Companies getting AI ROI right are fixing their data first

23

TL;DR:

Agentic AI ROI increasingly depends on data architecture, not model sophistication. New research links composable systems with sharply higher returns, while broader evidence shows fragmented data and infrastructure remain major obstacles to scaling enterprise AI.

Article:

Companies seeing stronger returns from agentic AI may have a less glamorous advantage than better models: modern data architecture. As businesses move AI agents from experiments into operational workflows, reliable access to connected, traceable enterprise data is becoming a critical determinant of whether automation creates value or simply scales errors.

The 2026 Enterprise Technology Report from the MACH Alliance, based on 600 senior technology decision-makers across seven markets, found that 78% of fully composable organisations reported measurable AI ROI, compared with 13% of companies in the early planning stage. It also found that 94% said composable architecture accelerated AI deployment, while 98% of mature adopters believed their infrastructure could support AI at scale.

The mechanism is straightforward. Agentic AI needs unified customer, product, operational and transaction data, along with data lineage showing where information originated, how it changed and when it was updated. Fragmented systems deny agents that context and make questionable decisions harder to trace. The CMSWire analysis identifies three priorities: integrated data models, accessible lineage and stronger cross-functional data collaboration.

“The competitive advantage goes to enterprises whose infrastructure lets those capabilities coordinate and share context,” said MACH Alliance President Jason Cottrell.

Broader evidence tempers the enthusiasm. A 2025 survey of 1,854 executives found just 10% of organisations using agentic AI were already reporting significant measurable ROI, while one in four cited inadequate infrastructure and data as a barrier.

The implication for CIOs is practical: unify data definitions, expose lineage programmatically and connect systems before granting AI agents greater autonomy. Better context may deliver more value than another model upgrade.