7% of leaders said their organization had established ROI from AI, even though nearly two-thirds said AI was already delivering meaningful business value

Feeling further along and being able to prove it are two different states, and most organizations are living in the first one.

One Number, Several Paths

Take a target like cutting resources per implementation by 20%. There is more than one legitimate way to get there: work that runs faster with the same people, a different mix of resources on each project, or catching quality issues earlier so people roll onto the next project instead of staying tied up fixing the last one. All three move the same top-line number. None of them show up the same way at the workflow level, which means deciding what success looks like has to include deciding which a redesign is walking, before assuming the connection is obvious.

The Fatigue Underneath

I have watched the fatigue that builds when this connection stays missing, on my own teams as much as anywhere else. Goals compound instead of replacing each other. Expectations shift. The pace of new asks keeps anyone from feeling like the last thing took hold before the next one arrives. Adding an AI mandate on top of that pile, disconnected from the metrics a team already lives by, reads as one more addition to a whirlwind rather than a way through it.

How I Build the Baseline

The approach I keep coming back to is the same regardless of the organization. Start with what the business already measures and already cares about, not a new number invented for the occasion. Look for where a workflow worth redesigning intersects with one of those measures, since that intersection is where a baseline is worth setting, and a surprising number of organizations do not have one yet, or have one that still needs to be formalized before it can be defended. Then decide up front how the change will be tracked.

At Epsilon, that last step meant going into the project tracking tools themselves and building in the categories that let time and process steps get pulled apart cleanly, original estimates against actuals, side by side. That structure is the reason I can name a specific number, 30 to 40%, about the workflows my team redesigned, rather than a general impression that things got better. The number came from the tracking, not from how the work felt afterward.

None of this requires a new system built from scratch. It requires deciding, before the work starts, which existing number the redesign is meant to move, and making sure the tools already in place can show the difference when it happens.

Has anyone connected the workflow you are about to redesign to the number leadership already watches? Can your tools show the before and after when the time comes?

Prepared by Forjic Technology. This is the Operational Readiness dimension of the Forjic AI Readiness Survey. For the full analysis behind all five dimensions, see Access Is Not Adoption. Source: KPMG International, Global AI Pulse Survey, Q2 2026, n=2,145.