Hiring eats an enormous amount of an owner’s time, and most of that time goes to work that is not judgment. Writing the posting. Reading 200 resumes to find 15 worth a call. Composing the same rejection email repeatedly. Trying to remember what the second candidate said three days later.
AI handles all of that. It should not handle the decision, and in several jurisdictions letting it try creates obligations most firms have not looked at.
Start with the criteria, not the resumes
The step everyone skips is the one that makes the rest work. Before a single resume arrives, write down what actually predicts success in this role at your firm.
Not “detail oriented.” Something checkable. For a paralegal in a high-volume practice: has handled more than 50 active matters simultaneously, has used our case management system or a comparable one, writes clearly enough to send a client email without editing.
This list is doing double duty. It makes AI screening possible, because you can only sort against stated criteria. And it forces you to decide what you are hiring for, which is the part firms get wrong long before any tool is involved.
Where AI earns its place
The job posting. Give it your criteria, the pay range, and a description of the daily work, and ask for a posting that is specific rather than generic. Include the pay range. Postings without one get more applications and worse ones, and several states require it anyway.
Resume screening against your criteria. Feed it the criteria and the resumes and ask for a ranking with one line of reasoning per candidate. The reasoning is the important output. A ranked list without reasons is a black box, and you cannot check a black box.
Then read the top group yourself, and read a sample from the middle. The middle is where you find the person whose experience is described in unusual words, which is exactly the candidate a keyword-shaped process discards.
Structured interview questions. Ask for questions that test each criterion behaviorally, plus what a good and a weak answer sounds like. Structured interviews predict performance better than unstructured ones, and the reason most firms do not use them is that writing them is tedious.
Interview notes. Record with consent, transcribe, and summarize against your criteria. This solves the real problem of comparing five candidates a week later from memory.
Rejection emails that sound like a person wrote them. Every rejected candidate is someone who might refer a client or apply again. The principles in AI email for lawyers apply: short, warm, specific, no filler.
The legal limits
This is where firms create problems for themselves, and a law firm getting an employment law question wrong internally is a bad look.
New York City requires bias audits and candidate notification for automated employment decision tools. Illinois regulates AI analysis of video interviews and requires notice and consent. Other jurisdictions have added or proposed requirements, and this area is moving quickly.
The line that matters in nearly every framework is between a tool that assists a human decision and a tool that makes the decision. Ranking resumes for a human to read sits on the safe side. Automatically rejecting candidates below a score does not.
Avoid anything analyzing faces, voice, or inferred personality traits. That category carries the most exposure and the weakest evidence that it works.
And keep records. What criteria you used, what the tool produced, what the human decided. If a decision is ever questioned, contemporaneous records are your answer.
What to keep entirely human
The decision. Reference calls, where tone tells you more than words. Any conversation about pay. And the judgment call on a candidate whose background is unusual, which is precisely where a scoring system is least reliable and a person is most valuable.
After the hire
The hiring win compounds if onboarding is ready. A new paralegal who can ask your documented procedures a question in plain language reaches productivity much faster than one who learns by interrupting a senior person. That is the payoff described in AI SOPs for law firms, and hiring season is when it pays off most visibly.
Do this today
Take the role you hire most often and write down five checkable criteria that predict success in it. Specific enough that another person could apply them to a resume and reach the same answer you would.
That list is the input for everything above, and most firms have never written it down, which is why their hiring feels like guessing.