Everyone Wants AI in HR. Few Are Asking What It Should Actually Do.
By IntraQ · · 9 min read

Why the next phase of HR technology should be measured by business outcomes, not AI features.
Walk through the exhibit hall of any major HR technology conference and the pattern is hard to miss. Recruiting platforms promise AI that screens and schedules. Onboarding tools generate personalized plans. Payroll systems flag anomalies. Employee service portals offer copilots that answer policy questions. A few years ago, AI was a differentiator. Today it is closer to an expectation.
That shift has created a new kind of pressure on HR executives. Boards ask about AI strategy. CEOs want to know what HR is doing with it. Vendors present roadmaps in which every module gains an assistant or an agent. Under those conditions, adopting AI can begin to feel like a goal in its own right, and the question of what the AI is supposed to accomplish can quietly fall away.
HR Has Entered Its AI Era
Much of this progress is real. Many HR technology vendors are doing serious engineering work and solving genuine problems for their customers.
At the same time, the market rewards visible AI functionality, and visible functionality is not the same thing as value. Gartner has a name for one version of this problem. In a June 2025 prediction that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing rising costs and unclear business value, the firm also described "agent washing," the relabeling of existing assistants and chatbots as agentic without significant new capability. The issue is not that AI in HR is overstated across the board. It is that buyers increasingly need a way to separate meaningful capability from labeling.
The Wrong Question: Does It Have AI?
Most HR technology evaluations follow a familiar path. A vendor demonstration showcases a polished assistant answering questions in real time. A feature comparison spreadsheet adds a row for AI capabilities. Procurement requirements ask whether the platform includes generative AI and what is on the roadmap. Each step treats the presence of AI as evidence of value.
A better question is harder to answer in a demonstration: What becomes measurably better because this AI capability exists?
The difference shows up quickly in practice. An assistant that drafts job descriptions is impressive in a demo, but if the organization's hiring delays come from slow approvals and scheduling, faster drafting changes little. By contrast, a less glamorous capability that reliably reduces the number of unresolved employee data exceptions each month may create far more operational value while drawing little attention on a sales call.
The evidence on enterprise AI value supports this caution. McKinsey's State of AI 2025 survey of nearly 2,000 respondents found that 88 percent of organizations reported regular AI use in at least one business function, yet only 39 percent reported any EBIT impact at the enterprise level, and most of those attributed less than 5 percent of EBIT to AI. Adoption has become widespread. Measurable business impact remains much less common.
Automation Is Useful. Intelligence Is Different.
Part of the confusion comes from treating different kinds of AI as interchangeable. Generative AI produces and transforms content, such as drafting communications, summarizing documents, and answering questions in natural language. Workflow automation executes known processes when defined triggers occur, such as routing an approval or assigning onboarding tasks. Predictive analytics identifies patterns in historical data to estimate future outcomes, such as attrition risk or hiring demand. AI agents combine these capabilities to pursue multi-step tasks with some degree of independence.
Operational intelligence addresses a different problem. It focuses on understanding what is happening across an organization's systems, recognizing which changes matter, and identifying the evidence needed to act on them. Automating an onboarding checklist reduces administrative effort, and that is valuable. Determining whether a workforce change creates a new obligation is a different kind of work. It may require connecting information across several systems, interpreting organizational context, and locating supporting records.
Neither approach is superior. They solve different problems and often complement each other. Automation is most effective when the organization already knows what needs to happen. Intelligence is most useful when the organization needs help recognizing what needs to happen in the first place.
The Hidden Cost of Feature-First AI Adoption
When AI is introduced without a clearly defined operational objective, costs tend to appear in places that feature comparisons do not measure. Several overlapping assistants may emerge across the HRIS, the service desk, and the learning platform, each answering similar questions from slightly different data. Ownership becomes unclear when no one is accountable for whether an AI output is correct. Inconsistent underlying data produces inconsistent answers. Productivity claims go unmeasured because no baseline was established.
Governance adds another layer. Every AI capability that touches employee data introduces questions about access, accuracy, and oversight. NIST's AI Risk Management Framework treats governance as a continuous organizational function rather than a one-time approval, which means each new AI feature carries ongoing responsibilities.
Gartner identified similar factors in a July 2024 prediction that at least 30 percent of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs, and unclear business value. None of these outcomes is inevitable. But they share a common root. AI was introduced before anyone defined what it needed to improve, so there is no clear way to recognize success or failure.
Perhaps the most subtle cost is automating an inefficient process. If an approval workflow has unnecessary steps, AI can make those steps faster without questioning whether they should exist. McKinsey's research points in the same direction: the small group of organizations reporting the most significant value from AI were nearly three times as likely as others to have fundamentally redesigned their workflows. Value tends to come from changing how work is done, not from layering AI onto the way it was already done.
What Should AI Actually Do for HR?
A more disciplined approach starts with a short set of questions that HR leaders can apply to any AI opportunity, whether it comes from an existing vendor, a new platform, or an internal initiative.
1. What problem are we solving? The starting point is a specific source of operational friction, risk, delay, or decision difficulty, described in the organization's own terms. "We want AI in employee service" is not a problem statement. "Employees wait an average of three days for answers to routine benefits questions" is.
2. What information does the AI need? Every AI capability depends on data, and HR data is distributed across many systems with uneven quality and strict permission requirements. Before committing, leaders should understand which systems the AI must access, whether that data is reliable and current, and what integration and access controls are required.
3. What should improve? Define the measurable outcome before deployment: reduced processing time, improved accuracy, fewer unresolved exceptions, faster recognition of meaningful workforce changes, or better employee service. Establishing a baseline is what turns a productivity claim into a productivity measurement.
4. How will we know the AI is correct? Fluent output is not the same as accurate output. Organizations should know what evidence supports an AI conclusion, how uncertainty is represented, how performance is monitored over time, and where human review is required. The distinction between an answer and its evidence matters most when conclusions affect employees.
5. Who remains accountable? AI can investigate, draft, and recommend, but someone must own the decision. Clear decision authority, escalation paths, and human oversight should be defined before deployment, not discovered after an error.
These questions do not slow AI adoption. They direct it.
A Practical Example: Two Different Approaches to AI
Consider a hypothetical 1,200-person organization operating in six states. It runs an HRIS, a separate payroll platform, a document repository for policies and acknowledgments, and an employee service desk. Leadership has asked HR to show progress on AI this year.
Approach A: Feature-first adoption. The organization deploys an AI chatbot that answers HR questions and summarizes policies. The investment delivers real value. Employees find answers faster, and service desk staff spend less time on routine inquiries. But the chatbot only responds when someone asks. When an employee transfers to a different state and the HRIS reflects the change while payroll still shows the previous work location, nothing prompts anyone to ask about it. Depending on the jurisdiction, that mismatch could affect withholding, required notices, or which policies apply. The chatbot cannot help with a question no one knows to ask.
Approach B: Problem-first adoption. Before selecting a tool, HR identifies a specific challenge. Important workforce changes are recorded across systems, but their implications are not consistently recognized, and discrepancies between systems are often found months later. HR defines success in measurable terms, such as the time between a change and its recognition, the number of unresolved discrepancies, and the share of findings supported by evidence. It then evaluates an intelligence-oriented approach that can investigate relevant changes, connect information across authorized systems, identify unresolved questions, and route findings to the right people for decision.
These approaches are complementary. The chatbot remains valuable for employee service. The difference is that in Approach B, the AI investment follows a defined operational need, and the organization can tell whether it worked. As explored in The End of Reactive HR, recognizing what matters sooner is a different capability from responding faster when asked.
The Next Competitive Advantage Is Organizational Understanding
Systems of record will remain essential. The HRIS, payroll, ATS, and document platforms hold the authoritative data that HR depends on, and nothing about the AI era changes that. What is changing is the opportunity to interpret information across those systems, which most organizations still do manually. Having more data does not automatically produce better decisions. Connecting it does.
IntraQ is being built as an AI-powered HR operations intelligence layer with that purpose: helping organizations investigate workforce changes, connect relevant information, and support evidence-informed human decisions. Its architecture follows a deliberate sequence: Investigate → Draft → Approve → Evidence. The design emphasizes read-only integrations with existing systems, human approval before consequential action, and traceable evidence behind every finding. These are the principles guiding the platform's development, and its capabilities continue to mature against them. It is not an HRIS, ATS, or payroll replacement.
Stop Asking Whether HR Needs AI
The question of whether HR needs AI has largely been settled by the market. The more important question is what AI should make possible that the organization cannot reliably accomplish today. For some organizations, the answer will be faster employee service. For others, it will be better hiring decisions, fewer manual reconciliations, or earlier recognition of consequential workforce changes. The answer should come from the organization's operating problems, not from a vendor's feature list.
AI deserves to be evaluated as a means of improving organizational performance, not as an achievement in itself.
The future of HR technology will not belong to organizations that deploy the most AI. It will belong to those that understand what their AI is actually accomplishing. To explore how a problem-first approach to HR operations intelligence is taking shape, visit IntraQ.