Stredoslovenská distribučná, a.s. (SSD) is a distribution energy company operating in Žilina, Banská Bystrica, and part of the Trenčín regions, where it distributes electrical energy to more than 770,000 customers
Ableneo team closely collaborated with our team during the three-month testing period and provided on-the-job training. The process of continuous testing & tuning led to an increase of already sufficient success rate and convinced us to proceed with the solution into production.
Summary
SSD processes over 20,000 building permits annually, facing high variability in document types and a pressing need to reliably capture critical information. By implementing a custom automated solution, SSD achieved 100% recall—ensuring no critical documents were missed, significantly reducing manual verification. The system also categorizes documents into 26 predefined classes with a 97% accuracy rate, enabling SSD to operate 24/7 while freeing its team to focus on higher-value tasks.
Challenge
What was the problem
SSD receives over 20,000 building permits per year from state and public administration, local governments, building authorities, and district offices, which are manually processed and forwarded for further processing or archiving.
The major challenge of the project was the variability of the incoming documents and the requirement to capture all critical documents, while minimizing additional control over false positives.
Solution
How we solved it
We solved it by conducting a thorough evaluation and analysis of 35 workload questionnaires to understand employee tasks and identify inefficiencies.
From there, we performed a deeper analysis of the six processes with the highest potential for automation. To showcase potential improvements, we developed comprehensive high-level process flowcharts, serving as a foundation for clearer responsibilities and streamlined workflows.
Our approach combined:
- Kognitos AI,
- prompt engineering
- consulting and interviews,
- workload analysis.
During the deployment of the solution into full production, testing was conducted over three months on 3,373 previously unseen documents.
Outcomes
What has changed
The solution captured 100% of critical documents (100% recall), while 27% documents required additional verification (precision 73%).
In categorization to 26 categories, 97% accuracy was achieved.
Thanks to this, the person who previously performed this mechanical work can now focus on tasks with higher added value. This solution operates 24/7.