solution accuracy
duration of price offer calculation
people - Human resources reduction
Imagine you produce environmentally sustainable packaging. You supply a wide range of products to a large number of long-term satisfied clients. The demand is just pouring in. But the surge in demand is not necessarily all positive. It takes too much time to process so many offers at once. Customers are getting more impatient by the day, and your staff are getting under pressure. Could artificial intelligence help solve this? Read how Gauss Algorithmic helped THIMM Packaging significantly reduce the complexity and length of the quote preparation process.
In the world of manufacturing, every minute counts. This is of course also true for THIMM, which is one of the European leaders in packaging solutions with 12 locations in 5 countries. The company is continuously investing in automation and digitalisation of production and uses the latest technologies such as digital printing and laser cutting to produce its packaging. One of the challenges was the preparation of quotations. It was necessary to coordinate the cooperation of three departments - sales, development and quality control. As a result, a single quote could take several days to prepare and clients did not always wait that long. It was therefore time for a radical change that would make the entire sales process significantly more efficient.
The idea to use artificial intelligence in the company emerged during a conversation between the company's IT manager - Tomáš Kumhera - and the CEO - Martin Hejl. They talked about how years ago they used to simply estimate the cost of producing boxes by doing basic calculations over the parameters of the boxes in demand. That's when Tomáš Kumhera realised that they could replicate this approach using modern methods and artificial intelligence to streamline the lengthy process of estimating production costs.
Tomáš Kumhera realized the potential of artificial intelligence during a webinar for the Czech-German Chamber of Commerce, where Gauss Algorithmic presented successful examples of AI in industry. The goal of the collaboration was clear after a quick chat: to transform the bidding process with an efficient, fast and accurate system.
New types of values often emerge during the collection of production data. This is due, for example, to the adoption of new machinery and technologies, which then get entered into the database. On the other hand, some older values drop out over time. AI models are generally sensitive to the amount of training data. If there is not enough of it, it may not be possible to build a good prediction model. In that case, it is necessary to decide not to use it for some inputs and wait until sufficient new data suitable for the problem is collected in the future.
The Gauss Algorithmic team is able to perform initial data analysis and recommend to the client what is appropriate to store in the database if they are not already doing so. They can also recommend what additional information needs to be stored compared to the current state and then evaluate how long it will take to have enough inputs to train the AI model.
In the case of THIMM, there was already a rich archive of long-term and systematically collected data. The challenge, however, was to decipher which data were key to building an accurate predictive model. So the Gauss Algorithmic data team set to work. They were not only looking for hidden relationships, but also testing different combinations of datasets.
„ We were working with a huge amount of data. Our task was to determine what data we needed and to test whether it was sufficient to achieve our common goal."
- Jiří Hroza, Data scientist Gauss Algorithmic
The Gauss Algorithmic team took a dual approach: neural networks and boosted trees (XGBoost). After thorough testing, it was shown that decision trees offer the same accuracy, but at a significantly higher speed.
„ We opted for decision trees. Both models worked, but the trees were faster. This is crucial for real-time predictions - a salesperson can process a quote even during a meeting with a client, for example."
- Jiří Polcar, CTO Gauss Algorithmic
An interesting finding was that the type of box, which was generally considered to be a major factor in predicting cost, turned out to be less significant. In fact, it influenced the cost of production significantly less than the quality of the material used. This finding underlines the importance of a data-driven approach as opposed to a purely intuitive one.
The Gauss Algorithmic team developed the complete app including the web interface and backend. Data from ready-made offers was also integrated into the solution, creating a holistic solution that goes beyond prediction alone. Beyond the highly accurate cost prediction, the user has the ability to compare the estimate with similar, already produced and therefore accurately calculated bids. The system also takes into account dynamic changes in raw material costs.
„ Actual costs for paper, ink and other raw materials are input into the model. We send this data to the Gauss Algorithmic team in real time, ensuring that the prediction always works with actual data."
- Tomáš Kumhera, vedoucí oddělení IT společnosti THIMM
The Gauss Algorithmic team didn't just focus on the technical side of the solution. Considerable attention was also paid to the clear user interface. The goal was to create a fast, intuitive and easy-to-use application for its end users.
The result was an interface that allows users to take full advantage of the AI model's potential - without the need for complex user training. Everything is based on the requirements of the end users and the real situations in which they will use the product. This has resulted in a practical interface that can be extended with additional functionalities in the future.
„In today's dynamic times, our customers expect to receive their offers without unnecessary delays. Implementing an AI model for creating quotes has thus brought us a major increase in efficiency and speed. The AI model has also contributed to allowing our developers and technologists to focus on innovation and creativity instead of spending time on routine tasks. This shift opens up new opportunities and contributes to the continued growth of our company.”
- Martin Hejl, CEO of THIMM
Results that speak for themselves
After six months of intensive work, the project has achieved great results:
„ We trained the model on 90% of the data and tested on the remaining 10%, which the model never came into contact with. It turned out that we achieved 96% accuracy, which is an excellent result."
- Jiří Hroza, Data scientist Gauss Algorithmic
This case study demonstrates the power of artificial intelligence in transforming traditional industrial processes. It effectively addresses the original challenges and opens the door to new possibilities of using AI throughout the plant thanks to careful data handling, machine learning expertise and close collaboration with THIMM's internal teams. The project is clear proof that AI, when implemented correctly, can bring about revolutionary changes, even in more traditional industries.
„ This project shows that successful AI implementation is not just about creating a model. It's about understanding the problem, analyzing the data, and creating a comprehensive solution that will truly improve the way companies work. For THIMM, this means not only saving time and resources, but also being able to respond to customer demands faster than ever before."
- Johnson Darkwah, CEO of Gauss Algorithmic
THIMM is a major provider of packaging and distribution solutions in the EU
