How to automate the hiring process, and which steps you should not

An automated hiring process is one where the repeated work happens without anybody starting it. "Hiring automation software", "automated recruitment software" and "recruitment automation software" name the same shopping trip. The disagreement is over how much of the process a vendor claims to take.

Most guides answer that the same way: all of it. Pipefy's step-by-step guide, opened on 30 September 2026, automates sourcing, resume screening, scheduling, background checks and onboarding, while warning you to "do not attempt to automate 100% of the process; the human element remains vital for assessing cultural fit and final decision-making".

CareerPlug's guide, opened the same day, begins with "You can't automate every part of the hiring process", then tells you to automate "Pre-screen questions on your application" and a "polite rejection email" to whoever those rule out.

Both know judgement should stay with a person. Both then put the one irreversible judgement, telling somebody no, on the automated side, because it is the easiest thing on the list to automate. Easy and safe are different tests.

Swipe the table sideways to see every column.

Four, and they are the four that are pure repetition. None needs a judgement made, only the same work done again.

• Inviting every applicant. The step most often lost, and lost for the dullest reason: somebody has to send it. Software sends it the minute an application arrives, to applicant one and applicant 140 alike.
• Holding the first conversation. A first round is the same handful of questions asked of everyone, with follow-ups when an answer is thin. That is a procedure, not a verdict, and a procedure is what runs unsupervised.
• Ranking the answers. A ranking is a reading order. It tells you whose report to open first, and it costs nothing when it is wrong, as long as it is only an order.
• Writing the report. Per-question notes, the quote behind each one, a transcript. Writing up twenty conversations is the part a tired screener skips first and the part you need later.

What the four share is that each gathers evidence. That is the line: gathering the same evidence from a hundred people is repetition, done identically at applicant one hundred. Judging it is your team's job.

• The decision. Who moves forward, who does not, who gets the job. Each is a judgement somebody has to be able to stand behind afterwards.
• The offer. Money, level, start date, and the conversation where a good candidate tells you what it would take. None of that is repetitive.
• The edge case. The applicant from a different industry, the career changer, the person whose last two years are a gap with a reason behind it. Automation handles the middle of a distribution and mishandles the ends.
• The candidate who writes to explain something. Every automated pipeline has an inbox attached, and the inbox is where it fails. An acknowledgement template sent to somebody explaining a visa timeline is worse than no automation.
• The role where the work sample is the real test. For a designer, a writer or an engineer, a short paid piece of work tells you more than any conversation. Automate the conversation, then ask for the work anyway.

A tool that offers to automate one of those five is not a stronger product. It is the same product with a safety rail off.

Three things, and none shows up in the first month, which is what makes automatic rejection such an attractive mistake.

You lose the ability to explain a rejection

When a person declines a candidate there is a reason and somebody can say it.

When a threshold declines a candidate, nobody read the application, so nobody can answer the email asking why, or the lawyer asking it in harder words. A score is not an explanation. It is the absence of one.

You never find out where your ranking was wrong

A ranking is only correctable if somebody sometimes disagrees with it and hires anyway. Do that a few times and you learn which signals your process undervalues.

Automate the cut instead and everyone it was wrong about leaves quietly, and no correction comes back. The process looks flawless because its errors were deleted before anybody saw them.

You hand a judgement to one conversation on one day

An eight-minute interview catches somebody on a Tuesday, possibly on a poor connection, possibly a week after something happened in their life. As evidence that is useful. As a verdict it is far too thin, and anyone selling it as one is selling the part they cannot deliver.

Here is the arithmetic for one role with 120 applicants. Every number is an assumption, not a measurement. Change them to yours and the total changes.

Assume a first conversation of 20 minutes plus 8 minutes of arranging it: messages, a reschedule, the occasional no-show. That is 28 minutes each, so speaking to all 120 is 3,360 minutes, or 56 hours. Nobody has 56 hours, so in practice about 30 get one, which is 14 hours, and 90 people are never asked a question.

Now hold the same interviews unattended. Each still takes about eight minutes of the candidate's time, so nothing got faster.

What changed is that your presence is no longer the bottleneck, so they happen at the same time instead of one after another. Your time becomes reading: five minutes a report for the top 20 is 100 minutes.

Swipe the table sideways to see every column.

Now argue with the assumptions, because that is where your answer is. At 40 applicants and a 10 minute screen, by hand is under seven hours and you should just do it.

The arithmetic decides whether you need any of this, not the category. For a team with no recruiter, the buying decision is in AI screening for small teams.

1. Put the genuine must-haves on the form as yes or no facts: right to work, location, a licence the role legally requires. Facts, never quality. This is the only place a rule may exclude anybody.
2. Automate the invitation. Everybody who applies gets the same next step, immediately, with no order-of-arrival advantage.
3. Automate the first conversation, structured and recorded, so the twenty you compare later were asked the same things.
4. Automate the write-up and the ranking, and treat the ranking as a reading order and nothing more.
5. Stop there and read. The shortlist is where a person starts working.
6. Automate scheduling and sourcing only if that is your bottleneck. For most teams with one open role it is not, and both are separate products.

The shape this automates is the candidate screening process, and the case against is the problems with AI hiring. The step-by-step version for a pile of two hundred is how to screen 200+ applicants without a recruiter. If what you need is somewhere to track people rather than a way to interview them, that is an ATS, not an AI interviewer.

Four things, the same four, and nothing else.

• The invitation. Applying starts the interview. Every applicant is invited automatically while your monthly allowance lasts, and your live jobs pause when it runs out rather than quietly skipping people.
• The interview. The AI recruiter holds a spoken interview in the browser, about eight minutes and never more than ten, in English, drawn from your role and the applicant's resume, following up when an answer is thin.
• The ranking. A match score, an overall score, and a recommendation from strong hire to no hire.
• The report. Scored sections with reasoning, a ledger of which claimed skills were demonstrated and the quote that shows it, risks, and questions for your own round, plus the transcript and recording. A complete sample report is there to read first.

Here is the list that matters more, because it is the one this category leaves out. Career1 is not an applicant tracking system and does not sync with one. It does not source candidates and does not schedule anything.

It has no reject action: the AI ranks and recommends, a person moves an application forward or not, and that person's name is on it. It runs in English only, and publishes no bias audit and no accuracy figure, having measured neither. The pricing is a free plan of 2 interviews a month, then $49 for 25, $99 for 60 and $199 for 150, stopping there with no overage.

So if you want one system that automates the funnel end to end, this is not it. Sourcing and scheduling it does not do; the rest are the steps nobody should automate. The resume screening happens inside the interview, which asks about the claims on the document rather than scoring its wording.