Hacker News Reader

Comment by cjbgkagh | Hacker News Reader

I was able to run some stats at scale on this and people who make mistakes are more likely to make more mistakes, not less. Essentially sampling from a distribution of a propensity for mistakes and this dominated any sign of learning from mistakes. Someone who repeatedly makes mistakes is not repeatedly learning, they are accident prone.

View on HN

My impression of mistakes was that they were an indicator of someone who was doing a lot of work. They're not necessarily making mistakes at a higher rate per unit of work, they just do more of both per unit of time.

From that perspective, it makes sense that the people who made the most mistakes in the past will also make the most mistakes in the future, but it's only because the people who did the most work in the past will do the most work in the future.

If you fire everyone who makes mistakes you'll be left only with the people who never make anything at all.

In this case it was trivial to normalize for work done.

It’s very human to want to be forgiving of mistakes, after all who has not made any mistakes, but there are different classes of mistakes made by all different types of people. If you make a mistake you are the same type of person, but if you are pulling from a distribution by sampling by those who have made mistakes you are biasing your sample in favor of those prone to making such mistakes. In my experience any effect of learning is much smaller than this initial bias.

Can you elaborate? What scale? What kind of mistakes? This sounds quite interesting.

A decade of data from many hundreds of people, help desk type roll where all communication was kept, mostly chat logs and emails. Machine learning with manual validation. The goal was to put a dollar figure on mistakes made since the customers were much more likely to quit and never come back if it was our fault, but also many customers are nothing but a constant pain in the ass so it was important to distinguish who was right whenever there was a conflict.

Mistakes made per call, like many things, were on a Pareto distribution, so 90% of the mistakes are made by 10% of the people. Identifying and firing those 10% made a huge difference. Some of the ‘mistakes’ were actually a result of corruption and they had management backing as management was enriching themselves at the cost of the company (a pretty common problem) so the initiative was killed after the first round.

This sounds really interesting but possibly qualitatively different than programming/engineering where automated improvements/iterations are part of the job (and what's rewarded)

What if you define a hard rule from this statistics that « you must fire anyone on error one »? Won’t your company be empty in a rather short timeframe? [or will be composed only of doingNothing people?]

Why would you do that? You’re sampling from a distribution, a single sample only carries a small amount of information, repeat samples compound though.

Or they are working in a very badly designed system which consistently encourages them to make mistakes

Reply on news.ycombinator.com