Is AI Really Changing Recruitment At All? A Year with LLMs…

You’re here for the controversial title, aren’t you. The answer is yes.

Yes, it is. But not always in the ways you’re thinking.

Over this year, we’ve plugged ourselves into a variety of AI tools and platforms - some have worked, some don’t work, some are getting expensive… so let’s have a look back at what’s ACTUALLY having an impact, and what’s likely going to happen next

It’s easier, and quicker, to find people

Sourcing is a time-consuming, not-too-fun task that you have to do. It’s nicer than putting out a job ad and saying “no” to every applicant from outside the country, because you have control over who goes into your longlist.

What AI lets you do now is upscale your sourcing skills - so you can say “I want someone who has scaled demand generation in a tightly regulated industry” and AI tools can go through the backgrounds of many candidates, quite quickly, and pick them out.

This sounds amazing, right? Hold your horses.

• Some of these tools use LinkedIn API which doesn’t give you everyone - because LinkedIn won’t allow it

• Naturally, you can’t run these queries over the whole dataset, because you’d need an ocean of water to cool the data centres just for that

• It still makes mistakes

• It still relies on crunching words from LinkedIn profiles, which even the most basic of sourcers knows is well, basic

We’ve got our own LLM, Manus, trained to headhunt using boolean search strings and OSINT techniques - for instance, it will go beyond LinkedIn results to find X (Twitter) accounts, a candidate’s footprint, blog articles, etc.

Actually, it’s easier to build a true picture of people more quickly

Yeah, this is the one. It’s easier to build up a picture of people. We were asked to find a leader who had not just platform experience and industry background, but who was either an athlete or a chessmaster. We found someone who was BOTH of those things, as well as a brilliant leader for a leading competitor.

We found him firstly through traditional, roll-yer-sleeves-up techniques, but the LLM built out the bigger picture for us.

And this made the outreach - and the subsequent interview we held with him - all the more interesting.

The problem with all this intensive research is the cost, however…

The AI costs are skyrocketing, actually…

A lot of recruitment companies are looking at replacing sourcers with AI because it’s more cost-efficient. Not because it’s better (it’s not, yet), but because of money. Fair enough. Capitalism. We all do it.

However - we still cannot replicate the handmade results that we get in LinkedIn’s Recruiter Pro solution. We can supplement it. We can get AI to evaluate our longlist and pull out the best candidates (in its opinion), and we can get AI to explore deeper into their backgrounds, cross-referencing with other open source techniques.

But the cost is accelerating. A Manus licence will set you back around $200 a month, and you can quite easily blow that in a week.

LinkedIn has introduced another level to its Recruiter licencing - and that’s upwards of £700 a month. It may well do some of what Manus is doing right now for us, but given the dreadful AI integration into their previous iterations, we won’t be upgrading any time soon.

So basically, the cost of AI is ramping up, and if everything is built on ChatGPT, Claude or whatever, you can expect all of them to ramp the prices up for the companies built on them, which has a knock-on effect.

We’ve probably seen the end of cheap AI.

But did we mention the admin?

It really wasn’t that long ago that we’d have to take our written notes, fire up the CRM, and fill in all of the fields such as current salary, expected salary, etc. etc. etc.

Of course we don’t do that any more. Now we’ve got a CRM that picks up on emails, InMails, video calls and every form of communication, and uses the AI to find that information and do it for us. And yes, it’s a huge time saver. The difference is that we can do 8 interviews in a day and the admin is done for us - so we can probably fit in a couple more than we used to.

In that way, recruitment is not much different from other industries where AI is simple doing some of the heavy lifting that would have taken us an hour, during which we’d have been pulling our hair out.

But do not use it for outreach

Now - as marketers - we’re quite proud of our outreach. We’ve worked on our messaging pretty hard, so that we often get up to 80% response rates from candidates regardless of whether they’re open to work or not. The benchmark is around 20%, btw.

We do that because we know what we’re doing, right?

We’ve looked at what LinkedIn’s AI messaging suggests, and we suggest back that it’s absolute pants. We also think that a lot of recruiters are using the AI suggestions from LinkedIn, which is why we get so many comments about how great our outreach is.

Game knows game, eh.

You cannot beat human interactions, and the human touch. There are some who believe that you can use AI recruiters to carry out screening interviews, which may or may not be a good idea - the jury is out. Many candidates hate it, and I’d rather find a solution that all candidates like rather than face the prospect of losing your best potential candidates.

And if we’ve learned anything, it’s that doing things at scale breaks things.

And just where are we going with this exactly?

For a lot of recruiters, AI is now baked in. The admin is mostly covered off and sourcing has got more colourful, with a proliferation of alternatives to LinkedIn Recruiter, which will eventually be sued into non-existence once LinkedIn has worked out what they’ve done well.

What is most likely to happen is that LinkedIn will eventually get their act together, as they own the whole database, and they won’t let anyone have full access unless they pay through the nose.

Sourcing will get more expensive with AI, and more frustrating without it.

Candidates know when AI is interjecting into the process, and they don’t always like it. That’s hugely damaging for employer brands, and when the numbers become clear, Talent teams will reject the tools because candidates are exiting the process. We know that our outreach gets at least four times the response rate of the AI-generated emails, so why would we change? AI isn’t going to close the gap on a well-written - and funny - message that comes from deep knowledge of the sector and (not at all creepily) the candidate.

I guess you could summarise it with this: nice tech. Don’t get carried away. You’ll still need to do the work.