Why the people decisions stay with a person
This job turns research into decisions about hiring, promotion, pay and training. A model can score a personality assessment in seconds. It cannot sit with a leadership team and defend a selection system when managers, lawyers and rejected candidates all push back on it.
Two tasks show the split clearly. Designing and validating selection tests is part statistics and part judgment: the math is routine, but choosing what a job really requires, and deciding which evidence of fairness is good enough, is a call someone has to own. Advising management on policy is the same shape. The analysis can be automated long before the conversation can.
It is also a small, senior field. The Bureau of Labor Statistics counts about 790 jobs in the United States, with median pay of $193,950 and employment projected to grow 6.5% from 2025 to 2035 (BLS, 2025). Most of that work sits inside consulting firms, large employers and government, much of it in professional services. Small fields move differently from big ones: fewer roles means fewer entry points, and junior analysis work is the first thing a tool absorbs.
What AI does, what it assists, and what it leaves alone
Start with the work AI can already carry. Scoring assessments at scale, cleaning survey responses into usable tables, and drafting first-pass summaries of engagement results are repeatable and well specified. Share of the task time in that group: 17%. The overall figure for how much of this job AI can handle today is the coverage score above; how coverage is measured explains what counts.
Then the assisted work. Analyzing job requirements and building training content are both faster with a model in the loop, and neither finishes without review. Someone has to check that the competency model matches the actual job, and that a training module does not quietly teach the wrong thing. Share of task time where AI helps rather than replaces: 50%.
Last, the work that stays with people: advising executives on organizational policy, coaching managers through a change they dislike, and standing behind a selection or promotion system when it is questioned. Share of task time in that group: 33%. These tasks are not hard because they are technical. They are hard because they need accountability, context and someone in the room.
How strong is the evidence?
Weaker than it should be. The evidence grade for this job is D, and a D grade means there is no direct, published test of AI against qualified industrial-organizational psychologists on their own tasks. Because of that, no parity number is given here: claiming one would be guessing.
What would settle it is specific. A blind comparison of model-built and psychologist-built selection systems, judged on validity and adverse impact. A test of AI-written survey interpretations against expert interpretations of the same data. Audited results from real hiring programs, not vendor case studies. Until something like that is published, the honest position is that the task-level reasoning stands and the head-to-head quality question is open. The scoring method shows how grades are assigned and what moves them.
When the picture could change
Most likely between 2037 and 2049 (8 in 10 of our scenarios). What that window measures is explained on the replacement-year page.
Two things could pull it earlier. First, there is no physical barrier at all: the robotics requirement for this job is none, so progress depends only on software and trust. Second, the cost gap is wide. Tool costs for the automatable parts run from roughly $80 to $7,840 a year, against $29,790 to $93,200 for the human hours those tasks replace. That gap is exactly the kind of pressure that thins out junior analytic roles first.
Two things hold it back. Employment law and professional standards put a named person behind selection and promotion decisions, and an unexplained model output is a liability rather than a shortcut. And adoption inside large employers is slow by design: assessment systems get reviewed, challenged and audited before they get rolled out. You can see both sides of this next to similar roles on the compare page.
How to stay needed in this job
Lean into the tasks in the needs-a-human group. Own policy advice to senior leaders, where the judgment and the responsibility sit together. Own the defense of a selection system: validity, fairness, documentation, and the conversation when a result is questioned. And own the coaching and change work, where the measure of success is whether people actually behave differently afterward.
Two skills matter alongside that. The first is auditing model-driven people decisions: knowing how to test an algorithmic screen for adverse impact and how to document what you found. The second is clear translation, turning messy data into a decision a non-technical executive can make and stand behind. The guide to AI skills employers want covers the practical end of that.
What to do: pick one assessment or survey process you run, automate the scoring and cleaning, and spend the time you free up on validation and stakeholder work.
If you are weighing a move, nearby work is worth a look: clinical and counseling psychologists, school psychologists and sociologists all share research training with different day-to-day tasks. The social scientists job family shows the wider set, and the list of jobs that mostly need a person shows what the task mix looks like further up the scale.