x
operating multinationally
across major Polish cities
and expert customer support
"We are very satisfied with the outcomes. The efficiency of the solution surprised us. We were 20x faster than with the manual work. And the cooperation with Future Mind was flawless. Always on time, always coming up with new ideas, a true partnership."
As Super-Pharm was implementing a Product Information Management, the company faced a need to categorize and attribute 30,000 products in total.
We could immediately see that with the current breakthroughs in AI, large language models could be used to expedite the process and get rid of manual product attribution.
Given our existing relationship with the client, we brought this up to Super-Pharm and were met with positive reception towards the idea of automating a resource-consuming process. So, we quickly got down to work.
In the process of implementing a Product Information Management Super-Pharm created new product categories and product attributes to bolster its information architecture and make search easier for users.
The product base contained 30,000 products in total, though, and in these new circumstances, they would all need to have attributes assigned in accordance with the new attributes and category tree. Super-Pharm lacked a tool that could make this overwhelming task easier to ensure the implementation could happen according to the ambitious schedule.
We knew the company already had plenty of data on their products, like current product descriptions and categorizations based on information from Super-Pharm’s e-commerce.
This led us to believe that large language models could be the ideal solution for Super-Pharm's challenge. We fed essential details such as the new category tree, attribute sets, product descriptions, and other parameters to a GPT model, which then estimated attribute values using custom prompts.
This approach allowed us to manage large volumes of repetitive tasks much more efficiently and quickly compared to relying on an external agency.
We started by building a proof of concept, which quickly demonstrated strong matching efficiency with potential for further improvement, highlighting significant time-saving opportunities. The initial accuracy in attribution was 40-50%, a promising foundation for further refinement and development.
After enriching the model with data, we leveraged Django, a Python framework well-suited for this task, to develop a basic content management system where users could define allowed attributes and review AI-generated assignments.
Attributes, along with other crucial data such as product details and descriptions, can be added manually or imported from Excel files. This functionality was widely utilized during the PIM implementation, significantly accelerating the processes of data import and export.
By deploying a large language model to analyze product descriptions and automate attribute assignment, we significantly reduced the manual workload and accelerated the implementation of Super-Pharm Poland’s Product Information Management system.
This solution not only streamlined the process but also freed human employees from the tedious and error-prone task of manual data classification, minimizing the risk of mistakes and improving overall efficiency.
demonstrating the synergy between human insight and AI's analytical capabilities.
To ensure consistency in information architecture, the algorithm was constrained by pre-established categories. However, it was also allowed to share additional suggestions with the team, leading to valuable recommendations.
After analyzing thousands of products, the model identified missing tags that could enhance user navigation, many of which were ultimately incorporated into Super-Pharm's official categorization.
As an end-to-end solutions team, we partner with Clients to help them understand their customers
All services
>
