Only One-Third Met Their AI Business Case. Read the Survey’s Fine Print

A headline that says “only one-third of enterprises realize AI value” sounds like a verdict on the whole market. The survey behind that claim is narrower—and still useful. It asked 502 senior AI decision-makers at large companies already running AI programs whether their most significant initiative had met its business case. One-third said yes.

HFS Research conducted the survey in partnership with Rackspace between August and September 2026. Respondents came from enterprises with at least $100 million in annual revenue across eight industries and four regions; all were running AI pilots or more advanced programs, and 83% were operating AI in production or beyond. The release says 33% had met or exceeded the business case for their most significant AI program. [^[1] , ^[2] ]

That result deserves attention, but it is not a representative estimate of every business’s AI experience. Companies that have not tried AI, smaller organizations, and firms without a senior AI decision-maker in the respondent pool are outside the sample. The study describes a group already far along the adoption curve—not a random slice of the economy.

The comparison groups are not a causal story

The release reports that 37% of organizations deploying AI in one function met or exceeded their business case, compared with 28% of those deploying across several functions. It also says 56% of organizations with AI embedded in standard operations had met or exceeded the business case. These are differences between respondent groups, not evidence that expanding AI causes returns to fall or that production maturity causes success. The organizations may differ in experience, investment, industry, project selection and how they define the business case. [^[1] ]

Other results point to operational questions worth asking. The release says 96% of respondents reported significant governance gaps and 44% lacked one consistent approach to running AI in production. Because Rackspace participated in the research and sells enterprise AI infrastructure and services, readers should distinguish the survey findings from the proposed solution: the release advocates a centralized operator role and managed services. That recommendation may suit some organizations; it is not a neutral conclusion proven by the survey alone.

This does not invalidate the research. It gives readers context for judging it. The methods note provides the field dates, respondent count, revenue threshold, geographic and industry coverage, and AI-adoption criteria. It also says the survey was conducted in partnership with Rackspace. Those details help bound what the numbers can support. [^[1] ]

Measure a project against its own promise

For an individual company, the right question is not whether its program matches a survey average. It is whether a defined workflow improved against a credible baseline. Before a pilot, write down the outcome the business case predicts: fewer minutes per case, fewer errors, faster resolution, lower cost per transaction, higher completion, or another measurable change.

Also record what can make the comparison misleading. Did the volume or complexity of cases change? Did staff receive training? Did another process change at the same time? Did reviewers handle only easy examples? A simple before-and-after measure can be useful, but it does not automatically isolate the effect of AI. A matched comparison or staged rollout may give a stronger signal when feasible.

Track quality and risk alongside speed and cost. A faster workflow that increases rework, privacy exposure, unsupported outputs or customer complaints may fail its business case even if usage rises. NIST’s AI Risk Management Framework organizes work around governing, mapping, measuring and managing risks; it is a resource for structuring the assessment, not a certification that a system is safe. [^[3] ]

The U.S. Government Accountability Office similarly frames AI accountability around governance, data, performance and monitoring. Those dimensions are useful reminders that “value” is not a single dashboard number: organizations need an owner for the process, a record of the data and system’s intended use, criteria for performance, and monitoring after deployment. [^[4] ]

Treat the survey as a prompt, not a forecast

If your organization is assessing an AI program, begin with one operational question and a baseline. Name an accountable owner who can see model behavior, the surrounding software, employee workflow, data and ongoing cost. Decide in advance what result would justify expanding, revising or stopping the pilot. Review the result with affected staff, not only the team that selected the tool.

The survey’s strongest lesson is not that every enterprise will miss its AI targets. It is that even AI-active companies often report gaps between a business case and realized outcomes. The next step is to make the local business case measurable and falsifiable. If a program is working, the evidence should show where it saves time or improves quality, what it costs to operate, and what risks need active management. Readers comparing model efficiency claims can also see our guide to cost per accepted task.