AI Adoption Is Widespread, but Scaling It Is Not

Enterprise AI has crossed the adoption threshold but remains far from universal scale. Nearly nine in ten organizations now report regular AI use in at least one business function, according to McKinsey’s latest global survey, while 44% say AI is scaling across the enterprise.

SAPinsider research shows a similar maturity gap among SAP customers, with adoption accelerating alongside SAP S/4HANA migration and cloud modernization while many organizations remain at early- to mid-stage maturity. The more meaningful distinction is increasingly how deeply AI is embedded into business operations and whether that use produces measurable outcomes.

Where Are Companies in AI Adoption Today?

Most companies use AI somewhere, but fewer than half have scaled it across the enterprise. Access to generative AI pushed experimentation into nearly every function, while embedding AI into core workflows has moved more slowly.

McKinsey’s 2026 survey puts enterprise-scale adoption at 44%, rising to 54% among companies with at least $1 billion in revenue compared with one-third of smaller organizations. The Stanford AI Index 2026 shows another side of that maturity gap: agentic deployment remains limited across many business functions even as general AI use becomes mainstream. Companies increasingly sit at different points on a spectrum from experimentation to deeply embedded operational use.

Why Do AI Pilots Fail to Scale?

AI pilots often fail to scale because production deployment introduces workflow, role, governance, and accountability questions that isolated experiments can avoid. A pilot can show that a model works without establishing how well it fits into everyday operations.

Deloitte’s 2026 pulse data illustrates the gap: 48% of respondents introduced AI without redesigning the workflows or roles around it, while only 12% report redesign at scale. McKinsey similarly found organizations that redesigned workflows early were 5.3 times more likely to report value creation. The scaling challenge tends to emerge when an experiment has to become part of repeatable, governed work.

Who Should Own Enterprise AI Adoption?

Enterprise AI adoption is commonly shared between business leaders and the technology teams responsible for delivery, data, security, and architecture. Business owners bring the process knowledge and accountability needed to identify where AI creates value, while IT and data teams provide the technical foundation.

Customer evidence from SAP Sapphire Madrid illustrates that model: Ericsson filtered use cases through business pain points and executive sponsorship before reaching 85,000 live users on Joule, while Fonterra brought business and process owners directly into use-case selection. McKinsey’s research on AI winners similarly identifies governance, talent, and decision rights as operating-model differentiators. In more mature deployments, ownership tends to extend beyond the teams implementing the technology.

How Should Companies Choose Which AI Use Cases to Scale?

Companies commonly narrow AI portfolios around use cases with a defined business problem, measurable potential value, and enough technical and organizational readiness to support production use. Generic model capability alone provides limited evidence that a use case will perform reliably inside a specific enterprise environment.

The Stanford AI Index shows rapid gains on agentic benchmarks, but benchmark performance does not establish how a system will handle a company’s pricing rules, approval paths, controls, or process exceptions.

Ericsson ranks use cases by impact and feasibility, while Fonterra brought business owners, process owners, SAP teams, and partners together to assess value and deployment readiness. More concentrated portfolios also make it easier to compare results across a manageable set of production use cases.

How Should Companies Measure Whether AI Adoption Is Creating Value?

Companies increasingly measure AI adoption through changes in business outcomes, reliability, and control alongside traditional usage measures. Licenses, users, and login counts show exposure, while measures such as cycle time, decision quality, output quality, errors, and cost per process provide a clearer view of operational impact.

Deloitte’s AI Pulse Check research finds organizations remain better at measuring AI costs than value. Companies with predeployment baselines for cost, quality, speed, or other business outcomes have a clearer reference point for judging whether later changes can reasonably be associated with AI.

What This Means for SAPinsiders

• Scale exposes weaknesses that pilots can hide. Small experiments can accommodate unclear ownership, fragmented data, and manual interventions. Those weaknesses become more consequential when AI moves into repeatable business operations.
• AI portfolios may require active pruning. Cheaper experimentation can leave companies with more pilots than they can operationalize. Mature portfolios are likely to concentrate resources around the use cases showing stronger evidence of value.
• Measurement increasingly begins before deployment. AI value is easier to assess when organizations already know the baseline for cost, quality, speed, or outcomes. Without that reference point, attribution becomes considerably harder.