Deeper AI Integration Drives Enterprise Returns, PYMNTS Study Reveals

Businesses have poured millions into acquiring AI models, copilots, and capabilities. The next investment dollar may deliver more value when spent fixing data quality, access controls, workflows, and organisational infrastructure than when buying yet another sophisticated model.

Research from PYMNTS Intelligence published in September revealed a significant gap between how extensively firms deploy AI and the value they capture from those deployments.

The 'AI at Work: Why Deeper Enterprise Use Produces Stronger Returns' report examined 75 distinct business tasks across multiple organisations. Firms with AI embedded in one or two functions applied it to an average of 41 tasks. Those with AI embedded in three or more functions used it across 40 tasks — essentially identical breadth.

The return profiles, however, diverged sharply. Among limited adopters, 55% reported generating returns from their AI investments. For deeper adopters operating across multiple functions, that figure jumped to 93%.

The data suggests AI value creation depends less on how many tasks receive AI treatment and more on how thoroughly organisations integrate AI into core business functions.

AI appears to be reaching a familiar technology inflection point where the constraint shifts from capability to infrastructure. The internet required data centres and logistics networks. Cloud computing demanded application migration and new security architectures. Smartphones spawned downstream markets in payments, identity, and mobile software.

Today's enterprises often lack the organisational design needed for capable AI systems to operate effectively inside them. This limitation becomes critical as AI transitions from answering questions to performing operational work.

An AI assistant drafting a memo can function atop messy organisational structures. An agent approving invoices, modifying customer accounts, initiating procurement workflows, or recommending multimillion-dollar capital movements cannot operate without clean underlying systems.

Companies with three or more embedded AI functions reported an average of 5.6 barriers to adoption, compared with just three among companies with no embedded AI. Organisations furthest along in AI deployment aren't encountering fewer problems — they're discovering more of their own organisational complexity.

Production AI exposes what pilot projects can ignore:

• Incompatible databases
• Fuzzy ownership structures
• Inconsistent policies
• Disconnected workflows
• Approval processes designed for humans manually transferring information between systems

The next wave of enterprise AI spending may appear in budgets as cybersecurity, identity management, cloud infrastructure, data governance, consulting, integration, or workflow software. Economically, however, much of it represents AI-enablement spending.

As access to powerful models continues commoditising, possessing AI itself provides limited competitive advantage. Competitors can purchase access to roughly equivalent intelligence.

What cannot be acquired overnight is the organisational architecture required to exploit that intelligence. This reality elevates clean data, interoperable systems, clear decision rights, and mature governance from corporate housekeeping to productive strategic assets.

The competitive question has shifted from which company has the smartest model — everyone can rent one — to which company can actually deploy it effectively in production environments.