In order to price an AI agent, CFOs must map the process it touches, redefine its role as both technology and labor, and shift from measuring seats to outcomes.
Bryan DeGraw
Associate Principal – Finance Solution Intelligence
Workday
When I’m talking to a CFO about strategic transformation, the first thing I tell them is to throw away their P&L reports, budgets, and org charts. I’ve gotten some shocked looks and raised eyebrows. But it’s the right place to start. Because the way we’ve done things in finance for decades is changing.
Every finance leader knows how to onboard a new hire. There's a role, a salary, a manager, a place on the org chart. Every finance leader also knows how to onboard new software: a license, a deployment timeline, a line in the budget. Between those two, a company can pay for almost any kind of work it needs to do. Until now. AI is challenging how companies budget, onboard and pay for agentic output.
The old system was built for people and software, and it can’t handle an agent that is both.
An agent handling collections or drafting journal entries doesn't have a skip-level manager or a reporting line. Nor does it have a license fee that scales. It has a workflow. The org chart shows titles, and the P&L shows costs, but neither shows the work itself—the sequence of steps an agent or a person actually performs.
The old system was built for people and software, and it can’t handle an agent that is both. So how do companies get ready to integrate this new digital workforce into the human one they have managed and supported for years? Pricing, funding, and managing an organization with both people and agents requires a new strategy and perspective.
Here’s what finance leaders need to understand about the economics of the new digital workforce.
Report
The first step is for companies to map their own processes. When Hackett benchmarks a function—general accounting, accounts payable, tax, treasury, planning—we hand the client a process definition first. We ask them to identify, across the whole organization, who is actually doing that work. Capture the FTEs, the time, the cost.
This analysis, depending on the size and scope of an organization, is usually more difficult than it sounds. And there are always surprises.
I remember a treasurer who told us, with total confidence, “I’ve got 10 people doing this. This will be a piece of cake.” We ran the data-collection exercise across the broader business and found roughly twice as many people scattered around the organization doing treasury-type work that never showed up on his chart.
Without this step, you will not accurately capture your own cost, especially with agents in the picture. As a result, your attempts to predict the savings from integrating agents into your workforce will be inaccurate.
Once you have captured the big picture of the work, you can start looking at where agents fit into the workflow and are adopting tasks in your process.
Once you have captured the big picture of the work, you can start looking at where agents fit into the workflow and are adopting tasks in your process.
It's important to understand what qualifies as a “digital worker.” The definition has broadened. An AI agent is treated as software in an engineering sense, but it becomes labor in an economic sense. From one perspective, it’s a liability, from another, it’s an asset.
We’ve found it beneficial to track digital labor like a balanced scorecard: part financial, part operational, with costs and benefits scattered across technology, labor, process, and business-performance lines. It’s not simply an agent with a name tag, so the scorecard approach gives you insight into this new category that is both software and labor.
Once you have your scorecard, it’s time to assess agent capacity, how agents fit into your organization, and then build a financial governance strategy for agentic AI within your operating model.
Traditional software economics were predictable: a license fee, a deployment cost, then a steady charge that didn’t move much with usage. Volume dictated fees regardless of outcomes.
Digital labor is different. Think of it like leasing a car with a mileage allowance, except the mileage is visible in real time—and a low bill isn’t automatically good news. Agents are employed to improve productivity, reduce costs, and increase quality of work. They need to be used to convert cost to value. Finance leaders need to know whether heavy usage reflects real value or an unbudgeted cost spike.
This is why the traditional approach to budgeting is changing. If digital labor doesn’t map to seat count, what should CFOs budget against instead? The focus shifts to the work—a unit of productivity versus a unit of software access. In practice, that means budgeting for outcomes an agent is meant to deliver:
This is very different than budgeting for the number of software licenses that are provisioned.
As budgeting starts to shift from seats to outcomes, it's easy to confuse faster cycle times with a lower cost base. Time saved isn’t necessarily cash saved. If cycle times go down and error rates improve, the same people may still be doing the same work. Unless headcount, or the underlying cost structure, actually changes, the business hasn’t realized the savings—it’s only made the existing team do the same thing faster. Finance leaders also need to take the “integration tax” into account. This is the time employees spend fact-checking AI, bridging systems, and managing extra reconciliations and exceptions that need manual cleanup before a transaction can flow through. It’s a cost that persists whether or not a company automates their process with AI, and one that has to be identified before the value of any new technology can be fairly measured.
Stop financing AI as a collection of tools and fund it as a portfolio of operating outcomes.
The Hackett Group’s 2026 AI World Class Finance study found that organizations redesigning entire processes around AI, rather than simply automating existing tasks, can create performance advantages of up to 75% across the enterprise. In order-to-cash alone, AI World Class organizations achieve:
This shows up only when process redesign, not just tool deployment, does the heavy lifting.
I encourage my clients to stop financing AI as a collection of tools. Fund it as a portfolio of measurable operating outcomes. Don’t think of it as buying a tool. Think of it as solving a problem.
Identify the needs first. In practice, that means starting with the business outcomes finance leaders are already accountable for:
Let the technology follow from there, rather than buying capability first and searching for a use case after. This will help you adapt your strategy, approach, and tools to make sure you get real value out of your new workforce of agents and people.
Report
A gap between effort and impact is the real ROI problem. Discover the solution.