Recruitment Automation: What to Automate First in Hiring

Hiring didn't get slower because recruiters got lazy. It got slower because the pile got bigger.

In May 2025, Greenhouse's chief product officer said recruiters have faced a 134% spike in applications per job since ChatGPT launched. By mid-2025, LinkedIn was taking in around 11,000 applications a minute. The team reading those applications is the same size it was before.

So everyone is reaching for automation. That part makes sense.

What doesn't make sense is where most companies start. They go straight for the AI that screens and ranks candidates, because reading applications looks like the most expensive step. It's also the only step in the hiring process that is currently in federal court.

There's a better order. It starts with the boring work nobody wants to do and stops right before the part that makes a decision about a person.

Follow one job opening and the problem shows up fast

Picture a 40-person company opening one role: a customer support lead. One person owns hiring, and they also have a real job to do.

The post goes live on Monday. By Friday there are a few hundred applications.

Now look at what eats that person's week. Reading applications is part of it. The bigger time sink is everything wrapped around the reading:

• Finding a slot that works for the candidate and two interviewers.

• Rescheduling when one of those interviewers gets pulled into something urgent.

• Answering "any update?" emails.

• Telling everyone who didn't make it that they didn't make it.

None of that is judgment. It's coordination. And coordination is where hiring time quietly disappears.

The data points the same way. Gem's 2025 recruiting benchmarks, built on more than 140 million applications, found the average time to hire rose from 33 days in 2021 to 41 days in 2024.

The decision on a candidate takes an afternoon. Most of those 41 days are waiting.

That waiting is what you automate first.

Start with the calendar, not the candidates

Interview scheduling is the least exciting item on this list. It's also the one I'd automate before anything else.

A 2026 recruiting statistics report cited in Guide's interview coordination benchmarks puts scheduling at roughly 35% of a recruiter's time. Guide is upfront that the number doesn't separate a quick phone screen from a five-person panel, so treat it as a rough signal. Even half of that would be a lot of hours spent playing calendar Tetris.

Back to our support lead role. Say you phone-screen 12 people and bring 4 back for a panel with two interviewers.

Done over email, every one of those is a thread with three to six messages in it. Done with a scheduling link, the candidate picks a slot that already works for everyone, gets a calendar invite automatically, and gets a reminder the day before.

You don't need an enterprise tool for this:

• Calendly's Standard plan is $10 per seat a month. Round-robin scheduling, which rotates phone screens across interviewers, starts on the $16 Teams plan.

• Most applicant tracking systems, including Greenhouse, Ashby and Workable, ship with candidate self-scheduling built in. Check what you already pay for before you buy anything new.

• Dedicated interview scheduling platforms like GoodTime price on the number of candidates you process a year rather than seats, and sell through quotes. That only starts to make sense at real hiring volume.

Some teams hand scheduling to a person instead, usually a part-time assistant. That works too, and I went through the trade-offs in what virtual assistant services actually include. For a single opening, though, a scheduling link does the job for the price of a lunch.

Why this goes first: it makes no decision about anyone. Every candidate benefits from it. And if you hate it, you switch it off tomorrow and nothing is broken.

Then fix the thing candidates actually complain about

Ask job seekers what they hate about hiring and AI screening isn't at the top of the list. Silence is.

In Greenhouse's 2024 job seeker research, 61% of candidates said they had been ghosted after an interview. On the employer side, Axios reported that only 4% of roughly 6,500 companies using Greenhouse contacted every rejected applicant during one quarter of 2024.

That isn't a technology problem. It's a "nobody has time to send 300 emails" problem. Which makes it a perfect automation problem.

For the support lead role, it means a handful of email templates tied to stage changes in your applicant tracking system:

• An application received email that says when they'll hear back, with a real date in it.

• A short note when someone moves to the next stage.

• Interview reminders with the time, the video link and who they're meeting.

• A respectful rejection email once a decision has been made.

Look at the wording on that last one. You're automating the message, not the decision.

A person still decides who is out. The software just makes sure the candidate hears about it instead of hearing nothing.

That line between automating a message and automating a decision is the most important idea in this whole piece. Keep it in mind for the next section, because that's where most companies cross it without noticing.

Parsing is fine. Ranking is where it gets risky

Once scheduling and updates run themselves, the pile is still a pile. This is where the temptation kicks in.

Vendors use "AI screening" to describe three very different things, and they carry very different levels of risk.

Parsing. Software reads each résumé and turns it into structured fields: job titles, dates, skills, education, location.

Your hiring person still reads and decides. They just do it from a clean table instead of 300 PDFs with 300 different layouts. Low risk, real time saved.

Knockout questions. A few objective, job-related questions on the application form. Are you authorized to work in the US? Can you work the 7am to 3pm shift? Do you hold the license this role legally requires?

If the answer is no, the application can stop there. This is defensible as long as every question is genuinely required for the job and you can explain why it's there.

Ranking and auto-rejection. A model scores each applicant for "fit" and drops the low scorers before a human ever looks.

On paper this saves the most time. It's also the version that has gone wrong in public.

Amazon found this out early. In 2018, Reuters reported that Amazon scrapped an experimental recruiting engine after discovering it penalized résumés that included the word "women's."

It had learned from ten years of mostly male hires. Nobody programmed it to do that. It picked it up on its own.

There's a newer problem on top of the old one. A growing share of applications are written with AI, so a ranking model increasingly scores how well a candidate's chatbot wrote a résumé, not how good the candidate is.

More automation on the reading side doesn't fix that. A person who reads 30 shortlisted applications properly does.

So for our support lead: parse everything, add two or three honest knockout questions, and let a human read what's left.

Why the screening step is the one in court

This is where it stops being a philosophical debate. Every major lawsuit and new law around hiring AI points at the same step: the moment software decides who moves forward.

The cleanest example is iTutorGroup. According to the EEOC, the company's application software was set up to automatically reject female applicants aged 55 or older and male applicants aged 60 or older.

More than 200 qualified people were turned away because of their age. The company settled in 2023 for $365,000.

The bigger case is Mobley v. Workday. Derek Mobley says he was rejected from more than 100 jobs at companies that screen applicants through Workday.

In 2024 the court let his claims move forward on the theory that Workday could be liable as an agent of the employers using its tools. In May 2025, his age discrimination claim was conditionally certified as a nationwide collective action.

Then in June 2026, the court largely denied Workday's latest motion to dismiss, letting California discrimination and disability claims proceed.

Read the agent part twice. The vendor isn't the only one exposed. If a tool rejects people on your behalf, the rejection is still yours.

The rulebook is filling in around this:

• New York City has required bias audits and candidate notices for automated employment decision tools since July 2023. A December 2025 audit by the New York State Comptroller called the city's enforcement ineffective. That kind of finding usually comes before tougher enforcement, not after.

• California's civil rights regulations on automated decision systems took effect on October 1, 2025. They apply the state's anti-discrimination law directly to these tools and require employers to keep related records for four years.

• Illinois amended its Human Rights Act so that, from January 1, 2026, using AI that discriminates in employment decisions is a civil rights violation, and employers have to tell candidates when AI is being used.

• Colorado replaced its original AI Act with a narrower law, SB 26-189, signed in May 2026. It takes effect on January 1, 2027, and hiring decisions are squarely in scope.

• The EU classes hiring AI as high-risk under the AI Act. The 2026 omnibus deal pushed those obligations to December 2, 2027. Delayed, not cancelled.

You don't need to memorize any of that. Just notice the pattern.

Not one of these rules cares about your scheduling link or your status emails. They all target the moment a machine decides about a person.

The 90-day order I'd follow

Here's how I'd roll this out, starting with that one support lead opening and keeping every step reversible.

Weeks 1 and 2: scheduling and reminders. Turn on self-scheduling for phone screens and panels, plus automatic reminders.

Before you switch it on, write down how many days it takes to get a candidate from application to first interview. That's your baseline.

Weeks 3 to 6: candidate communication. Set up the stage-based email templates from earlier. Rejection emails can go out in batches, but a person approves each batch before it sends.

If you send offer letters, this is a good moment to template those too. Contract automation covers what that kind of document automation handles well and where it still needs a human check.

Weeks 7 to 12: parsing and knockout questions. Turn on résumé parsing in your applicant tracking system and add two or three objective knockout questions.

Your hiring person still reads every application that passes. They just get through the pile faster.

Then measure three things: days from application to first interview, how many candidates drop out partway through, and how many hours a week your hiring person gets back. If those three moved in the right direction, the automation worked.

Somewhere around day 90, someone will suggest adding AI scoring to rank the pile. If you go there, go slowly:

• Ask the vendor for its most recent independent bias audit, and read it properly.

• Check selection rates by group. The long-standing rule of thumb from the federal Uniform Guidelines is the four-fifths rule: if one group is selected at less than 80% of the rate of the highest group, that's a red flag worth investigating.

• Keep a human review on every rejection the model recommends.

• Tell candidates you use it. In New York City and Illinois you have to anyway.

Automate the waiting, keep the judging

AI doesn't replace recruiting. It replaces specific tasks inside it, which is the same pattern that shows up across almost every job, as I wrote in will AI take your job.

In hiring, the tasks it replaces best are the ones nobody would ever put on a job description: chasing calendars, sending updates, retyping résumés into a system.

The tasks it handles worst are the ones that decide whether someone gets a shot at a job.

Give the first group to software. Keep the second group with a person whose name is on the decision.

If you're working out which other repetitive workflows in your business are worth automating, this is how we think about AI automation.

Questions people ask about hiring automation

Is it legal to use AI to screen résumés?

In most of the US there's no outright ban, but the employer stays responsible under federal anti-discrimination law no matter which tool made the call. Some places add their own rules: New York City requires bias audits and candidate notices, California applies its anti-discrimination law to automated decision systems, Illinois requires notice, and Colorado's new law starts on January 1, 2027.

What's the cheapest part of hiring to automate first?

Interview scheduling and automated status emails. A scheduling tool like Calendly starts at $10 per seat a month, and most applicant tracking systems already include self-scheduling and email templates you may not be using.

Does automation make the candidate experience worse?

Not if you automate communication. Most candidates complain about silence, not software. Automated updates and reminders usually make the process feel more human, because people hear back.

Should a small business use AI video interview scoring?

I'd skip it for now. It's the hardest kind of AI to explain to a candidate, and in Illinois the AI Video Interview Act already requires employers to give notice, explain how the AI works, get consent and delete the video on request.