AI adoption does not always produce financial returns in enterprise operations

Podcast – AI Adoption Does Not Mean AI Returns

Insights Podcast

Insights

Podcast – AI Adoption Does Not Mean AI Returns

June 24, 2026 | Podcast

Read the transcript

Transcript

An organization can report strong AI adoption and still produce zero financial return.

I am Josh, and welcome back to the City Shift Finance podcast.

Today we are discussing why AI deployment produces adoption without producing returns, and why the financial consequences of that gap tend to arrive well after the investment that created them has already been made.

Adoption metrics were designed to measure deployment progress, and they do that accurately, but they were never designed to measure financial impact, and organizations that treat them as a proxy for return are measuring the wrong thing with a high degree of precision.

The condition begins in how organizations have been measuring progress.

Capital has been committed, systems have been deployed, and usage is being reported upward through the organization. The deployment appears to be on track because the signals being measured, how many people are using the AI, how frequently, across how many functions, are all moving in the right direction, and leadership is receiving confirmation at each reporting cycle that the investment is working as intended.

What those signals do not measure is whether the financial performance of the organization has changed as a result, and by the time leadership begins asking that question, the investment has usually been running long enough that the cost is already embedded in the operating structure and the teams responsible for the deployment have moved on to the next phase.

Financial underperformance is usually the first signal that reaches the numbers.

"Organizations often discover pricing pressure in their financial statements months after customers have already changed their buying decisions."

The AI-generated outputs are entering the same approval structures, the same review cycles, and the same decision processes that existed before the deployment began, and those structures were built for a world where information arrived slowly and decisions moved at the pace the governance was designed to support. Approval chains built for weekly cycles receive daily outputs and process them weekly.

Review structures designed for human-speed information receive AI-speed information and apply the same governance they always have, and the technology has changed the pace at which information enters the organization without changing the pace at which the organization is permitted to act on it. The gap between what the deployment assumed the organization would do with its outputs and what the organization is actually doing with them is where the return quietly disappears, and by the time that gap surfaces in a performance review, the investment producing it has already shaped the decisions that followed.

Operational planning deteriorates from the same source.

When deployment is built against an assumption that the organization will naturally absorb and act on AI-generated outputs, every planning cycle draws from that assumption without testing whether it holds, and the financial projections attached to the deployment were built against a version of the organization that would receive faster information and convert it into faster decisions.

That version of the organization does not yet exist because the decision structure was never redesigned to support it. Each output the system produces is internally consistent with what the deployment was designed to generate, and the organization continues running on a planning assumption that is coherent in isolation and increasingly disconnected from how decisions are actually being made.

The projections will require revision, and those revisions will keep coming until the assumption is corrected rather than the forecast.

AI deployment also changes the labor cost equation in ways that do not appear immediately.

When AI generates outputs that previously required significant staff time, the expectation is that labor costs will fall or that the same headcount will produce materially more value, and in the early stages of deployment that expectation appears to be confirmed by the volume of output the system produces. In practice, the staff time freed by AI-generated outputs gets absorbed by the review and governance activity that the unchanged decision structure requires, and headcount stays flat because the back-end process still requires the same number of people to move a decision from output to execution, while the volume of information requiring review has grown faster than the governance structure can process it. The payroll line stays fixed while AI expense is added on top.

"Organizations that protect margins do not wait for the numbers to explain what the market has already made clear."

Let me show you what this looks like in practice. A finance function deploys AI across its core workflows, and within the first quarter adoption is above 80% and outputs are being generated faster and in greater volume than the function produced before. Leadership considers the deployment a success and reports progress upward, while the financial statements continue reflecting the same performance levels they showed before the deployment began.

The outputs are entering a review and approval process that has not changed, and the decisions those outputs were supposed to inform are still taking the same amount of time to reach execution, because the governance structure was never updated to match the pace the technology created. The organization is carrying the full cost of the deployment and none of the return, and the gap between those two things has been widening quietly across every reporting cycle since the deployment went live.

The organizations that produce measurable financial returns from AI share a recognizable characteristic.

They treat the deployment and the organizational redesign required to capture its output as a single project, and neither moves forward without the other. Before any deployment begins, they identify the specific financial line item the deployment is expected to move, define what movement on that line would look like within a defined period, and redesign the decision process that sits between the output and the outcome. The governance structure is updated before the deployment goes live, approval cycles are rebuilt to reflect the pace at which AI-generated information arrives, and accountability for acting on outputs is assigned before the first output is produced. The distance between what the AI produces and what the organization does with it stays narrow enough that the financial performance reflects the investment, and the return appears in the numbers because the conditions required to produce it were built into the structure before deployment began.

That discipline is what AI returns actually require.

Not higher adoption rates or more functions brought onto the system, but the organizational capacity to receive what the AI produces and convert it into a financial outcome at the pace the AI makes possible, and to build that capacity before the deployment goes live rather than in response to a return that has failed to appear.

The organizations that lose that capacity lose it through deployments that proceed without a corresponding change to how decisions are made, through governance structures that continue operating at the pace they were built for while the information feeding them accelerates, through a period where each output appears useful in isolation while the cumulative effect on financial performance quietly widens the gap between what the deployment was projected to produce and what the business is actually generating.

The deployment keeps running, the adoption metrics keep climbing, and the organization keeps measuring progress against a version of itself that the financial statements are not yet describing, and by the time that gap becomes a performance problem, the financial return that the deployment was supposed to produce has already been spent on the cost of running a structure that was never designed to capture it.

Thanks for tuning in.

About the Host

Josh is the Director of Strategy at City Shift Finance, leading the firm’s Insights podcast and episodes on financial, workforce and operating decisions.

Podcast – AI Adoption Does Not Mean AI Returns

Podcast – AI Adoption Does Not Mean AI Returns

Outcomes Defined by Impact

Sign in

Create your account

Verify your email

Sign in

Contact us

Contact us

Contact

What we do

Recent work

Who we are

Featured

Sign up to download