On April 23, KPMG called a meeting.

The firm's US audit partners were informed that approximately 100 of them — roughly 10% of the partnership — would be leaving. Not because performance was poor. Not because the business was struggling. KPMG's audit practice is growing. Over the past two years, it has picked up more listed audit clients than any of its Big Four competitors. ¹

They were leaving because the economics of how an audit gets done are changing.

AI is increasingly handling key steps of the audit process, spurring firms to rethink how work flows through junior and senior teams alike. ¹ Less work requiring large junior teams means less supervision required. Less supervision required means fewer senior partners needed to run it. The partnership structure built around the old model is being resized to fit the new one — whether the firm planned it that way or not.

Two weeks earlier, on April 7, EY made a different kind of announcement.

The firm rolled out enterprise-scale agentic AI across its entire global audit practice — embedding AI agents directly into the audit platform used for 160,000+ engagements worldwide. ² The rollout integrates 150 AI agents into audit workflows, with the goal of supporting all end-to-end audit activities by 2028. EY describes this as part of a multibillion-dollar commitment to audit quality and transformation. ²

Same industry. Same technology pressure. Same economics reshaping the profession.

One firm is being changed by it.
The other is choosing to direct it.

That gap — between being reshaped by AI economics and deliberately steering them — is where most AI strategies are currently breaking.

And it's not visible on any adoption dashboard.

Two Scoreboards

For the past several years, the dominant metrics in AI have been deployment metrics.

How many tools are running. How many employees are using AI. What percentage of workflows have been touched. How much output is being generated.

These are the numbers that show up in press releases, earnings calls, and board updates.

They are also largely disconnected from financial performance.

There are, in reality, two scoreboards.

The first measures activity.
Adoption. Usage. Volume.

The second measures outcomes.
Cost. Revenue. Margin.

Most organizations are winning the first scoreboard. Very few are winning the second.

The Illusion of Progress

Adoption is no longer the constraint.

Across large organizations, AI is everywhere. Training is complete. Tools are deployed. Workflows are AI-enabled. From the outside, it looks like transformation.

Inside the financials, it often looks like very little has changed.

Costs haven't moved in proportion to productivity gains. Revenue hasn't accelerated at the rate investment would suggest. Margins remain largely intact.

This creates a dangerous dynamic: the presence of AI begins to stand in for the impact of AI.

Activity becomes evidence. Usage becomes proof. And the question that actually matters — what changed financially? — gets pushed aside.

When Measurement Drives Behavior

Once activity becomes the metric, behavior follows.

In several large organizations, AI usage has already been incorporated into performance expectations. How often people use it. How much. How many workflows include it.

The response is predictable.

Teams introduce AI into processes where it adds little value. Outputs are generated because they can be — not because they're needed. Work expands to justify the tool.

Usage goes up. The business doesn't.

This is not new. It is the same pattern seen in every major technology cycle. The tooling changes. The mistake doesn't.

The Only Questions That Matter

Every AI initiative should be able to answer three questions:

• What cost does this remove?
• What revenue does this create?
• What time does this compress — and how does that translate financially?

If it cannot answer one of these, it is not transformation. It is activity.

These are not complicated questions. They are uncomfortable ones. Because answering them requires tracing AI from the tool it lives in to the P&L line it should move — and most organizations have never built that connection.

What the Numbers Say

The evidence is now consistent across every major research source published this year.

McKinsey's latest Global Survey on AI found that 60% of organizations still have not seen enterprise-wide EBIT impact from their AI programs — and identifies the reason precisely: most companies are deploying AI in ways that are more visible than valuable. ³ Horizontal tools — copilots, chatbots, summarizers — help people work faster but, in McKinsey's words, "rarely change a P&L." ³

PwC published its AI Performance study on April 13, 2026, drawing on 1,217 senior executives across 25 sectors. The finding sharpens the picture further: 74% of AI's economic value is captured by just 20% of organizations — with top performers generating 7.2 times more AI-driven revenue and efficiency gains than the average competitor. ⁴

The other 80% are not failing at AI. They are succeeding at the wrong scoreboard.

WRITER's 2026 enterprise AI survey of 2,400 global executives closes the loop: 97% report AI has been beneficial — yet only 29% report meaningful organizational ROI. Nearly half describe their company's AI adoption as a massive disappointment. ⁵

97% say it's working. 29% can show it in the numbers. That gap is not a technology problem. It is a measurement problem.

The Signal

The companies that will define the AI era are not the ones with the most deployments.

KPMG's partnership restructuring was not a choice the firm made. It was a consequence the economics made for them. EY's investment in agentic audit infrastructure is a choice — a decision to direct the same pressure toward redesigning how audits work rather than waiting for the model to force restructuring.

Both paths lead to a different business. Only one of them is deliberate.

AI doesn't become real when it's deployed. It becomes real when it shows up in the numbers.

The question is not whether you're on the right scoreboard.

It's whether you know which one you're playing on.