AI Pays Off When You Redesign Around Value, Not Workflows

AI adoption tends to progress through three levels of work. Each level changes what the organization improves and how success should be measured. Here's how they break down:

1. There's the task: a single activity AI assists or automates—the kind of thing you'd measure in minutes saved or output per person.
2. There's the workflow: a multi-step process redesigned from end to end, measured in cycle time and error rate.
3. And there's the value flow: the full path from a signal or customer need to a business outcome, crossing whatever functions and systems it has to cross to get there.

A lot of enterprise AI investment still sits at the first level. The current wave of advice— redesign your workflows—moves it to the second.

The third level is where growth, margin and customer experience get made or lost, and few companies report organizing, funding or measuring much of anything there yet. In our own research into how 1,550 enterprise AI decision-makers are using the technology, only 18 percent said AI is fully integrated across their enterprise, and 42 percent said their organization isn't built to capture the value AI can already create.

Gartner found something similar looking at finance functions specifically this year: In a March 2026 survey of finance leaders, 45 percent said their AI investments lean toward productivity, and only 20 percent said they lean toward decision quality. For Gartner, pilots and use cases in production are signs of progress, but they are not proof that AI is delivering the value boards expect.

Organizations are counting the wrong things, at the wrong altitude.

A workflow redesign, on its own, tends to inherit the boundaries of whichever function owns it. A faster claims process is still a claims process. A faster onboarding flow still ends with onboarding. Real value rarely respects those lines—a customer need, a product launch, a resolved claim moves through several functions and systems before it turns into anything measurable, and a lot of that value gets lost along the way.

There's a name for this problem: the seam tax. It's what gets lost every time work crosses from one function into another. A handoff that sits in someone else's queue, information that gets re-entered or summarized until it starts to lose its original meaning, a decision that waits because no one owns the outcome, only their piece of it. The cost can look like a delay, duplicated controls or a team hitting its KPI while creating a worse result further down the line.

It's hard to put one clean number on the seam tax across every company, but its presence is consistent in our research. Asked what's limiting their ability to scale AI, leaders point to integration across systems (38 percent) and data fragmentation (36 percent) more than anything else, with governance and control friction (29 percent) and organizational silos (26 percent) close behind. Each of those is a symptom of the same tax, exacerbated at the handoff between functions, unreachable by a workflow redesign.

Leaders know this, even if they don't always put it in these terms. Sixty-five percent expect real progress scaling AI over the next year or two. Fifteen percent say their organization is actually equipped to deliver it. Asked what's really holding AI back, the same group was about twice as likely to point at how their organization runs as at the technology itself: 22 percent blamed the operating model, 11 percent blamed AI's capability. Nearly half, 47 percent, said AI is already good enough to meet what the business needs from it today. The organization is what's holding back value from AI investment.

A workflow redesign is straightforward to write down. Getting to a value flow is both harder to write down and harder to build. It means mapping the business logic actually buried inside a process, connecting it across whatever systems and functions it touches and standing up a working version fast enough that leaders can see real value before they commit to scaling it everywhere.

This is where our approach differs from the general advice to break down silos. We don't start by redrawing the org chart. We start with a practical sequence:

• Choose one, meaningful value pool
• Establish the process baseline, including its current cost, speed, quality and business impact
• Connect the systems, data, rules and context behind it
• Redesign the work by deciding what AI should automate, what stays with a person and what shouldn't be done at all
• Build and test a working version against a business case with real numbers attached

The sequencing is the whole point. A commercial bank used it to take a manual, document-heavy lending process and cut most of its cost out, on a path toward halving time-to-cash. A health insurer used the same approach to move from reacting to claims after the fact to catching risk earlier, saving roughly $15,000 for every emergency visit it prevented. A global insurer used it to replace a patchwork of local AI experiments with one governed architecture, cutting cross-border deployment time by 80 percent.

In each case, the win came from following one path all the way through—from the task, into the workflow, out to a measurable business outcome—rather than optimizing any single piece of it in isolation.

None of this works if AI just gets pointed at whatever a team is already doing. As our CEO Nigel Vaz has put it, the goal isn't a faster caterpillar—it's a butterfly. Running AI through an existing process without asking whether that process still deserves to exist is how a company ends up with expensive automation and no real change in how the business performs. The organizations getting value out of AI are the ones willing to rebuild decision-making, workflows and customer experience around what's newly possible, instead of layering AI onto what's already there.

The harder part of that shift is rarely technical. Whether a redesigned value flow makes it past the pilot comes down to:

1. Incentives, KPIs and whether the people doing the work trust it enough to actually change how they operate.
2. Whether it can move past one team and into the rest of the business comes down to something else entirely—connected data, modernized systems, shared context and governance built for scale.

Both of those have to be engineered together, deliberately, rather than treated as separate workstreams that happen to share a kickoff date.

The market has already moved from tasks to workflows. The next question is whether companies can make the shift from workflows to value flows: identifying the handful of paths that actually determine growth and resilience, and organizing funding, ownership and measurement around those paths instead of around whichever function happens to own a piece of them.

Functions still matter. They remain the home for expertise, standards and control. But the value flow becomes the place where transformation is funded, owned and measured.