The problem with AI recruiting software no vendor will tell you

September was a busy month for Truffle with a lot of demos. And one of our sales team was on vacation so it meant more time for me doing founder sales.

Doing a lot of demos in short succession makes it easier to spot patterns and I wrote about some of them here (as well as how we’re trying to solve them).

One meta thing I’m starting to notice though that is harder to solve: we’ve got to stop slapping AI on bad processes.

This is hard to say as someone who founded an AI recruiting software company but I think more people in my position need to call this out if they’re to be taken seriously.

I have seen Truffle absolutely transform hiring for companies, but it’s most often in companies who nailed down their processes.

So here’s what no AI screening vendor will tell you (but probably should).

What AI recruiting software doesn’t do

Recruiting vendor pitches (including ours to some degree) are all about better and faster. You take the tasks recruiters do and you hand them over to a product that can make them (sometimes) better and most definitely faster.

In Truffle’s case you can review hundreds of applications in a day in more granular detail instead of a more human, surface-level volume. That should mean that you have more time to dig into the ones that actually meet your requirements, and you’d expect it to lead to better hires or a better candidate experience or happier hiring managers or some combination of all three.

But if our customers’ hiring process is still designed for a world in which someone could only review a maximum of 50 applications a day, they’re creating problems in other parts of their organization.

More specifically, new tools layered on top of old processes means that humans become the new bottleneck.

For example, if I can get 20 candidates interview-ready in a week instead of 5, that’s awesome, but if my hiring manager only has the capacity to interview 5 people a week, it actually just drags out the process and moves the bottleneck further down the funnel.

How to design hiring processes for AI

Which begs the question: if we were coming in fresh with clear goals and no preconceived notions around how to achieve those goals, how might we design our processes?

I suspect they’d be at least a bit different than the processes that are largely unchanged from how hiring happened 30 years ago.

That’s probably the actual challenge talent teams should be tackling right now rather than figuring out more points at which they can simply infuse another AI tool.

These are thorny problems which is why companies aren’t rushing to tackle them. I don’t pretend to have all the answers but here are a few thoughts:

1. Capacity isn’t zero

Even if you can screen more candidates with AI you still need a human somewhere in the loop. Therefore you need to think about the full hiring team’s capacity.

You should also consider doing things like removing stages if it duplicates work already done. You’d be surprised how many companies do one-way interviews and then ask the same questions on a phone screen.

This makes us and Walter sad.

2. Interview fewer people

When I see people crap all over async interviews or anything that asks people who apply to a job to do more than submit a resume I get a little sad.

If you have hundreds (or even thousands) of people applying, the likelihood of your resume getting lost in the shuffle is very high. If companies have the ability to go deeper in a larger pool, you’re more likely to surface individuals that otherwise wouldn’t have gotten a second look.

It sounds counterintuitive, but the point is to make those live interviews count. Give more people a fair look early, then spend time with the candidates where there’s a reason to keep talking.

3. Use deadlines

“Waiting for feedback” is a killer for everyone involved. I think more hiring processes should set deadlines for when certain things will be done, e.g. first-round interviews completed by X date.

You can also set expectations like “all interview feedback submitted within 24 hours”.

4. Better measure hiring outcomes

This is on Truffle’s roadmap to help better identify whether hires actually worked out. This would become a nice self-learning driver for our AI for your specific company’s hiring needs and taste.

In the meantime it’s definitely something we recommend our customers do.

Fixing hiring isn’t just a software problem

I read a few manifestos that promised to fix hiring using AI. Sadly a very human process won’t be fixed simply by throwing AI at it.

But AI can, when combined with better process design, improve things. I’m a real witness to it.