to changing requirements. Automation played a key role in scaling the platform effectively, while keeping costs at an acceptable level.

If we were to present the current scale of the project’s operations in numbers, how would they look?

We are currently processing 10 million receipts a day in more than 6,500 stores. How many AI models are we running here? You have to multiply the number of stores by the number of products. Interestingly, while data collection takes place constantly while the stores are operating, the analytical and forecasting parts run only at night.

What were the final results of implementing this AI project, what did you achieve?

We delivered measurable business value to the client. This included improving sales forecasts, optimizing the supply chain and reducing waste. Forecasting has to be effective to be meaningful. We are constantly optimizing the supply chain, minimizing the number of overstock and shortages in stores. The entire change has had a positive impact on the environment by reducing the amount of food wasted. We have also increased staff productivity, as previously employees had to move around the store and manually check for expired products. Of course, we have also increased sales. If we look at our statistics, we can see a 5%-7% increase in sales. By minimizing out-of-stocks, we could see a 3%-7% decrease in losses and a 30%-50% reduction in the average number of out-of-stocks.

Why do you think the solution turned out to be so effective?

Flexibility, openness and focus on delivering real business value were key concepts for us during the work. By constantly adapting to changing needs and using modern technologies, we were able to transform the technological framework into concrete opportunities for our client.

Before starting work on the project, we were mindful of building trust in the relationship with the client and the importance of the data discovery stage, conducting Data Discovery Workshops. These are the starting point in any AI project and provide the opportunity to find and identify potential business value for the client.

Tomasz Mirowski – graduate of the Faculty of Information Technology at the Opole University of Technology. Completed postgraduate studies in Big Data and Advanced Data Analytics Systems at the WSB Academy. Holds numerous licences and certifications from, among others, Microsoft, Hortonworks and IBM, proving cloud competence, as well as knowledge of the Hadoop platform. Worked for one of the largest Polish banks as an IT expert at the beginning of his career. In 2017, joined 3Soft as an IT Architect. Since September 2021, responsible for the Data Management Department as Chief Technology Officer. On a private level, a husband, father and enthusiast of music, whatever the genre. Also a fan of new technologies and electronic gadgets.