Unifi Partnered with Zindi’s Generative AI talents to Solve Complex Data Extraction Challenges
Unifi has now developed an advanced tool for collating environmental, social, and governance (ESG) data from complex corporate reports, by tapping into the power of the generative AI talents through a Zindi competition. After mixed results from traditional consultancy firms, Unifi partnered with Zindi to host a GenAI competition on Zindi, where solutions are built by a community of skilled data scientists. This competition was specifically designed to address Unifi’s challenge of extracting ESG metrics found in PDF documents that include both text, charts, and images.This resulted in innovative approaches that significantly improved their productivity and efficiency.
PROBLEM
Unifi’s main constraint in scaling their services was the inaccurate measures and slow process of extracting ESG data from PDF reports. This data extraction is essential in providing comprehensive insights to their clients. Unifi had previously looked to a number of consulting firms for solutions but they found the outcomes unsatisfactory.
SOLUTION
Motivated to get the best results possible and to reduce the time and cost of the manual process, Unifi and Zindi hosted a challenge on Zindi’s platform with the objective of developing an AI solution for parsing annual reports in PDF format to extract predefined ESG metrics. Throughout the competition, Zindi guided participants by hosting a webinar to explain the problem in depth and conducting a Q&A session to address participants’ concerns.
The challenge was open for four months and it was accessible to all skilled AI practitioners across the globe. After the competition ended, with the help of Zindi’s automated evaluation and ranking systems, a real-time leaderboard which helps participants track their progress, the top 10 candidates were shortlisted based on their accuracy score.
Reflecting on the competition, the director of Unifi stated, “Zindi team struck a good balance preparing the competition in such a way that the submissions were comparable but the intrinsic challenges of the problem were not diluted.”
IMPACTS
With submission of the participants’ code, documentation and 10 minutes video explaining their solution and highlighting iterative learnings, the top three solutions were selected by JG and the team at Unifi. As a result of these participants offering a diverse range of approaches, Unifi is now taking each of the solutions and incorporating the different approaches applying them to the models that Unifi has running in production to drive the efficiency gains.
“This process gave us confidence knowing that the solutions proposed would be judged on not only solving the problem in an pragmatic way, but also practically deploy considerations,” said JG Cowper, Director Unifi. “Thanks to the Zindi team for helping us accelerate solving this problem for our clients.”
The winning solution achieved an accuracy of 53%, marking a meaningful improvement over the 48% accuracy achieved by the solution Unifi had previously sourced and had been using previously.
The partnership between Unifi and Zindi shows how innovative collaborations can be of great value. By tapping into Zindi’s network of generative AI experts, organisations like Unifi can source a guaranteed solution for their AI driven problems.
