Enterprise AI got better, but the ROI didn't

2026: Broader adoption, stubborn economics

Fast forward to today. The models are undeniably better, deployment is broader and AI agents are being rolled out at scale.

McKinsey's global survey of 1,719 respondents across 97 countries shows that 44% of organizations reported scaling AI across the enterprise, up from 38% a year ago. Employees are undeniably feeling the lift: 80% said AI improves their individual productivity, while 50% said it helps them make better decisions.

Yet, the balance sheet tells a much colder story:

• EBIT impact is stuck in neutral. Only 37% of leaders can attribute any positive EBIT (earnings before interest and taxes) effect to AI, a statistic that hasn't moved since last year.
• The high-performer ceiling hasn't budged. Just 6% of organizations qualify as true AI "high performers," meaning AI accounts for at least 5% of their EBIT. That number is flat compared to 2025.

Granted, NANDA's 5% and McKinsey's 6% aren't direct apples-to-apples numbers. The studies used different methodologies. But the pattern is crystal clear: Adoption is sprinting ahead, while bottom-line financial return is barely crawling.

The bottleneck has moved from answering to operating

The key finding in the latest data isn't McKinsey's 6% high-performer ceiling. It is what that 6% does differently.

Nearly three-quarters of AI high performers fundamentally redesign their business workflows around AI, compared with only about one-quarter of other respondents.

Consider a standard enterprise request: Which completed jobs haven't been invoiced?

A modern LLM can understand that sentence. Executing it across an enterprise is not that easy.

The system must be able to do the following:

• Identify completed work in a field-service application.
• Match those jobs to accounts in a CRM.
• Cross-reference accounting records.
• Reconcile inconsistent customer IDs.
• Handle exceptions.
• Verify permissions.
• Prepare the transaction.
• Create an auditable record of the write-back.

A smarter model improves reasoning. It doesn't automatically reconcile disconnected systems of record, resolve missing customer data, handle API failures, enforce permissions or safely recover when a system drops halfway through a workflow.

That is the difference between AI that answers and AI that operates.

The CIO playbook: Closing the execution gap

To move from productivity gains to enterprise ROI, technology leaders must shift their focus from simply buying smarter models to redesigning how AI performs work.

1. Move beyond rigid pipelines to governed orchestration

Linear API chains break when data is incomplete, systems fail or exceptions occur. Use orchestration that maintains workflow state, supports retries and fallback paths, and lets AI reason through exceptions while keeping execution bound by deterministic rules.

2. Unify the primary data and context layer

Autonomous execution breaks down when core systems use conflicting names, IDs and definitions. Establish consistent customer, product, job and billing identities, or a reliable reconciliation layer, so AI can determine that records across CRM, ERP, finance and operational systems represent the same business entity.

3. Establish action-level governance and write-back guardrails

Moving beyond read-only AI requires fine-grained execution policies, identity-aware authorization, scoped and short-lived credentials, and human approval for high-risk transactions. Every write-back should be attributable, auditable and reversible where possible.

4. Target cross-functional hand-offs first

Look for processes spanning multiple disconnected applications where employees still copy, reconcile, re-enter or verify information manually. These handoffs are where AI can move beyond individual productivity to remove operational friction.

5. Measure the process, not the AI

Stop defining success by active users, prompts or query volume. Measure the business process, including the following:

• Reduced days sales outstanding
• Shorter order-to-cash cycles.
• Lower cost per transaction
• Fewer exceptions.
• Faster resolution time.
• Greater throughput per employee.

If the metric cannot eventually connect to revenue, cost, cash, risk or capacity, it is difficult to call it enterprise ROI.

The bottom line

Enterprise AI clearly got better over the past year. Adoption increased. Agents scaled. Employees became more productive. What didn't materially improve was the share of organizations capturing significant financial value.

The lesson for 2026 is increasingly clear: High performers aren't winning because they have access to secret models. They are much more likely to be doing the unglamorous work of redesigning workflows and connecting AI to enterprise execution.

The next enterprise AI advantage won't go to the companies that deploy the smartest chatbots. It will go to the CIOs who successfully bridge the gap between AI intelligence and enterprise execution.