What if the biggest constraint on AI-led HR transformation is not the capability of AI, but the way decisions around people are designed?
HR functions are under pressure to make decisions faster, more consistently and with better evidence. At the same time, those decisions are becoming more complex. Skills are changing, workforce models are evolving, employee expectations are rising and managers are grappling with more information than they can realistically process.
The natural response has been more technology. AI can search, summarize, recommend, predict and act. Agentic AI extends that capability further by moving from answering questions to coordinating and executing work.
Yet better technology does not automatically produce better decisions.
That is because HR decisions still operate at the intersection of people, information, process and judgement. Improving them therefore requires us to understand not only what AI can do, but how people make decisions in the first place.
This is where behavioural economics becomes particularly relevant.
The overlooked constraint in HR decision-making
Herbert Simon's concept of bounded rationality starts with a practical observation: people do not make decisions with unlimited information, unlimited time and unlimited cognitive capacity.
Instead, we make reasonable choices within the constraints around us.
Consider how that plays out across HR. A recruiter reviewing hundreds of applications may gravitate towards profiles that resemble previous successful hires. A manager preparing performance reviews may give disproportionate weight to recent events. During succession planning, highly visible employees can come to mind more readily than equally capable colleagues whose contributions are less visible.
These patterns are not necessarily signs of poor intent. They are often consequences of information overload, limited attention and the way choices are presented.
That changes the question HR leaders need to ask of AI.
Rather than simply asking, “What can we automate?”, a better starting point is:
“Which constraints are preventing better decisions, and how can technology reduce them?”
That distinction matters.
From process automation to decision architecture
For years, HR technology has primarily focused on digitizing transactions and automating workflows. AI introduces another possibility: improving the decision architecture around those workflows.
Suppose a manager is making a promotion recommendation. Giving that manager more dashboards may simply create more information to process.
A better-designed system could instead bring together relevant performance evidence, skills information, role requirements and historical context; highlight missing or contradictory data; surface alternatives that may otherwise be overlooked; and make the basis of its recommendation clear.
The manager still decides. But the conditions under which the decision is made improve.
The distinction becomes increasingly important as agentic systems enter enterprise workflows. Deloitte's HR technology outlook points towards increasingly integrated interaction between people and technology, while its broader work on agentic AI argues that organisations are unlikely to unlock its value simply by automating processes designed for humans. Work itself needs to be reconsidered.
For HR, that means moving beyond task automation to decision redesign.
Agentic AI can expand the decision space
Most first-generation HR assistants have been reactive. Someone asks a question and the system retrieves information or generates a response.
Agentic AI changes the operating model.
Agents can potentially work towards an objective, coordinate steps across connected systems and workflows, identify exceptions and escalate issues within defined boundaries. The industry is consequently examining how human-agent collaboration could reshape HR work itself. Gartner, for example, has published research specifically examining agentic AI across the enterprise and HR function, including human-agent partnership.
Consider a few practical applications:
- detecting stalled recruitment activity and bringing exceptions to a recruiter’s attention
- coordinating onboarding activities across HR, IT and workplace functions
- identifying payroll exceptions for review before they become employee issues
- bringing verified skills and role requirements together when identifying development options
- guiding managers through policy-intensive employee queries while escalating cases that require specialist judgement
The common thread is not autonomy for its own sake. It is the ability to reduce the search, coordination and information-processing burden surrounding a decision.
That is where agentic AI connects directly with bounded rationality.
It can widen the information available while narrowing what a person actually needs to examine.
More autonomy requires better judgement about where autonomy belongs
There is an important tension here.
The same recommendation designed to reduce cognitive load can create automation bias if users begin accepting it without proper scrutiny.
This is why human oversight cannot simply mean inserting an approval button at the end of an automated workflow.
HR leaders need to distinguish between decisions on the basis of impact, reversibility and risk.
An agent coordinating an onboarding checklist is fundamentally different from a system influencing who is shortlisted for a role. Resolving a routine policy query is different from recommending an employee for promotion or determining compensation.
As impact rises, so should the rigor surrounding:
- the evidence used by the system
- transparency of the recommendation
- authority given to the agent
- human review and override
- exception handling
- traceability and auditability
This is becoming more than a design preference. Regulation is placing scrutiny on AI used in employment, recruitment and worker management. Under the EU AI Act, a number of employment-related applications fall into the high-risk category, reinforcing the importance of human oversight, transparency and governance in HR AI deployments.
The implication for CHROs is clear: AI governance and HR operating-model design can no longer develop on separate tracks.
The business case needs to move beyond cost per interaction
There is another reason why decision design matters: economics.
The business case for AI in HR can easily become concentrated on transactional efficiency. How many employee queries can an assistant handle? How much manual effort can be removed? How quickly can a process be completed?
Those are useful measures, but they capture only part of the value.
A serious economic assessment also needs to consider the full cost of the operating model: technology consumption, integration, data readiness, governance, monitoring, security, adoption and the human oversight required to operate it responsibly.
The benefit side deserves the same discipline.
For HR leaders, value could include:
- fewer avoidable employee contacts and escalations
- shorter process cycle times
- improved first-time resolution
- reduced administrative follow-up
- capacity released for higher-value HR work
- more consistent access to policies and information
- better visibility of exceptions and decision risk
The real economic question is therefore not:
“Is an AI interaction cheaper than a human interaction?”
It is:
“Does the redesigned HR service produce a better outcome at an acceptable risk-adjusted cost?”
That is a much more useful business case.
Five questions before putting an agent into an HR decision
Before deploying AI into a people process, HR leaders can apply a simple decision-design test.
- Decision
- Constraint
- Evidence
- Autonomy
- Value and risk
What decision or outcome are we trying to improve, and who remains accountable for it?
What is making the current decision difficult: lack of information, limited time, cognitive load, fragmented processes or inconsistent execution?
What information should AI surface, and can users understand the basis and limitations of the recommendation?
What can the agent execute independently, what should it recommend, and what must be escalated for human judgement?
How will we measure efficiency and experience alongside fairness, adoption, exceptions and control effectiveness?
These questions shift the conversation from deploying AI features to designing a better HR system.
The next transformation of HR is a decision-design challenge
AI will undoubtedly change HR. Agentic systems will broaden what can be coordinated, analysed and executed without constant manual intervention. Current HR technology research already points to closer integration between human work and digital capabilities.
But the more consequential question is what organisations do with that capability.
Bounded rationality reminds us that human judgement will always occur within constraints. Behavioural economics helps identify where those constraints influence choices. Agentic AI creates new ways to reduce some of them. Good decision design determines whether that capability improves outcomes or simply accelerates an imperfect process.
For CHROs, this suggests a different ambition for AI: do not simply automate the work around HR decisions. Redesign the conditions in which good decisions are made.
That may ultimately be the more meaningful measure of intelligent HR.
How Infosys BPM can help
Infosys BPM helps enterprises transform HR operations by combining domain expertise, technology, AI, analytics and process improvement across the employee lifecycle. Our HRO services can help organizations streamline HR processes, enable better-informed decision-making, enhance employee experience, and strengthen compliance and operational efficiency, with a focus on measurable HR and business outcomes.
Ready to rethink how AI can support better HR decisions?
Connect with Infosys BPM to explore how we can help transform your HR operations.
