chIPA: Bulgaria's digital communication revolution powered by the Institute of Public Administration

The problem

Bulgaria's Institute of Public Administration (IPA) faced significant communication challenges related to the high volume of frequently asked questions received by phone and email. These routine inquiries concerned IPA services, courses, events, and administrative procedures, and handling them manually required considerable staff time and effort. This created inefficiencies in communication, slowed down response times, and limited the capacity of employees to focus on more complex tasks. At the same time, there was a need to modernise communication channels and strengthen digital skills within the institution in line with broader public administration trends. The project therefore aimed to address both operational inefficiency and the need for greater digitalisation.

The solution

To address these challenges, IPA developed and introduced chIPA, a chatbot designed to provide fast, accessible, and efficient communication support. The digital assistant was developed using GPT-4o-mini and integrated into a system that combines a vector database — a type of database optimised for searching by meaning rather than exact keywords — with a Maximal Marginal Relevance (MMR) algorithm to retrieve relevant information while minimising repetition in results. MMR is a technique used in information retrieval that balances how relevant each result is to a user's query with how different it is from results already selected, so that the chatbot can pull together responses that are both on-topic and varied rather than repeating the same point in slightly different ways.

The project began with a needs assessment, supported by focus groups with IPA staff and external users, in order to better understand the most common communication problems and identify user needs. The results confirmed that a large share of inquiries were repetitive and could be handled through an automated digital tool.

Based on these findings, IPA established a working group to design and develop the chatbot. The internal costs were primarily associated with staff time and resources — IPA employees participated in the needs assessment, testing, and validation phases, which required a significant time commitment. The IT specialists and technical consultants who developed the chatbot's prototype and ensured its integration with IPA's existing systems worked on the project as volunteers. The solution was created to answer frequently asked questions, provide information about courses and events, and guide users through administrative procedures. The aim was to reduce pressure on traditional communication channels, improve access to information, and provide support at any time.

Before its full launch, chIPA went through a pilot testing phase with a selected group of users. This made it possible to identify technical issues, verify the quality of responses, and collect feedback for improvement. After the first adjustments, the chatbot was tested internally by IPA employees and then by selected external users from different administrations. This step-by-step approach ensured that the system was functional, relevant, and user-friendly before wider deployment.

After testing and validation, chIPA was officially launched on the IPA website at the end of 2024. Since then, its development has continued through regular monitoring, user feedback, and ongoing updates, ensuring that the chatbot remains effective, relevant, and up to date.

The impact

The integration of chIPA led to improvements in communication efficiency, user engagement, and digital capacity at IPA. These outcomes were determined based on initial usage data and qualitative feedback.

1. Reduced pressure on traditional communication channels
The chatbot helped reduce the number of frequently asked questions handled through phone calls and emails and improved response times by providing 24/7 access to information. This was tracked through reductions in phone calls and emails, increases in chatbot interactions, and user satisfaction ratings.

2. Improved accessibility and user satisfaction
chIPA made IPA's services, courses, events, and administrative information more accessible at any time. User satisfaction was measured through post-interaction feedback, while qualitative input provided insights into the chatbot's perceived ease of use and the relevance and usefulness of the information provided. Engagement was also tracked by comparing the number of users interacting with the chatbot against total website visitors.

3. Early steps towards digital transformation
The chatbot supported the automation of routine administrative communication and contributed to more efficient internal processes. It also played a role in strengthening employees' digital competencies by encouraging the use of new technologies and supporting training efforts. This was measured through time saved by employees on routine tasks and the number of successful administrative queries processed by the chatbot.

Ongoing monitoring
The long-term effects of chIPA are expected to be monitored through continuous data collection, ongoing analysis, and regular feedback from both users and employees, ensuring that the system remains effective and responsive over time.

Lessons learned

• Involve employees and users from the start — and keep them involved. One of the main lessons is that involving both employees and service users from the very beginning helps ensure that the chatbot responds to real needs and challenges. At IPA, needs assessments, focus groups, and continuous feedback enabled improvements to the chatbot over time, making it more user-friendly, relevant, and effective. This shows that institutions should not limit stakeholder involvement to the planning stage, but should maintain it throughout implementation and further development.
• Anticipate technical challenges — including hallucinations. When creating digital assistants, institutions will encounter various challenges, from technical issues to instances of "hallucinations" — where the chatbot generates responses that sound plausible but are inaccurate. This happened with chIPA as well. The implementation team managed this through constant monitoring and adjustments, mainly outside standard working hours; phased involvement of testers, starting with five people and the Executive Director personally, then expanding to other IPA experts and volunteers from different institutions and communities across the public administration system; extremely strict control over correct answers; regular enrichment of the knowledge base; and adherence to the principles of ethics, integrity, and good cyber security practices.
• Plan for continuous improvement. chIPA is not a static tool. Regular monitoring and updates are important to maintaining its relevance and effectiveness. IPA ensures the chatbot evolves alongside changes in user behaviour, new technological developments, and adjustments in public service processes. This proactive approach ensures the chatbot stays relevant and efficient in meeting its goals.
• Start with a pilot before full deployment. Another important lesson is the value of testing the chatbot on a smaller scale before fully launching it. The pilot phase allowed the team to check functionality, identify bugs, verify answers, and collect direct feedback for improvement. For other institutions, this demonstrates that starting with a pilot can reduce risks, solve problems early, and lead to a more reliable and effective final product.

At the time of chIPA's implementation in late 2024, the EU AI Act had already been adopted but had not yet entered into force. Moreover, there was no specific Bulgarian AI legislation in place at that stage, only a general national concept outlining the development of artificial intelligence until 2030.