Which Recruiting Tasks Can AI Phone Control Take Over First?

The daily repeat is really three kinds of work

I know a woman who does recruiting. She has four or five job platforms open at the same time.

She says the first thing she does when the day starts is posting the same job description on each of five platforms. The field names differ. The content is word for word the same.

Then comes the greeting. Each platform’s recommendation list has dozens of people on it. She opens them one by one, glances at the CV, and sends one line: “Hi, your background looks like a good match for this role. Do you have a moment to talk?”

The same sentence, sent dozens of times.

She put it this way once. The hardest part of this job is judging people. The part that takes the most time is the work before that judgement.

Below, the phone work in recruiting is taken apart, so you can see which parts are judgement and which parts are only moving information around.

Posting and moving covers publishing a job on several platforms, editing the job details, and changing the status.

Filtering and collating covers picking people out of the recommendation list, gathering CV details into a table, and preparing the invitation material.

Conversation and judgement covers the greeting, the interview invitation, the salary talk, and follow-up and rejection.

The third kind takes the most thought, and the volume is small. What eats the time is the first two, and those two barely need a judgement.

Filtering can be handed over, and bulk blasting is a poor fit

Bulk blasting does nothing useful, and the platforms explicitly restrict that kind of behaviour.

Send a hundred copies of the same sentence and the reply rate is low enough to ignore. Read twenty CVs carefully and send twenty messages written for those people, and the result is much better. Using a machine to pile up the number of greetings is the wrong direction from the start.

The painful part sits before the greeting: out of dozens or hundreds of recommendations, pick the few worth talking to.

That step only reads. It can be handed over:

Open the recommendation list on this platform. Pick out people who
updated their CV in the last week, whose target city is Hangzhou,
and who have 3 to 5 years of experience. Screenshot them and send
them to me.

After that step, fifty pages of CVs become a list of about ten people. The time goes from paging through to find who is worth a chat, to glancing at the list and deciding who to talk to first.

Candidate details are scattered through the messages on each platform, and the formats differ. Getting them into one table means copying by hand.

This part can go to a machine:

Turn the candidate messages on these three platforms into one table:
name, years of experience, target role, target city, source platform,
and how far the conversation has got. Append it to this sheet.

The screening conclusion stays with you. You decide who moves to the next round. That is the part of recruiting a person does. The machine gives you a tidy table. The judgement is yours.

For the mechanics of getting data from a screen into a table, see turning phone screenshots into a spreadsheet.

An invitation that goes out wrong costs the most. Time, place, the contact person, and what to bring: get any one of those wrong and the other person makes a wasted trip. You lose them, and they tell other people.

A workable split is that the machine prepares and a person sends:

Put this candidate's basic details, three available time slots,
and the interview location into one document.

Once the material is ready, you check it once and then send. One wrong send is a broken promise. The extra confirmation is at most one extra click.

What can be handed over completely is the notification part. A reminder to the candidate the day before the interview is a pure notice. It involves no judgement.

Follow-up needs layers.

Notification-type follow-up can be handed over: a reminder the day before an interview, and a batch nudge when there has been no reply a week after they applied.

Conversation-type follow-up depends on the case, and a person should do it: asking how they are leaning, and explaining why the process is slow.

Judgement-type follow-up stays with a person: salary, whether they will join, and reassurance after a rejection.

Anything that involves money or a promise stays with a person. A machine saying “salary is negotiable” and you saying “salary is negotiable” out loud carry different weight.

Keep accounts on separate phones, and hand over the read-only step first

A recruiting role naturally means accounts on several platforms, and sometimes more than one account on a single platform. Where those accounts sit is easy to overlook.

Switching logins back and forth on one device, over a short time, is itself unusual behaviour. Spread the accounts across several devices, and let each device handle its own few. When you need to handle them together, work by device group. Logging in and out on one machine is a poor arrangement.

For how to arrange the devices, see one person managing 20 phones. If you want a health check on the accounts first, see a daily health check for multiple accounts.

Each platform has its own rules on automated behaviour. The rules differ, and so does how hard they are enforced. Read the platform’s rules page before you start. Slower is the safer pace.

The order goes like this:

First, hand over filtering. Pick people from the recommendation list on hard criteria. It only reads, the risk is zero, and what you get back is the time spent paging.

Second, hand over collating. CV summaries and invitation material. Keep a confirmation by a person.

Third, messages. Only the notification kind: reminders and nudges. Persuasion stays with you.

Judging whether a job can be handed over still comes down to those three questions. Is the answer fixed? Is there a loss if it is done wrong? Does it need a judgement on the spot? Hand it over only when all three pass.

The software is completely free and runs on your own computer. For stringing several recruiting apps into one task, see how to automate a task that spans three apps.