DataTaunew | comments | leaders | submitlogin
How to Consistently Hire Remarkable Data Scientists (firstround.com)
11 points by ebellm 3289 days ago | 7 comments


4 points by ramsey 3285 days ago | link

I have a data driven job hunting algorithm. Let's see how Sailthru matches up ;)

There's no mention of ever evaluating a candidates github code and contributions. That's a bit like me firing off a resume without tailoring it to the employer, don't you think?

The job posting

https://boards.greenhouse.io/sailthru/jobs/48920?t=awnye3

is extremely generic. There's no indication of what a recruit would actually be doing. It's keyword bingo.

Points are deducted for mentioning foosball. It sounds like brogrammer paradise already.

500 peoples' time is wasted to hire three? About the same odds as a lottery scratch off, but it takes all week and has none of the fun instant gratification.

There's absolutely no mention of salary range. I'm probably earning more than this "compelling and irresistible" offer.

Result: I'd probably self select out of this process.

On the plus side, you do get points for not asking ridiculous questions like "How many golf balls do you think it would take to fill a school bus?"

Disclaimer: Intended as light hearted criticism. I hope you take it as such.

-----

4 points by sciencedaedalus 3288 days ago | link

I interviewed with Jeremy's team at Sailthru using this process.

Result: Take Home -> Data Day -> No Offer.

It was an overall great experience. Afterwards I felt like I had a really good idea of the "fit". I understood the kinds of projects I'd be working on, I understood how their team functions, and most importantly I understood that I likely wouldn't be a good fit.

If I have a differentiating skill; it's Data Mining. A skill that I wasn't really able to show off.

From the article (regarding the sample data set):

"... I would suggest keeping this sample reasonably clean to ensure that they don’t waste valuable hours on munging that would otherwise have gone into analysis or modeling."

That pretty much kills my wow-factor. Most of the problems I work on don't start with a structured data set to build from, generally the biggest challenges I face (and enjoy) involve identifying and structuring the data needed to answer a question.

Hence the understanding that I wouldn't be a good fit.

Data day was a lot of fun, I certainly didn't consider it a waste of time; even knowing right off the bat I wouldn't be able to really shine. As the candidate, I garnered much insight about the job's nature and got to meet some awesome people.

Advice to other companies considering implementing a similar system:

1. Do it.

2. Tailor Data Day to select for the kind of skill set you need the most i.e If you want a candidate like me, be more withholding with your data :)

-----

3 points by jeremystan 3288 days ago | link

Great feedback and I really appreciate your positive words on the process despite the "no offer" outcome.

I will think more about expanding the variety of data included in the data day (we certainly have a ton of it!). But I worry about candidates getting lost in the complexity or pursuing too many rabbit holes. Having a narrower dataset increases the chance that the candidate can produce something meaningful in a fixed amount of time. It also increases the comparability across data days.

But I agree that a wider and less structured set of data would be a more realistic experience.

Best, Jeremy

-----

3 points by roycoding 3289 days ago | link

Our process may generate too many true negatives — i.e. candidates who are well suited to data science (true) but who do not receive an offer (negative).

Isn't that a false negative? :)

-----

2 points by jeremystan 3288 days ago | link

Good catch (and an embarrassing slip-up). First Round was kind enough to fix it.

-Jeremy

-----

2 points by elyase 3289 days ago | link

I agree with @roycoding, the test (making an offer) has two possible outcomes, positive (Yes offer) and negative (No offer). The test will be right(True) or wrong(False) depending on the actual quality of the candidate:

Good candidates | Yes Offer -> True Positive

Bad Candidates | Yes Offer -> False Positive

Good Candidate | No Offer -> False Negative

Bad Candidate | No Offer -> True Negative

-----

2 points by roycoding 3289 days ago | link

I think this is an interesting article. My team has just started looking to hire another data scientist and we have begun discussing strategy. What the author describes probably won't work exactly for us, since we have a much smaller company (i.e. no recruiters or support staff). I do like the idea of sending a take-home "test" to all comers.

FYI, our job ad is here: https://jobs.lever.co/zoomer

-----




RSS | Announcements