4 ms·

If you're hiring people you don't have high conviction in, then of course you'll get relatively worse outcomes. I would question why. I do think companies need

If you're hiring people you don't have high conviction in, then of course you'll get relatively worse outcomes. I would question why. I do think companies need to set a target aptitude percentile they're trying to hire, as it has to be somewhat aligned with comp percentile to have success. You can't try to hire 99th percentile engineers with 50th percentile pay and expect to land any hires But if you set a lower target skill percentile to match lower comp, then you're actively choosing to lower the bar, so there should also be a lower bar for considering somebody a poor hire. At least in our case we were able to get by targeting ~80th percentile engineers at like 60th percentile comp. Mostly through hiring people that didn't know their worth in the market, but also being a good place to work, with low attrition (non comp factors)

mlyle 4y ago
It seems like you've been hiring in domains that are a bit less technically demanding than the ones I was in. Even with a whole lot spent on recruiting and careful resume screens, about 5% of interviewees were people I was 95% confident could do distributed systems work; another 10% of interviewees were people that I was 90% confident could do the work. Turning the comp knob doesn't affect these numbers that much. So the acceptance criteria to get a reasonable hire rate and not spend all of your time interviewing becomes that 90% bar and being able to imagine working with that person without any obvious toxic or crazy behavior. In turn, that becomes about 15-20% not working out after hire. Setting the bar to 95% and really sure of cultural fit would be hiring at 1/3rd to 1/4th the pace, and you'd still have 7-8% not working out, which isn't that optimum of an operating point. I'd also have missed the absolute best performer I ever hired.
- adam_arthur 4y ago
  Sure the domain of the work matters a lot. But the concept of skill distribution applies to all domains. You can hire a 10x engineer in a basic web dev role the same as you can hire a 10x in a distributed systems one. It's possible distributed systems work is inherently more difficult to assess, but I'm not sure. It seems the process just needs to be different to assess that skillset. If your best hire is somebody you had less confidence in, then it's a sign your confidence function is not fully tuned to what's causal to success. But you're right that you can't know everything after an interview. Work ethic cannot be measured directly as you stated, but usually it can be inferred with a reasonable level of confidence through behavioral cues of the candidate. A perfect correlation between hire confidence and performance on the job implies you have a perfect hiring process. Obviously impossible in practice, but you can get a lot closer than the average hiring process, or FAANG. Obviously to iteratively improve your hiring process, it's important to view it this way. Unfortunately most confuse and forget about what the relationship between hiring and on the job performance should be, and see hiring as a separate siloed game and view signals in the context of that game rather than the job. Finally, it's pretty bad for company culture and morale to have a high fire rate. It's worth going through more interviews to limit the bad hires. But obviously the tradeoff needs to be quantified and deliberately decided on. If your average tenure is < 2 years, there's something wrong
  - mlyle 4y ago
    > If your best hire is somebody you had less confidence in, then it's a sign your confidence function is not fully tuned to what's causal to success. I think it's unreasonable to think that you can guess so well as to how performance will be over 6 months-1 year+ periods in several hours of measurement. > If your average tenure is < 2 years, there's something wrong My average tenure was much higher than that. But again, about 1 in 6 didn't make it a year. Turnover rate past that year was very low. I'm not sure why you think having 15-20% of people not working out in the first year (after which turnover drops) translates to an average tenure of less than 2 years. It implies one greater than 5 years. > If your best hire is somebody you had less confidence in, then it's a sign your confidence function is not fully tuned to what's causal to success. If I had a fully tuned confidence function, it would output 1.0 for candidates that would work out and 0.0 for candidates that wouldn't. This isn't realistic. Nor is it realistic for it to monotonically increase with candidate quality (in which case it could be transformed to the above function with a threshold). > Finally, it's pretty bad for company culture and morale to have a high fire rate. I don't think getting up to 1 in 6 (first year) freaks people out too bad. Rates across a broad variety of industries and employers are more than 1 in 3 quitting or being fired in the first 18 mos. I think one can do significantly better than this, but I also think 1 in 10 isn't attainable and question your numbers.

mlyle 4y ago

It seems like you've been hiring in domains that are a bit less technically demanding than the ones I was in. Even with a whole lot spent on recruiting and careful resume screens, about 5% of interviewees were people I was 95% confident could do distributed systems work; another 10% of interviewees were people that I was 90% confident could do the work. Turning the comp knob doesn't affect these numbers that much. So the acceptance criteria to get a reasonable hire rate and not spend all of your time interviewing becomes that 90% bar and being able to imagine working with that person without any obvious toxic or crazy behavior. In turn, that becomes about 15-20% not working out after hire. Setting the bar to 95% and really sure of cultural fit would be hiring at 1/3rd to 1/4th the pace, and you'd still have 7-8% not working out, which isn't that optimum of an operating point. I'd also have missed the absolute best performer I ever hired.

adam_arthur 4y ago
Sure the domain of the work matters a lot. But the concept of skill distribution applies to all domains. You can hire a 10x engineer in a basic web dev role the same as you can hire a 10x in a distributed systems one. It's possible distributed systems work is inherently more difficult to assess, but I'm not sure. It seems the process just needs to be different to assess that skillset. If your best hire is somebody you had less confidence in, then it's a sign your confidence function is not fully tuned to what's causal to success. But you're right that you can't know everything after an interview. Work ethic cannot be measured directly as you stated, but usually it can be inferred with a reasonable level of confidence through behavioral cues of the candidate. A perfect correlation between hire confidence and performance on the job implies you have a perfect hiring process. Obviously impossible in practice, but you can get a lot closer than the average hiring process, or FAANG. Obviously to iteratively improve your hiring process, it's important to view it this way. Unfortunately most confuse and forget about what the relationship between hiring and on the job performance should be, and see hiring as a separate siloed game and view signals in the context of that game rather than the job. Finally, it's pretty bad for company culture and morale to have a high fire rate. It's worth going through more interviews to limit the bad hires. But obviously the tradeoff needs to be quantified and deliberately decided on. If your average tenure is < 2 years, there's something wrong
- mlyle 4y ago
  > If your best hire is somebody you had less confidence in, then it's a sign your confidence function is not fully tuned to what's causal to success. I think it's unreasonable to think that you can guess so well as to how performance will be over 6 months-1 year+ periods in several hours of measurement. > If your average tenure is < 2 years, there's something wrong My average tenure was much higher than that. But again, about 1 in 6 didn't make it a year. Turnover rate past that year was very low. I'm not sure why you think having 15-20% of people not working out in the first year (after which turnover drops) translates to an average tenure of less than 2 years. It implies one greater than 5 years. > If your best hire is somebody you had less confidence in, then it's a sign your confidence function is not fully tuned to what's causal to success. If I had a fully tuned confidence function, it would output 1.0 for candidates that would work out and 0.0 for candidates that wouldn't. This isn't realistic. Nor is it realistic for it to monotonically increase with candidate quality (in which case it could be transformed to the above function with a threshold). > Finally, it's pretty bad for company culture and morale to have a high fire rate. I don't think getting up to 1 in 6 (first year) freaks people out too bad. Rates across a broad variety of industries and employers are more than 1 in 3 quitting or being fired in the first 18 mos. I think one can do significantly better than this, but I also think 1 in 10 isn't attainable and question your numbers.