How the Social Discovery Group Talent team works ✾

Recruiting has an odd paradox built into it.

The more roles you have open at once, the less time you get to spend on actual recruiting. ⏳

Conversations with candidates get pushed aside by Slack, the ATS, spreadsheets, statuses, emails, notifications, and post-interview comments. You copy something here, paste it there, remind someone, then open the same candidate profile again to see whether the status has changed.

Each of these tasks takes minutes, not hours. But those minutes quietly add up to a big part of the working day.

At some point, we at Social Discovery Group started looking at this from a different angle: does a recruiter really have to do all of this by hand? 🤔

❌ We didn’t ask “Can AI replace recruiters?” That question never seemed very useful to us.

A much more practical question is this: which parts of a recruiter’s job need their experience and judgment, and which parts are manual only because that’s how the process happens to be set up?

That question gave us a very pragmatic approach to AI and automation. 🤖

First, kill the busywork. Then talk about AI 🎯

It’s easy for AI to become a goal in itself.

You plug in one more tool, write one more prompt, add one more integration. You end up with a beautiful diagram that nobody wants to maintain.

🙅‍♀️ That’s not how we work.

We keep automation down to earth. Every new workflow has to pass three tests:

• Does it save the team time?
• Does it make our ATS data better?
• Does it keep the candidate experience at least as good as before?

If the answer to any of these is “no,” we don’t count it as a success.

It also helps to separate AI from plain automation.

When the system creates a candidate profile, sends a notification, or moves a message from Slack into the ATS, that’s automation. 🔀

When you need to make sense of text, pull out the key points from an interview, draft an email, or tailor a message to a situation, that’s where AI comes in.

In practice, the two work well together. But they solve different problems.

Scenario #1. AI writes the first draft, but doesn’t get the final say.

Let’s start with the obvious one: writing. ✍️

We use ChatGPT and other AI tools to draft job descriptions, candidate emails, and posts.

The idea is simple. Instead of starting from a blank page every time, the recruiter gets a first draft to work from. 💡

There’s an important catch, though. We never treat AI-written text as ready to send.

A model is good at building structure, suggesting wording, cutting a long text down, or adapting it for an audience. But it doesn’t know every detail of the role. It wasn’t in the conversation with the candidate. And it isn’t accountable for what gets sent. 🤷

So the final check is the recruiter’s job. ✔️ Check the facts. Cut the corporate jargon. Delete the sentence that sounds like HR, marketing, and a big company’s press office wrote it together.

Above all, make sure the text matches the real role and the context of the conversation.

The model is simple: AI speeds up the draft, and a human owns the result. 🤝

Scenario #2. You can automate a rejection, but not the decision to reject.

Here’s a problem that seems to affect all of recruiting, not just our team. Candidates should get feedback on time, but in practice this is the step that most easily gets lost in the daily rush. ⏰

That’s especially true when lots of roles, interviews, and other processes are running at once. The decision about a candidate has been made, but the email may not go out right away.

This is still an area we’re working on. We’ve already built the automated communication flow and are now gradually rolling it out. Our goal is that no candidate goes without feedback just because a recruiter didn’t have time to send the message.

The logic is fairly simple. When a recruiter makes a decision and moves a candidate to the rejection stage, the system checks which step they reached: initial screening, HR interview, technical interview, or something else. 🔍AI then writes an email that fits that stage, and the candidate gets it automatically.

AI doesn’t make the rejection decision. That stays entirely with the recruiter. Automation only kicks in afterwards, to tell the candidate the outcome promptly and properly.

This split matters to us. The machine helps with communication, but it doesn’t replace a professional decision. ⚖️

There’s also a clear goal for the future. It’s not just about cutting a recruiter’s manual work. It’s about making the process more reliable, so that timely feedback becomes the standard instead of depending on how busy the team is.

Scenario #3. Managers keep writing in Slack, and the ATS still gets the information.

Here’s another small problem almost every hiring team knows.

Hiring managers often leave feedback on candidates right in Slack. That makes sense: they work there every day, so it’s easier to jot a few lines after an interview than to log into the ATS.💻

And that’s fine. Slack is open all day, and it’s easy to type a couple of sentences right after a call. It’s much easier than going into the ATS just to fill in a field.

But then the manual work begins. The recruiter has to take the message from Slack and paste it into the candidate’s profile.

We decided not to make managers change their habits.🤷‍♂️

Now they just add a specific emoji reaction to a Slack message, and its text is automatically copied into the ATS.

✅The manager keeps working where they’re comfortable.
✅The recruiter doesn’t have to copy and paste.
✅The candidate’s history stays in the main system.

In my view, workflows like this often do more practical good than any flashy AI demo. The technology solves a specific problem that comes up every single day.

Scenario #4. A referral shouldn’t turn into a manual scavenger hunt.

Our referral program is another good example of how much time simple but repetitive tasks can eat up.

The company runs both an internal and an external referral program. You can read more about how Social Discovery Group`s Referral external program works here. 📎

On paper, it’s simple: an employee recommends someone, a recruiter reaches out, and the candidate goes through the process.

In reality, there are lots of small steps in between. You have to assign a recruiter and notify them, create the candidate in the right pipeline, record the source and the referrer’s name, and then remember to keep the employee updated.

We used to do much of this by hand. ✋ Now it’s all one automated workflow:

An employee refers a candidate ➜ The system assigns a recruiter and notifies them in Slack ➜ The candidate automatically shows up in the right ATS pipeline, with the source and referrer’s name on their profile ➜ When the status changes, the employee gets an update.

The full history stays on the candidate’s profile.

It’s easier for employees too. They don’t have to message a recruiter to find out what’s happening with the person they referred. The status updates arrive automatically.

Interviewers’ internal ratings and comments, of course, stay inside the hiring process. 🔒 That’s a part of automation people sometimes forget: you need to automate not only the action, but also who can see what.

Scenario #5. AI Notetaker: not a “smart interviewer,” just a good note-taking assistant

We also use an AI Notetaker to record and transcribe interviews.

The obvious rule comes first: an interview is recorded only if the candidate agrees. ✅

After the call, the transcript is turned into a structured summary in our own format. It covers experience, relevant skills, motivation, strengths, questions and concerns, expectations, and next steps.

There’s one rule we care about a lot: if a topic wasn’t discussed, the model must not make it up. 🚫

No “filling in” a candidate’s motivation. No covering a blank field with a nice-sounding guess. If the information isn’t there, that should be visible.

The recruiter gets a solid starting point, checks it, adds their own observations, and adjusts the wording. ✔️

As a result, less time goes into piecing the conversation back together from notes, especially when the interview was an hour ago, and five other things have happened since.

Where we draw the line.

As much as we love automation, there are things we deliberately keep away from AI. For example:

❌the final assessment of a candidate;
❌the decision to hire or reject;
❌interpreting motivation and career expectations;
❌choosing between candidates;
❌reviewing sensitive or contentious wording.

Automation can prepare, move, remind, and organize. But the decision stays with a person. ✅ For us, this isn’t just a nice principle. It’s a working rule.

Why sometimes it’s better not to automate at all.

We’ve learned something else along the way: automating everything isn’t always a good idea.

Some processes have so many exceptions that the automated workflow becomes a maintenance project of its own. 🔧

You add a rule for one case. ➕ Then an exception for another. ➕ Then an exception to the exception.

At some point you realize it would have been easier to just do it by hand.

So being able to automate something isn’t a reason to automate it. If a manual check is faster, more reliable, and easier to follow, we leave it to a person.

So what does the candidate get? 👀

Candidates don’t need to know how many integrations are running behind the scenes.

For them, the benefits are much more down to earth, and that’s exactly why they matter more:

• They hear back faster
• The email matches the stage they actually reached
• Their interview is recorded only with their consent

And a person makes the decision about them, not an algorithm.

When automation works well, you barely notice it. ✨That might be the best test of all.

How we measure success

Our goal isn’t to show how many AI tools our recruiters now have.

What we care about is how many unnecessary steps have disappeared from the process:

• Less manual copying between systems.
• Less information stuck in Slack.
• Less switching between tools.
• Fewer cases where candidates wait for an answer just because the team is overloaded.
• Significant time saved (e.g., 16 hours a week or 768 hours a year across 16 recruiters on referrals).

The team feels the biggest difference at peak times, when lots of roles are open at once, and there are too many processes to comfortably keep in your head. 💬

What we took away from all this

First: automate the action, not the decision. ⚙️

AI can draft an email or organize interview notes. But the professional assessment of a candidate stays with the recruiter.

Second: information needs one home. 🏠

For us, that’s the ATS. If automation leaves you with yet another chat, spreadsheet, or file to search through, you’ve probably solved one problem and created another.

Third: candidates shouldn’t pay for your internal complexity. 🎯

They don’t care how the team syncs Slack with the ATS. They just want a clear answer and to know where their application stands.

That’s probably the real test of good automation. Not how impressive the technology looks, but whether it made things a little easier for the people on both sides of the process.

What’s Next 🚀

We’re not planning to automate recruitment entirely. To be frank, we don’t see the point.

Instead, we keep breaking the process into small pieces and asking one question about each:

Does this really need a human, or is it just an action we can hand over to the system safely and transparently? 🤔

If it’s the second ➜ we automate it.
If it’s the first ➜ we leave it to a person.

After all, a recruiter’s job isn’t to update ATS statuses as fast as possible. It’s to talk to people, understand context, work with hiring managers, and make decisions where an algorithm alone isn’t enough.

Everything else, where possible, can be lifted off their plate. 🙌

Written by Maria Khandros, Head of Recruitment, Social Discovery Group