AI-Powered Treasury: Georgia’s Innovative Approach to Managing Payment Risks 

This article is published in association with IMF.

Davit Gamkrelidze, Giorgi Mchedlishvili, Graham Prentice, Alok Verma

Imbibing the spirit of Georgia’s national epic, “The Knight in the Panther’s Skin” by Shota Rustaveli, which celebrates the virtues of exploration and overcoming challenges, the Georgian State Treasury^[1] has embraced innovative modern technology to manage risks in Treasury payments.

Georgia has established a robust public financial management system (PFMS) to manage the government’s financial information. PFMS is a suite of four core and integrated systems, developed by the LEPL^[2] Financial-Analytical Service (IT Department): (i) e-Treasury, (ii) e-Budget, (iii) e-DMS, and (iv) e-HRMS. Payment and accounting processes are digitalized in the e-Treasury system and are conducted online. The payment process in the e-Treasury system is centralized, with all payment orders prepared by spending entities and submitted to the Treasury for processing. Payment orders are segregated for processing through two different channels, called the red channel and the green channel, based on pre-configured rules and validations in the e-Treasury system.

While the red channel payments are scrutinized primarily by Treasury operators, green channel payments are processed automatically and sent to the commercial banks though the Central Bank of Georgia. However, there are risks in payment processing through both channels—primarily related to data entry errors for payment amounts, beneficiaries, and account classifications—which can lead to erroneous payments and accounting and undermine the Treasury’s credibility.

The planned expansion of the Treasury Single Account to general government entities is expected to increase the volume of transactions processed centrally by up to 33 percent. This surge could potentially exacerbate the existing risks in payment processing due to the increased workload of the Treasury operators and a greater number of transactions passing through the green channel. 

In this context, in collaboration with the Fiscal Affairs Department of the IMF^[3], the Georgian Treasury leveraged its large fiscal datasets (average annual payment transactions above 7 million) in the e‑Treasury system to develop a prototype AI application aimed at improving the quality of review in the red channel and identifying anomalies in the green channel. The AI-based payment risk detection system was developed using these high-quality fiscal datasets to enhance data-driven decision-making and mitigate payment system risks. The system scrutinizes payments in both channels:

• Green Channel: The risk detection system, built on 42 AI models, learns underlying patterns from historical data to detect anomalies in green channel payment orders. This system successfully identified errors in 6 out of 8 previously known erroneous payment orders, demonstrating its effectiveness in filtering and scrutinizing payments before they are sent to the bank. When finalized and deployed, this capability would make the green channel more effective and allow the Treasury to route more transactions through it, thereby reducing the burden on the red channel.
• Red Channel: The second AI system, applied to the red channel, learns from past trends of scrutinizing payment orders by Treasury officers. This supervised learning-based detection system was able to identify over 80 percent of rejected payment orders in the payment dataset.

These models have been installed in a testing environment and will undergo further validation before being deployed live. The AI applications are expected to augment existing mechanisms, enhance the quality of Treasury reviews, and expand the coverage of the green channel, resulting in a faster, more trustworthy and more efficient payment and accounting system in Georgia. The eventual implementation of the finalized applications is anticipated to significantly improve overall payment risk management within the Treasury.

The Georgian Treasury’s development of a model for AI in payment risk management, while still work in progress, demonstrates the huge potential benefits of AI in public financial management. 

^[1] State Treasury is part of the Ministry of Finance of Georgia.

^[2] “LEPL” stands for “Legal Entity of Public Law”.

^[3] FAD’s support utilized funding from the Swiss State Secretariat for Economic Affairs (SECO) that supports building institutional capacity to ensure the appropriate adoption of digital innovations in PFM and leveraging digital innovations for data-based analysis.