2,418 Profiles. Now What?

More candidates was never the goal. The right ones were.

AI sourcing tools are very good at the first part. And that's exactly how they make the second part harder.

What Happened at Sourcing Summit Europe

Viktor Malyi wrote a long recap of Sourcing Summit Europe in Amsterdam this month. It opens with a sourcer telling him that neither Claude nor ChatGPT works for their sourcing, because they're "totally unpredictable".

A few numbers from the talks stayed with me.

One search, one lost week. Yves Greijn showed 2,418 profiles from a single search. By his own maths, that's about a full working week of doing nothing but reading them.

One day, 5,426 profiles. Guillaume Alexandre showed his own AI setup finding and analysing 5,426 profiles in a single day. He told a room full of sourcers that the "finding people" part of their job is going to machines.

Half the people, gone with one filter. Guillaume also showed that of about 8,000 people working at Rolex, 47.5% aren't tagged with the luxury industry on LinkedIn. One filter, and half of them disappear. Put an AI ranking on top of that, and you never even know who you missed.

Same brief, different people. Viktor ran the same sourcing brief through Claude three times. The lists shared only about 15% of the same people. So who was the top match, the one from Monday or the one from Tuesday?

Yves summed it up in one line: "finding people got cheap, knowing who matters is the actual job now."

Where Sourcing Ends, Interviewing Starts

Here's what the recap doesn't say. Every profile that makes it through sourcing ends up somewhere. Usually, in an interview.

If your tool hands you 2,418 people and no reason for any of them, the decision doesn't go away. It just moves down the funnel. Recruiters screen more people. Engineers sit in more interviews. And engineering hours are the most expensive hours in your hiring process.

A shortlist with no reason behind it is just a longer list.

It goes the other way too. If your filter quietly dropped half the good people, no interview in the world will find them. The best interview process can only pick from who's in front of it.

A Profile Is Not an Interview

Before AssessDev, I was a startup CTO for 8 years, and I sourced and hired engineers myself. Some of the best engineers I hired had LinkedIn profiles that said almost nothing. Some of the most impressive profiles fell apart in the first 15 minutes of a technical interview. (I think I've got a pretty good LinkedIn profile. I wonder how I'd perform in an interview.)

A profile tells you what someone wrote about themselves. An interview tells you how they think and shows you what they can do. Those are not the same thing, and I wonder if any amount of AI ranking on top of profiles could change that.

The Honest Trade-Off

Viktor and Yves both build sourcing tools, so they have a view. But the problem they describe is one I see too.

And I'm not against AI sourcing. In fact, I'm secretly rooting for it. If it solves the problem we've all had these past couple of years, I'll be the first to cheer. It just doesn't remove the decision. It moves it.

How to Use AI Sourcing Without Drowning Your Interviewers

Three things, small enough to start next Monday.

1. Agree what good looks like before anyone searches. Diane Rocher's first rule at the summit was that no AI fixes a bad intake meeting. Write down the must-haves and the deal-breakers with the hiring manager first, and use the same list for the sourcing brief and the interview scorecard. Then ask for fewer people, with a reason for each one. If someone was picked for "built a payments system at scale", that's your first interview question: tell me about it. Follow up until you get to the nitty-gritty parts.
2. Test the tool before you trust it. Viktor's test is simple: run your AI sourcing on one role two or three times and compare the lists. If they barely overlap, you know how much to trust it. Don't trust the filters either. People don't keep their profiles up to date, so look at who the search didn't return, not only who it did. And read what the machine gives you, every time. Diane shared that one of her tools reported "success" while delivering only half a job description. A successful run is not a successful result.
3. Use Leandro's 10-people test. Leandro Gomes da Silva had one quality test for outreach: could I send this message to 10 people unchanged? If yes, rewrite it. His reply rates went from 30% to 60% at one company, and from 19% to 40% at another. I'd use the same test for interview invites and for feedback after an interview. If your feedback could go to 10 candidates unchanged, it isn't feedback.

One more thing while you're running Viktor's test: try it on your interviews too. Would two of your interviewers score the same candidate the same way? If not, you have the same problem, just with people.

Your Checklist

• ☐ Must-haves and deal-breakers are agreed with the hiring manager before sourcing starts.
• ☐ The sourcing brief and the interview scorecard use the same list.
• ☐ Every shortlisted person comes with a reason.
• ☐ That reason turns into a question in the first interview.
• ☐ We've run the same search twice and compared the lists.
• ☐ We know how many profiles we get per role, and how many we actually contact.
• ☐ Outreach, invites and feedback pass the 10-people test.
• ☐ A person reads everything before it goes to a candidate.

Finding people got cheap. Deciding who matters didn't, and that decision always ends up in an interview. It's still owned by people, real human beings like you and me.

So, how many profiles does your sourcing tool hand you for one tech role, and how many of them do you actually interview?

Sources

• Viktor Malyi, How people sourcing will work in 2027: insights from Sourcing Summit Europe. LinkedIn, 10 October 2026. Talks quoted from it: Yves Greijn, Guillaume Alexandre ("The Great Pivot"), Leandro Gomes da Silva ("The Message Is Automated. But The Reason to Reply Isn't") and Diane Rocher.

This post first appeared in Shortlisted by Kat. Photo by Markus Winkler on Unsplash.

From Shortlist to Hire

Start with what good looks like, then let a senior engineer test it in a structured interview.