Confident and Unready: HR's AI Is Outrunning the Data Underneath It

There is a number in this year's State of Data Integrity and AI Readiness study from Precisely and Drexel University's LeBow College of Business that should stop any HR leader mid-roadmap. Surveying more than 500 senior data and analytics leaders across U.S. and EMEA enterprises in late 2025, the researchers found 88% saying their organization has the data readiness AI requires. In the same instrument, 43% named data readiness as the most significant barrier to getting AI aligned with business objectives. The pattern repeats for infrastructure (87% confident, 42% blocked) and for skills (86% confident, 41% blocked).

That is not a survey error. It is what an organization looks like when the executive summary and the implementation team have stopped comparing notes. And HR is unusually exposed to it, because HR's data is the messiest data in the building: records assembled over decades by different people in different systems, with job titles, manager hierarchies, compensation history, and skills tags that were never standardized because nobody needed them to be.

Then the co-pilots arrived, and suddenly everybody needed them to be. "Confidence in AI does not automatically translate into ROI," Precisely Chief Data Officer Dave Shuman said in the study's release. In HR, it does not automatically translate into a defensible decision either.

Adoption Sprinted. The Review Process Walked.

The clearest picture of the gap comes from Traliant's 2026 AI Governance Gap research, conducted by Researchscape among 512 U.S. HR professionals at organizations of 100 to more than 1,000 employees between May 1 and May 15, 2026. Sixty-two percent of HR teams use AI regularly; 21% have it embedded directly in core workflows, and just 10% are still confined to pilots. Adoption, in other words, is settled.

Governance is not. Sixty-five percent of HR teams use AI for compliance and policy work (the highest-stakes category available), yet only 39% have established a formal process for reviewing AI output for bias, accuracy, or legal risk. Seventy-eight percent say output gets reviewed; barely half that number can point to a process that makes the review consistent. Fifty-one percent train employees on responsible AI use, and 45% extend AI literacy training to everyone. On the EU AI Act, 30% are aware it may apply and have prepared for it, while 18% are aware and have done nothing.

"Governance can't be an afterthought once AI is embedded in everyday workflows." – David Ashman, Chief Product and Technology Officer, Traliant

SHRM's State of AI in HR 2026 report, drawn from 1,722 completed responses collected December 5–23, 2025 through its Voice of Work panel, points the same direction. Forty-nine percent of organizations have an AI policy, but only 25% call it clear and future-proof and 54% say theirs is too restrictive to be useful. Fifty-seven percent of respondents were unaware of state-level AI regulations affecting their work, and 52% of organizations do not involve HR in setting overall AI strategy.

The failure mode here is not recklessness. It is a governance layer written for a pilot being asked to carry production traffic, in a function left out of the room where strategy got set.

Bad Inputs Do Not Announce Themselves

The reason data quality is the load-bearing problem here, rather than one item on a long list, is that AI failures on people data are quiet. A model that summarizes an employee's tenure from a record where the rehire date overwrote the original hire date does not throw an error. It produces a clean, confident, well-formatted answer that is wrong, and it produces it inside a promotion review, a retention risk score, or a reduction-in-force analysis where nobody is going to re-derive the underlying figure by hand.

Precisely's respondents put data quality at the top of the data integrity priority list, cited by more than half of leaders. The study also isolates what actually moves trust: organizations with a data strategy or formal governance program report high trust in their data 71% of the time, compared with 50% of those without one. That 21-point spread is the closest thing in this year's research to a lever. Sixty-three percent have established some form of AI governance, which means better than a third have none.

Employees have noticed the confidence gap even where they cannot see the data. BambooHR's State of the Workforce 2026 research, based on 1,248 respondents surveyed March 24 through April 9, 2026, found 81% of leaders reporting a productivity increase from AI while 49% simultaneously said AI has not delivered tangible value and is overhyped. Eighty-nine percent of employees want greater transparency. And 74% of leaders believe their people already have the AI skills they need, a belief that sits awkwardly beside Traliant's finding that fewer than half of organizations train everyone in AI literacy.

"The opportunity in front of organizations isn't to slow innovation down. It's to make innovation more human, more transparent, and more sustainable." – Brad Rencher, CEO, BambooHR

Transparency is the part that has a data prerequisite. You cannot explain how a recommendation was reached if you cannot say where its inputs came from, when they were last verified, or who is accountable for correcting them. That is a records question before it is an AI question.

What Data Readiness Actually Requires

None of this argues for a moratorium. The organizations reporting high data trust are not the ones that waited; they are the ones that built governance alongside deployment rather than after it.

• Audit the records before the model. Pick the three fields your AI use cases depend on most (usually hire date, manager, and job title) and measure completeness and duplication across every system that writes to them. You cannot govern outputs derived from inputs you have never counted.
• Write the review process down, not just the policy. Only 39% of HR teams have a formal process for checking AI output for bias, accuracy, and legal risk, even though 78% say review happens. The difference between those two numbers is what fails an audit.
• Get HR into the AI strategy conversation. With 52% of organizations excluding HR from overall AI strategy, per SHRM, the function that owns the data and the employment-law exposure is often the one not consulted about either one.
• Treat AI literacy as coverage, not a course. Forty-five percent of organizations extend AI literacy training to all employees. Partial coverage in a tool embedded in core workflows is the same as no coverage for whoever was skipped.
• Date-stamp your compliance posture. Thirty percent of HR teams are prepared for potential EU AI Act applicability and 57% of SHRM respondents were unaware of relevant state AI rules. Assign an owner and a review cadence rather than a one-time assessment.

The uncomfortable truth in this year's data is that nobody is going to be forced to confront a data foundation problem by their AI tools. The tools will keep answering. They will answer fluently, instantly, and with the same tone of certainty whether the underlying record was verified last quarter or last decade. Eighty-eight percent of data leaders believe they are ready; 43% know they aren't. The gap between those two numbers is exactly the size of the room where HR's next bad decision gets made: politely, in a well-formatted summary.