AI for HR and recruitment, and where it must stop

What an AI hiring and HR team actually is

An AI hiring team is a group of AI workers that carry the administrative weight around employing people. Reading applications against written requirements and putting them in a sensible order. Finding an interview time across four calendars. Replying to the people you did not take. Building the contract, the equipment list and the first week for somebody who starts on Monday.

This is the department where the boundary matters more than the capability. Everything here touches somebody's livelihood, and the honest position is that the workers below do the paperwork and a person makes every decision about a human being. Not as a courtesy, and not as a legal precaution, but because a system that ranks people on patterns can quietly reproduce whatever bias was in the patterns.

The other half of this department has nothing to do with hiring. It is the constant drip of questions from people who already work for you: how much leave do I have left, how do I claim this expense, where is the form, what is the policy on working from home. Those are answerable from a written handbook, and answering them is not a good use of anybody's afternoon.

The team, and what each one does

Nine jobs. Read the limits attached to the first three carefully, because they are the point.

• The application sorter. Reads every application against the requirements you wrote down and groups them by which requirements they meet. It reports what it found and it never rejects anybody.
• The candidate replier. Acknowledges every application, tells people where they are in the process and lets them know when a role is filled. Silence is the most common complaint candidates have, and it is entirely fixable.
• The scheduler. Finds interview times across everybody's calendars, sends the invitation, sends the reminder and rearranges when somebody cannot make it.
• The interview pack builder. Prepares the same set of questions for every candidate for a role, tied to the written requirements, which is both fairer and more useful than whatever anybody thinks of in the room.
• The interview note-taker. Writes up what was said so the interviewer can listen instead of typing. Scoring stays with the people who were in the conversation.
• The onboarding builder. Produces the offer and contract from your own templates, the equipment list, the accounts to create and a first week that has actual content in it.
• The trainer. Turns your written procedures into training material and short checks, so a new starter learns the real process rather than whatever the nearest colleague remembers.
• The policy answerer. Answers leave, hours, expenses and benefits questions from your written handbook, and says plainly when something is not covered rather than guessing.
• The records clerk. Keeps documents in order, notices when a contract or a visa or a certification is coming up for renewal, and routes the approvals that go with them.

A Tuesday inside an AI hiring team

8:00am. Sixty applications arrived overnight for one role. The sorter has read all of them against the six written requirements and grouped them: meets all six, meets four or five, meets fewer. Nothing has been discarded and every application is still readable in full.

8:45am. You spend twenty minutes on the first group and ten on the second, because the second is where people who wrote a modest application sometimes sit. You pick eight. That decision is yours and the system recorded only what you chose, not a recommendation you followed.

9:30am. The replier writes to everyone. The eight get an interview invitation. The rest are told the role is progressing with other candidates and that their application was read. Every message goes out the same morning, which is unusual enough that people notice.

11:00am. The scheduler books all eight across three interviewers and two time zones, sends invitations and sets reminders. This normally takes somebody most of a day of messages.

1:00pm. The pack builder produces the interview questions. Same questions, same order, every candidate, drawn from the requirements. Your interviewers can still ask anything else they like.

2:30pm. A current employee asks how much annual leave she has left and whether she can carry it over. The policy answerer replies from the handbook and links the section. Nobody in management was interrupted.

4:00pm. Monday's new starter is handled. Contract drafted from your template and sent for a human check, laptop request raised, accounts listed, and a first week built from your written procedures rather than from an empty calendar.

4:45pm. The records clerk flags that two certifications expire next month. It is a small thing and it is the sort of small thing that becomes a large thing when nobody is watching.

What stays human in hiring and HR

Every decision about a person stays human. Who to interview, who to hire, who to promote, what to pay, and how to handle somebody who is struggling. These are not tasks with a correct answer waiting to be computed.

Rejection stays human, or at least stays with a person's name on it. Being turned down by a machine is a bad experience and people talk about it. A template a person approves and sends is fine. An automatic rejection based on a score is not.

Scoring against anything outside the written requirements stays out entirely. Where somebody studied, how their name reads, what their photograph looks like, how long a gap sits in their history. If it is not in the requirements you wrote, it is not a factor, and the requirements should be short enough to defend out loud.

Anything involving employment law stays with people who know your jurisdiction. Notice periods, contracts, dismissals and leave entitlements differ by country and change. A system holds your written policy. It does not know whether that policy is lawful.

How hiring connects to other departments

Hiring is where an operating system either compounds or stalls. Everything the operations department wrote down as a procedure becomes training material here, which means the written process pays for itself twice: once when work runs on it and again when somebody new learns from it.

The connection people underrate is the one to adoption. A business installs AI systems and then hires people who were never taught to use them, and within a year the new half of the team is working the old way.

• The operations team supplies the written procedures, which become both the training material and the checks a new starter works through.
• The finance team receives approved starters, leavers and changes as clean records rather than as a message somebody remembers to send.
• The support team is usually where new staff start, and its written answers double as the fastest onboarding material in the business.

What an HR system costs and what changes the number

An audit, then a build, then a monthly fee while it runs. The size of the build depends almost entirely on how much of your people process exists in writing. A business with a handbook, templates and defined roles is a short project. A business where terms are agreed verbally and vary by person has to settle those questions first, and that is not a task a supplier can complete for you.

Hiring volume is the other factor, and it cuts both ways. A business hiring continuously gets value from the sorter and the scheduler immediately. A business hiring twice a year gets more from the policy answerer, the trainer and the records clerk, which serve the people already employed.

• Whether a written handbook, contracts and role requirements already exist.
• How often you hire, and how many applications a role attracts.
• How many countries you employ people in, since the rules differ.
• Whether onboarding procedures are written down or carried in people's heads.
• Which existing system holds employee records, and whether it can be connected.
• Whether Wobble operates it monthly or hands it to your own team once the handover is done.

Hiring decisions this must never be allowed near

Automatic rejection. It is the one thing every buyer asks for and it is the thing to refuse. A pattern-matching system trained on who you hired before will favour people like the people you hired before, and the businesses that most need to widen their hiring are the ones this quietly narrows.

Anything scored on characteristics you did not put in the requirements. If you cannot say out loud why a factor matters for the job, it must not influence an ordering. The protection is that the requirements are written first, kept short, and applied the same way to everyone.

Video interview scoring on tone, expression or confidence. Those measure how comfortable somebody is on camera, which correlates with things that have nothing to do with whether they can do the work.

A business hiring two people a year does not need most of this. The sorter and the scheduler will sit idle, and installing them is buying a machine for a job you do twice. Start with the policy answerer and the trainer, which earn their place every week.

What it costs, the decisions it is barred from, and who holds the records

No price is published for this, and what it costs is set almost entirely by how many of your policies exist as documents rather than as custom. A company with a written leave policy, a probation process and an offer template is buying a short build. A company deciding each of those case by case is buying the work of writing them down, which is worth doing whether or not any software follows.

The barred list is short and absolute. It does not rank candidates, it does not reject anybody, it does not decide pay, and it does not form a view about a performance issue. Each of those waits for a person with a name, because a hiring decision taken by a system is a decision with nobody accountable behind it, and that is a legal position as much as an ethical one. Every shortlist is a queue for human approval rather than an outcome.

Interview scheduling, acknowledgement replies, document collection, reference chasing and building a new starter's first week run unattended, and they are where the time actually goes.

You own the candidate records and the personnel files, in your own accounts, under whatever retention rule you set. Ali owns marketing and sales at Wobble and has hired into the same market you are hiring in, which is why the candidate-facing writing gets his attention rather than a template's. Wobble works from Karachi across 25 engagements in six countries, each published on the work page, and stays accountable for the workflows it built.

Common questions

What does AI do for HR and recruitment?

It carries the admin: sorting applications against written requirements, replying to every candidate, scheduling interviews, preparing consistent questions, writing up notes, building onboarding, answering policy questions from your handbook and tracking document renewals. People make every decision about people.

Can AI screen CVs and reject candidates?

It can read applications and group them by which written requirements each one meets. It should not reject anybody. Automatic rejection removes a person from your process with no human ever having seen them, and a system that learned from your past hires will favour people resembling them.

How do you stop it being biased?

By writing the requirements down before applications open, keeping them short enough to defend out loud, and letting nothing outside that list influence the ordering. Every application stays readable in full, and a person picks. Bias is not removed by the tool, it is limited by the process around it.

Can it handle onboarding for a new starter?

Yes, and it is the part most businesses feel first. Contract from your template for a human check, equipment and accounts raised, and a first week built from your written procedures instead of a fortnight of shadowing whoever is least busy that day.

Will it answer staff questions about leave and pay?

It answers from your written handbook and says plainly when something is not covered rather than guessing. Anything about an individual's pay, performance or circumstances goes to a person. The policy answerer exists to stop simple questions interrupting managers, not to have difficult conversations.

Is this worth it for a small team?

Not the hiring half, if you hire twice a year. The parts that pay for a small business are the policy answerer, the trainer and the records clerk, because those serve the people already employed and they work every week rather than during a recruitment round.