Only a third of AI-active enterprises met their business case, HFS survey finds

Getting AI into use is not the same as getting the expected business result. Only 33% of 502 senior decision-makers at AI-active enterprises said their most significant program met or exceeded its business case. The HFS Research–Rackspace study, published November 12, 2026, puts responsibility for running deployed AI at the center of its proposed fix.

The finding is not a verdict that the remaining two-thirds got nothing from AI. HFS says 38% partially met their business case, while 11% said their program had not delivered. Those distinctions matter: falling short of an investment’s full justification is different from reporting no delivery, and the headline success figure should not erase partial gains.

The sample also sets clear boundaries. HFS surveyed respondents in August and September 2026 at companies with at least $100 million in annual revenue, across eight industries and four regions. Every company was running an AI pilot or something more advanced; 83% had AI in production or beyond. These are respondents’ assessments of active programs, not a measure of AI adoption across the whole economy.

Different stages, different results

The business-case result changes substantially with deployment stage. Companies using AI in one business function reported stronger success than those deploying it across several. The highest success rate belonged to organizations where AI was embedded in standard operations. HFS’s original report says just 24% of the surveyed enterprises had reached that embedded stage.

37%AI in one function

Respondents deploying AI within a single business function.

28%AI across several functions

Respondents deploying AI across multiple business functions.

56%AI embedded in operations

Respondents whose AI was embedded in standard operations.

These are different groups of companies, not the same companies tracked through successive stages. HFS explicitly cautions that the comparison does not establish that operating maturity causes stronger returns. Its explanation is that expansion exposes fragmented ownership: companies have bought models, platforms, infrastructure and services without assigning someone to run those pieces together.

The benefits respondents did report were concentrated in efficiency. Workforce productivity improvements appeared in 58% of responses, and measurable cost reductions in 42%. Customer and sales outcomes were less common: 28% reported better customer experience, while 21% reported measurable revenue growth. Productivity gains were therefore much more widely reported than revenue gains, even within this already AI-active sample.

The controls worry leaders more than the technology

Governance—the oversight and rules controlling AI use—was a near-universal concern. Significant gaps were reported by 96% of respondents, and 78% saw governance failures as a bigger threat to scaling AI than technology failures. Separately, 44% said their organization lacked a single, consistent approach to running AI in production.

HFS’s report gives that concern a concrete shape. Monitoring what autonomous AI does in production was the most-cited governance gap, at 55%. Meanwhile, 59% said regulators actively reviewed their AI use. The concern is therefore not simply whether a system produces useful output, but whether the organization can oversee what it does once deployed.

One operator does not have to mean outsourcing

HFS recommends assigning day-to-day responsibility for live AI performance, costs and governance to an operator. It calls the role a Services-as-Software Operator, but its definition allows three organizational forms:

• An internal team responsible for operating the combined system.
• Existing providers brought together under one remit.
• A managed-services partner coordinating the technology and partners already in place.

The recommendation aligns with the commercial positioning of study collaborator Rackspace, which describes itself as an operator spanning infrastructure, applications, data and AI. The survey was blind and named no vendor, platform or provider. That design separates respondents’ answers from a named supplier, although the proposed remedy also fits Rackspace’s business.

There is reported buyer appetite: 61% preferred one accountable partner across models, platforms and infrastructure. That preference rose to 74% among production-stage firms whose programs had not delivered. And 73% said they would pay a premium over raw infrastructure for fully managed AI with governance accountability. These are stated preferences, not completed purchases.

HFS says providers must earn that premium through clear operational responsibilities and measurable outcomes. The enterprise still retains policy, decision rights and legal accountability. Assigning an operator changes who runs AI day to day; it does not transfer the company’s ultimate responsibility for it.