Why AI screening did not reduce my workload
Most of what is sold as AI screening takes everything in, attaches a score, and passes everything through. The count coming out equals the count going in, which means the work you were told would disappear has been relabelled and handed to whoever sits after the screen.
Ask AI about this
This is the complaint I hear most often from agency owners who bought an AI screening tool. It works, in the sense that every application now carries a score and a summary. The consultants still open every CV, the shortlists take as long to build as they did, and the hiring managers are complaining about the same things. The tool did something, and the something was not less work.
I have spent a few years building screening and matching into a recruitment CRM, and I think the explanation is structural rather than a matter of which vendor you picked. Almost every screening product on the market is built to relabel, and a relabelled pile is the same size as the pile you started with. The difference is one design decision, and it is worth understanding before you buy another one.
A filter is only a filter if things stop at it
Take eight applications and put them through a screen. A screen that filters sends three forward and stops five, and the five are told, and the consultant's pile is now three. A screen that relabels sends all eight forward with a number next to each one. The consultant's pile is still eight. It is a better-organised eight, and the top of it is probably better than the bottom, but nothing has left, and nothing leaving is the definition of no work removed.
Once you look for it, you see the relabel pattern everywhere. Applicant scores, fit percentages, match rankings, AI summaries at the top of each CV. All useful, and all built so that the number of things the next human has to look at is exactly the number that came in.
Why vendors ship the relabel version
The relabel version has two properties a vendor loves. It can never be blamed for a wrong rejection, because it does not reject. And it looks like work was done, because every item now has a score. A product that removes has to be right often enough that the agency is comfortable letting it tell a person no, and the vendor has to stand behind that. Most vendors would rather ship a number and let you carry the decision.
There is a quieter reason too. A screen that removes reduces the pile, and a smaller pile is a worse demo. Vendors are measured on how much their tool touches, and a tool that touches eight and passes eight looks busier than one that touches eight and passes three. The industry has optimised for the wrong picture, and the picture is the one on the sales deck.
Where the work actually went
The work moved, and it moved in the worst possible direction. Because everything is now marked as reviewed, consultants trust the top of the list and send it on faster. The hiring manager receives more CVs than before, with a higher share of ones that do not fit, and does the actual sorting in the interview room. The screen has moved the work from the cheapest person in the chain to the most expensive one, and the most expensive one has the least time.
This is also why the second thing owners report, that clients have gone quieter, is the same problem. A hiring manager who receives six CVs and finds three wrong reads the next batch more slowly. After a few rounds of that, your submissions go into a folder for Friday afternoon. The relabel screen left the work where it was and spent the client's patience as well.
What a screen that removes has to do
The version that reduces load is harder to build and harder to sell, and it comes down to three commitments. It has to say no to some applications, which means a human has set a threshold and takes responsibility for it. It has to say no the same day, with a reason, because a candidate left hanging is a reputation cost and a queue cost at once. And it has to be right often enough that the consultant stops re-reading the rejected pile to check, because a rejection nobody trusts is a relabel with extra steps.
For what it is worth, here is how we drew the line in Recruitly, and why. The applicant AI review puts a skills score, a language score and an overall assessment on every application, and that review is a sort order. It never rejects anyone on its own. Rejection is a consultant's action, with a reason, and the candidate email goes out on a schedule the consultant chose. The AI scorecard drafts the qualification against the job and the consultant changes what is wrong. So the removing is done by a person, fast, with the AI having done the reading, and the relabelling stays where it is harmless, as an order to work through the pile.
Keeping a person in the removing step is a deliberate choice, and it makes for a quieter demo. We think an agency's name is on every rejection it sends, and a model should be allowed to draft that, and read for it, and never sign it.
How to tell which one you have
You do not need a vendor call to find out whether your screening tool removes work or relabels it. A week of watching the pipeline answers it.
- Count what goes in and what comes out at the screening stage for one role. If the numbers are the same, it is a label.
- Check how long a rejected candidate waits to be told. If the answer is "when the consultant gets to it", nothing is being removed at the point the tool runs.
- Ask the hiring manager how many of the last ten CVs were wrong for the role. If the share went up after the tool, the sorting has moved to them, and you will see it next in their reply times.
- Ask your consultants whether they still open every CV. If they do, the tool has not changed their week, whatever the dashboard says about it.
A tool that relabels can still be worth having, as long as you know that is what it does and stop expecting it to reduce anyone's workload. The candidate experience cost of a longer, better-organised queue is real, and so is the client cost of receiving the top of it unread. The screen that actually reduces work is the one where something leaves, and where the person whose name is on the rejection is the one who pressed the button.
Written for agency owners who bought a screening tool and are still opening every CV.
