A CFO should evaluate AI opportunities by financial impact, not technology novelty. Map where manual work concentrates in dollars and hours, match proven AI capability to each workflow, and sequence deployments by expected return inside 90 days. The opportunities that clear 2x ROI and free staff capacity for higher-value work go first.
How should a CFO evaluate where AI fits in a health system?
The first question is not what AI can do. It is what your people are doing that AI should absorb. Every health system has processes that do not work the way they were designed. The gaps get filled by people. Your best people quietly carry your most broken processes, and the cost is invisible because it is absorbed by existing staff, not billed as a line item.
The CFO's job is to make that cost visible. Before evaluating any AI tool, price the status quo:
- Staff hours consumed by manual workflows. How many hours per week does your revenue cycle team spend on rework, retyping, and chasing denials? What does that cost at loaded labor rates?
- Denial rate and rework cost. What percentage of claims are denied? What is the average rework cost per denied claim? At $25 to $181 per claim, the math compounds fast.
- Revenue cycle lag. What is your average days in accounts receivable? How much faster would claims be paid if they were clean on first submission?
- Prior authorization backlog. How many physician hours per week go to prior auth? At 13 hours per physician per week, the opportunity is measurable.
What is the ROI of AI in healthcare operations?
Most health systems that rigorously measure AI returns report at least 2x ROI. That is the bar. The return shows up in three places:
- Denial reduction. Fewer claims entering the appeals queue because they were scrubbed and verified before submission.
- Staff capacity returned. Hours redirected from rework to higher-value work. This is the return most CFOs underestimate because it does not show up as a line-item savings. It shows up as capacity.
- Cash flow acceleration. Claims paid faster because they are clean on first submission. Days in AR drop. Cash flow improves. The balance sheet moves.
Every deployment should start with baselines you sign off on, so returns are measured in your numbers, not vendor claims. Work is sequenced to prove value inside the first 90 days. If a workflow can not clear 2x ROI, the recommendation is to stop, in writing.
How does a CFO build the business case for AI?
Start with the cost of the status quo. Quantify what your best people are spending on work that should not exist. Price the hours. Price the denials. Price the lag. Then map each AI opportunity to a specific workflow with a baseline, a 90-day proof window, and a dollarized return.
The business case writes itself when the before-state is priced honestly. The problem is not that the numbers are hard to find. The problem is that nobody has been asked to find them. The administrative cost is absorbed so deeply into the operating budget that it does not have a name. It is just the way things work.
AI changes what is possible, but the financial case is built on the cost of what is already happening. Here is the structure:
- Baseline the current state. Dollars and hours. No estimates. Pull the actuals.
- Match proven AI capability to the workflow. Not what AI might do someday. What it is doing in health systems today.
- Set a 90-day proof window. Deploy, measure, and evaluate. If it clears 2x ROI, scale. If it does not, sunset.
- Assign an owner. Every workflow needs a named owner inside your organization. No owner, no deployment.
- Price the capacity returned. This is the number your board cares about. How many hours come back to your organization, and what do those hours get redirected to?
Where do AI deployments fail in health systems?
AI deployments fail when they are tool-first instead of work-first. A vendor sells a product, the health system buys it, and nobody maps the workflow it is supposed to fix. The tool sits on top of a broken process, and the broken process stays broken.
Pilots stall because they are unmeasured. There is no baseline, no owner, no proof window. The pilot runs for six months, nobody can say whether it worked, and it fades into the category of things we tried.
Staff resistance kills more AI projects than technology does. When your best people feel threatened rather than supported, they will quietly ensure the tool does not work. Adoption and change management are not footnotes. They are part of the deployment, and the CFO should budget for them.
What should a CFO ask an AI vendor?
Ask for the baseline and the proof window:
- What workflow does this tool replace?
- What is the current cost of that workflow in dollars and hours?
- What does the 90-day measurement look like?
- Who owns the deployment inside our organization?
- What happens if it does not clear the ROI bar?
If a vendor can not answer those questions, the tool is not ready for your operations. The right answer from a vendor is not a product demo. It is a work map, a baseline plan, and a measurement framework.
The bottom line
The CFO's role in AI adoption is not to buy technology. It is to price the status quo, sequence the opportunities, and hold every deployment to a measurable return. The technology exists. The proof exists. The question is whether your organization maps the work, installs the tools, and measures the results, or continues absorbing the cost with your best people's time.
Want help building the business case?
Schedule a Strategy Call. Twenty minutes, no cost. We'll map where manual work concentrates in your organization and what AI can absorb today.
Schedule a Strategy Call