Most talent functions are now being asked what their AI plan is. The useful answer is not a tool selection. It is a view on which specific tasks in the hiring process are constrained by human throughput and which are constrained by judgment.
Throughput tasks are where the gains are real and measurable: scheduling, first-pass consistency checks against defined criteria, market research synthesis, drafting that a human then edits, and reporting that currently takes a day a week to assemble.
Judgment tasks are where the risk sits: assessment of potential, cultural and structural fit, and any decision a candidate could reasonably ask you to explain. In regulated environments the ability to explain a decision is not optional, and a model that cannot be interrogated cannot be the basis of one.
Three governance questions should be settled before any tool is deployed. What data goes into it and under which lawful basis. What decisions can it influence, and who reviews them. How is adverse impact monitored, and what happens when it is found.
The sequencing point matters more than the tool. If time in stage is eleven days because a panel cannot align diaries, AI-assisted screening will not help. Fix the measured constraint first, then automate what remains.
The organisations getting value from this are not the ones with the largest deployments. They are the ones that measured their process first, chose two tasks, applied the technology to those, and can now state the before and after figures.
Automating a broken process produces the same outcome faster, at greater cost and with less explanation available.
