Inspira Grupa is a pioneer in the field of internet-based business operations in Serbia and is among the rare domestic companies that are successfully developing their own digital products in as many as six areas: employment, the automotive sector, real estate, insurance and sales of tyres and auto parts.
All businesses involved have their own development sectors, but also one common element in the form of a centralised data science team consisting of six employees. They are tasked with working together with businesses to develop products for each site, based on the intelligent use of data. Here CorD Magazine discusses the work of the team and the latest major project it created with PhD Stevan Ostrogonac, senior engineer for artificial intelligence, and Srđan Mijušković, senior manager for product development.
What’s the latest project to emerge from the work of the Data Science team?
Stevan: Over the course of the past two years, working with the support of the Innovation Fund, Inspira Grupa has developed a system for recommending content (the so-called Rekomender), within the scope of the Innovation Co-financing Programme.
The project itself was implemented within Inspira Grupa’s central Data Science team, with the great assistance of our businesses’ development and IT teams. The system has already sprouted in the businesses of Poslovi Infostud, HelloWorld and 4Zida. The Rekomender serves close to two million users, who receive personalised content selected from among tens of thousands of adverts.
In order to generate recommendations, the system utilises user activity data and content data, as well as textual descriptions of content. The concept solution received 1st prize in the Path of Nikola Tesla Award competition of the Union of Engineers and Technicians of Serbia, while it is currently in the patent process.
What is meant by content personalisation, specifically for users on sites within the group?
Stevan: The architecture of the system that we’ve developed encompasses various needs of individual businesses. As such, for example, the Poslovi Infostud site gains the greatest advantage from so-called collaborative filtering, which enables the user to find a job that they are qualified for despite that job being formally classified in a category that isn’t in the interest of that user. For instance, a GP may receive a recommendation for a job as a professor of medicine, despite that job being categorised under “education” and not “medicine”.
Textual description-based recommendations are particularly important for the site 4zida.rs, as the system extracts relevant information pertaining to real estate properties from texts and integrates that information into the process of generating recommendations.
How do you gauge the success of a product that functions on the basis of smart data management?
Srđan: In our approach to measuring the success of a product, we focus on key performance indicators like data management system efficiency, user experience, number of users, time spent using the product and returning customer rate. In cooperation with our businesses, we use a variety of methods to gauge success, including collecting user feedback, analysing data on product usage, conducting market research and analysing competitors.
Specifically, at Poslovi Infostud the greatest importance is attached to the number of views of adverts, as well as the number of job applications generated accordingly. A comparative test conducted on this site showed that the number of views was 30% higher, and the number of applications 35% higher, for recommendations provided by the Rekomender compared to recommendations provided by the most similar adverts.
With more than 20 years of experience, previously as Infostud and now as Inspira Grupa, you are recognised among the public as a company that knows how to work with data. What are your plans with particular regard to the application of AI and similar systems?
Srđan: Alongside the Rekomender, our focus is also directed towards innovating and creating new products based on machine learning. By carefully monitoring the needs of our users and industry trends, our aim is to continue pushing the boundaries and offering solutions that are always aligned with the latest technologies. In the period ahead, we will test new functionalities for the needs of the Poslovi Infostud website and we hope to present them to our users soon.