Ministry of Finance and Economy: the AI-ONE platform for AI transformation (AX) in national fiscal administration

Background

Bringing AI into a public institution is not a matter of adding the latest technology. It has to run reliably in a secure environment cut off from the outside, and every result has to come from clear evidence such as laws, policies and internal documents. The entire process must stand up to audits and after-the-fact verification.

AI only takes hold in real work when security, reliability, controllability and explainability are secured together.

The Ministry of Finance and Economy’s AI-ONE platform is a project built on these requirements: it integrates scattered administrative data and puts generative AI to work in day-to-day tasks. Rather than simply adding a chatbot, it focuses on raising real-world usefulness by combining an AI specialized for the ministry’s work with the standards of the government-wide common AI.

Platform overview

The structure combines the standardization and scalability of the government-wide common AI with the work fit of the ministry-specific DocuLLM built by Cogniterm. It uses the common infrastructure as its foundation, while raising the document understanding, evidence search and context-aware answers that fiscal and economic administration actually needs to a level suited to the ministry.

The weight of this project lies in document centralization. Documents and data scattered across many places were consolidated into a central repository, together with a system for collecting, cleansing and processing them so AI can use them. Rather than introducing a single feature, the project ties the centralization of document assets and the use of AI into one.

The same structure can serve as a reference model for other public institutions.

Key features

01 Knowledge use built on document centralization

Scattered internal documents are managed in one consistent system and turned into a knowledge base that AI can use.

• Centralization of document assets scattered across G-Drive, web servers and PCs
• A foundation for AI use through document collection, cleansing and processing
• A system that runs from document management through search, use and drafting support

02 Combining the common AI with a ministry-specific AI

A ministry-specific DocuLLM sits on top of the scalable government-wide common AI to improve its fit for the work.

• A stable operating model built on the common AI infrastructure
• Document understanding and answers refined for the context of fiscal and economic administration
• Both standardization and practical, ministry-tailored usefulness

03 Evidence-centered answers

A search and answer system grounded in documents produces results that can be explained and verified.

• Answers generated by automatically referencing relevant evidence in documents
• Document-based Q&A for policy, fiscal and economic administration
• Sources that can be checked alongside each result

04 Operations built for audits

A public-sector operating model that accounts for sources, history, and review and approval flows.

• Tracking and management through stored query, answer and work histories
• A control structure in which staff review and approve drafts generated by AI
• The audit readiness and operational visibility required in secure public-sector environments

How it operates

Reflecting the security requirements and audit framework of public institutions, the platform is designed so that every answer comes from source documents and its sources and work history are managed. Even when AI produces a draft, the final review and approval rest with the responsible staff member.

• Sources cited in every answer to make it verifiable
• Query, answer and work histories stored to support tracking and management
• A review and approval process by staff after AI generates a draft
• Stronger links between document management and AI use through document centralization
• Stable operation even in environments with external connections blocked

Expected impact

The project goes beyond building one system: it offers a model for a generative AI platform that public institutions can use in daily work. By centralizing scattered documents and data and running the common AI and a ministry-specific model together on top of them, it reduces repetitive work and strengthens evidence-based administrative support.

Because document centralization and a ministry-specific AI are implemented within a single structure, this case can serve as a reference for other public-sector projects with similar requirements.