The financial case for AI begins with a contradiction: not all of its value is financial, but none of it should be left undefined. The companies that realize a return focus on economically meaningful workflows, define what better looks like, and establish how progress will be measured before changing the work.
Over the last three years of AI transformation work, the question I've gotten the most is the one about ROI. It comes up in nearly every engagement, which makes sense, but the answer is more nuanced than the question makes it sound.
If we think about AI in economic terms, the financial case begins by admitting that not all AI value is financial while simultaneously refusing to leave any of that value undefined. (After all, what’s measured gets managed and all that.)
But the reality is that not every benefit will appear on a P&L. Some of the value will show up as additional capacity, lower risk, better work, or a less frustrating employee experience. Those gains can eventually create financial value, but they do not become financial automatically. An hour saved only becomes economically meaningful when the organization decides what to do with that hour.
That does not mean AI cannot produce a measurable financial return. McKinsey studied twenty companies it considered leaders in technology and AI transformation. On average, their generated approximately $3 in incremental EBITDA for every $1 invested. Importantly, those companies concentrated their efforts in only one to three high-value business domains.
To be sure, the headline numbers are compelling, but we’re sort of burying the lede here. I actually think the most important finding is the last one: these companies didn't distribute AI randomly, sit back, and wait for value to appear. They identified the parts of the business where improvement would matter most, redesigned the work, and held leaders accountable for the results.
That is where I think the financial case for AI begins: with the economics of the workflow you intend to change.
Find the Economic Leverage Point
What you’re really looking for are economic leverage points in the business: places where better performance would materially affect the organization. In an industrial company, that might be estimate accuracy, equipment utilization, project throughput, maintenance downtime, win rate, or margin leakage.
Once you identify the leverage point, you trace it back to the workflows that influence it. Then you ask how those workflows could be redesigned around the right combination of people, AI, data, and systems.
This distinction matters because not every workflow deserves to be transformed. Saving someone ten minutes on a low-value administrative task may be helpful and you may decide to do it, but redesigning the estimating workflow so the company responds faster, protects margin, and wins more of the right work could materially change the economics of the business. As a result, you may decide to change that workflow first.
Nevertheless, when you find the right workflow, AI can create four kinds of value:
Financial value. The workflow directly increases revenue, improves margins, reduces costs, or strengthens cash flow.
Capacity value. The same people can complete more work, respond faster, or spend more time on the judgment-intensive work that creates greater value.
Risk and quality value. The organization reduces errors, rework, inconsistency, compliance exposure, or operational risk while producing more reliable work.
Human value. The work becomes less repetitive, frustrating, or cognitively exhausting. People gain confidence, exercise better judgment, and spend less time on tasks that drain their energy.
In best case scenarios, a single redesigned workflow may create all four types of value. But before the work begins, leaders need to identify which type of value is primary, which measures will indicate progress, and what result they expect to see.
The point here is not to force every benefit into a dollar figure. Rather, it’s to make every expected benefit visible enough to evaluate.
Define What Better Looks Like
AI transformation consultant Conor Grennan recently approached this question from another direction. Unlike most technologies, he argued, AI does not arrive with its own obvious measure of value.
A new machine can be measured in units per hour. A CRM can be evaluated through conversion rates and sales cycles. But what is the output of an analyst whose job is to understand the market? Or a manager whose most important contribution is keeping eight people moving in the same direction?
Grennan argues that the question leaders eventually have to answer is not simply, “What changed?” It is: “What did you expect to change?”
Asking that question matters because AI may already be creating value that the organization isn’t prepared to see. If no one established the baseline, defined the expected improvement, or decided which measure should move, there is no meaningful way to evaluate the result.
Before redesigning a workflow, establish how it performs today. Decide which of the four kinds of value you expect the new workflow to create. Set a target and a time horizon. Then watch what happens when people begin using AI inside the actual work.
Some workflows will become faster. Others will become more reliable or less frustrating. A smaller number may be reinvented entirely. The organization’s job is to recognize those changes, measure them honestly, and turn what works into a repeatable way of operating.
The ROI of AI appears when the organization knows what it wants to improve and changes the work accordingly.
Related Insights
