Readying federal CFOs to unleash AI for cost recovery and operational efficiency

Recurring Costs – Salaries & Benefits

As agencies work to achieve their mission amidst resource constraints, finding new ways to incorporate AI to handle routine tasks so human workers can focus on larger items is critical.

So, what does the average workload look like for a government worker? According to a Deloitte analysis of O*NET Work Activity data and Office of Personnel Management Fedscope employment data in 2024, several of the most time-consuming tasks across the federal government are documenting and recording information, handling and moving objects and getting information.

In fiscal year 2024, roughly 43% of the $6.9T US government obligations were made up of up 3 main costs: grants (25.08%), contractual services (8.91%), and salaries & benefits (8.78%).

The good news, per a similar Deloitte study, is that over 50% of the sub-tasks within these activities have medium to high finance process automation potential. By leveraging AI to handle many of the routine or repetitive aspects of these tasks, government workers could focus on mission-critical priorities and higher value work.

One use case for automating repetitive tasks within the government: workflow management.

Solution 1: Intelligent Workflow Management

AI can significantly enhance workflow management and finance workflow automation by automating and optimizing various aspects of the process. For instance, the use of AI-driven tools like FEMA OCFO Response Augmentation Suite, used for automating the generation of initial and preliminary responses to questions, has reduced processing times of 15 budget questions per day on average, which previously took thousands of hours per year to research and respond to.

An agentic AI solution (Figure 1) can begin with an automated, 24/7 monitoring of incoming requests, upon which a manager AI agent can triage a series of AI agents to categorize, assess, and assign the task to the best matched staff (by availability and expertise) as well as AI agents to support documenting task or request statuses and notes and updating its knowledge repository in the background. Large Language Models (LLMs) are used to process and encode incoming requests to check against existing documents and databases, and LLMs are used to decode and generate output into natural language for the assigned staff as well as the final response to users.

By streamlining the initial stages of workflow handling, AI can significantly reduce the time spent on manual data entry and retrieval, allowing staff to focus on more complex tasks. Intelligent workflow management promotes efficient and effective case resolution through optimizing task assignments and reducing the need for task managers to start from scratch. Throughout the workflow management lifecycle, AI can help enhance service delivery by improving efficiency, accuracy, and resource allocation.

Let’s visualize what this means: Download Infographic

Non-Recurring Costs – Grants and Contracts

Non-recurring costs are primarily in procurement—for example, in fiscal year 2024, the US government collectively spends $3.2 trillion on grants and contracts. In addition to reviewing existing and pending agreements, it is equally important to create a sustainable review cadence, continuously evaluating future procurement needs for each period of performance and ensuring that funds are awarded intentionally.

To thoughtfully determine future awards, federal agencies and CFOs would need to have a comprehensive understanding of current and future drawdowns by vendors and grantees to accurately forecast undisbursed funds. However, vendors and grantees’ drawdown patterns are most often non-linear and therefore difficult to forecast (e.g., time and materials contracts, 5-year project period of performance grants). Additionally, many agencies rely on labor-intensive processes to review and certify open obligations, using mass emails, spreadsheets, and Enterprise Resource Planning (ERP) data extraction. Given that agencies can have thousands or millions of unliquidated obligations (ULOs) and undelivered orders (UDOs) at a time, this approach can significantly drain resources and divert staff from larger tasks such as identifying trends or forecasting issues.

Consequently, significant amounts of expired funds are often returned to the Treasury because ULOs were never identified or there was insufficient time to repurpose the money. On average, federal agencies collectively return approximately $24 billion worth of budgeted funds to the Treasury each year. With the right analytic capability and faster processing time, agencies could reduce undisbursed funds and de-obligate them faster for investment in other mission-critical activities.

Solution 2: Obligation Insights

AI can leverage the vast amount of historical financial data that government agencies possess to assess the risk of any UDO or ULO by the end of the period of performance. Paired with advanced machine learning (ML) models, the solution can be extended to chart out “low risk” and “high risk” spending patterns (Figure 2), making it easy for Contracting Officer's Representatives (CORs) and Project Officers (POs) to identify awards with potential spending risks and save significant time in gathering, analyzing, and visualizing data.

AI can be used for early detection of spending or performance issues. With minimal data analysis work from staff, the government can proactively identify issues, reach out to vendors or grantees to provide technical assistance if needed, and liquidate obligations before funds expire, effectively recovering funds to invest in other mission-critical needs. Alternatively, the government can decide to reduce future awards to grantees or vendors who require less funding in future performance years.

Let’s visualize what this means: Download Infographic