Yes, Everyone Has AI. But How Do You Measure It?

In 2024, telling your board you had AI in production was a differentiator. In 2026, it isn't.

What CEOs, CFOs, and COOs say more than anything else right now is some version of: we know we need AI. The AI we’ve implemented seems to be working. We know we need to be measuring it. We do not know exactly how.

Institutions that are figuring it out are pulling ahead of those that aren't. So let’s break down what the ones who figure it out actually do.

The baseline most institutions do not have

Measuring the real cost of a single workflow is difficult across labor, handoffs, rework, QC audits, and the opportunity cost of what the team is not doing while they do that work.

The first time an institution runs that calculation, the number is almost always higher than expected. For example, one regional credit union we work with looked at a single workflow and came back with a per-file cost in the low single-dollar range. That became the number they are now improving against.

Without that baseline number, every AI vendor sounds equally impressive on a sales call, and no one is accountable to a specific outcome after the contract is signed.

The four metrics that actually matter

Once an institution has a baseline, four metrics shape whether an AI investment is working.

• Per-workflow cost. Where you started, where you are now, how it is trending.
• Workflow-level accuracy. The percentage of complete workflows that run end-to-end without surfacing an exception. This matters more than field-level accuracy, which can be very high while workflows still break constantly.
• Capacity freed and redirected. Hours per week your team is no longer spending on the automated work, and what they are doing instead. If those hours do not get redirected to higher-value work, you have not created value. You have simply rearranged it.
• Deployment velocity. How long it takes to go from a signed contract to a live workflow, and how long each subsequent workflow takes after the first. If workflow two is not meaningfully faster than workflow one, your vendor is not compounding.

CFOs that can report these four to their board every quarter have a defensible view of whether their AI investment is working. 

The vendor question has shifted

Two years ago, the question for an AI vendor was whether they could do it at all. In 2026, nearly all of them can do something. The real question is whether the value they claim actually holds up when you measure it.

Vendors with real production results welcome the measurement conversation. They have no trouble sharing their per-file cost before and after, their workflow-level accuracy, their average deployment time, and what current customers are doing with the capacity they got back. The vendors without real results either do not have the numbers or give numbers that cannot be independently verified.

It’s key to dig into measurement and ROI conversations before signing, not midway through implementation.

What this looks like at the board level

Boards are getting sophisticated about AI faster than most people may have expected. We are long past the board demanding AI implementation and sophistication. We are at the second phase of ROI measurement and true impact. 

To think about this at the board level, you need to have a plan in place before you start talking to vendors. What processes are costing us the most right now? What would success actually look like in dollars and hours?

Two things that work right now as you implement:

1. Build the measurement framework before you sign the next AI contract.
2. Make the next renewal conversation a measurement conversation. If a vendor cannot show the numbers in writing, you have your answer.

One more thing

ROI of AI is really a question about operational discipline. The institutions that will come out ahead over the next three years are the ones that treat AI like every other operational investment. Manage it, measure it, and hold the vendors accountable to it like you would any other operational change.