At Kondana, we are excited to keep expanding YALI, our Identity and Access Management (IAM) solution, with features that delight our customers. One recent area of focus has been reporting and event monitoring, capabilities that customers rely on heavily but often find time-consuming to use.
The Friction Point
The challenge was straightforward but recurring. When a customer needed a report, for example on login activity, they typically relied on their internal teams to generate it. A request would go in, a query would be written, a draft shared, and then several revisions would follow. This cycle often repeated four or five times before the right report was finally produced.
This back and forth created inefficiency: customers spent time clarifying requirements, and their teams spent time rewriting queries. We saw an opportunity to remove this friction.
Building Smarter
Late last year, we began experimenting with an AI-powered chatbot integrated into YALI. The idea was to let users type questions in natural language and receive real-time responses without requiring technical query writing or multiple rounds of handoffs.
Our first version was built using prompts. We created structured instructions for specific use cases such as login issues and password resets. While it proved the concept, this approach had limitations. Each new case required us to revisit earlier ones, prompts grew unwieldy, and the system sometimes misinterpreted identifiers like usernames versus system IDs.
To overcome these constraints, we rebuilt the chatbot using a Retrieval-Augmented Generation (RAG) approach. The most intensive part of this process was designing effective training data, but it made a significant difference in accuracy and reliability. The result was a chatbot that could handle a broader range of reporting queries with far less manual intervention. More importantly, it gave customers a faster, more intuitive way to access insights, reducing the cycle from days to minutes.
Learnings
This project gave us several valuable insights:
Training data is foundational. The quality and structure of training inputs directly influence chatbot performance.
Prompts have limits. They are helpful for prototyping, but sustainable solutions need proper context retrieval and fine-tuning.
Natural language opens doors. Querying enterprise systems without technical expertise is not only possible, but increasingly practical.
Automation matters. By reducing repetitive report generation cycles, customers can free up teams for more meaningful work.
What we learned through this implementation is that AI integration is not about adding a feature, but about rethinking how problems are solved. At Kondana, we continue to integrate AI capabilities thoughtfully into our solutions, not as standalone experiments but as tools that remove friction, improve efficiency, and deliver practical value for customers.