When good metrics are bad
Photo by Luke Chesser on Unsplash

When good metrics are bad

For all the value they bring, business metrics can also lead to trouble. There are the people who rely too much on metrics (and ignore other context). Then you have the poor choices of metrics (which measure the wrong thing). Or the ill-fitting presentation thereof (like that fancy dashboard that no one looks at any more).

Something you rarely hear about, though, is when a good metric is bad.


Yesterday Solomon Kahn kicked off a thread on metrics and it reminded me of an underappreciated aspect thereof: simplicity.

If you work alongside data scientists (or quants, or similar roles) you probably hear the phrase reduction of dimensionality a fair amount.?This is a nice way of saying "we're going to take a large amount of info, then boil it down to a smaller set of info, so that it's easier to digest."?(This usually improves ML modeling, so that's a Good Thing.)


A metric is also a form of dimensionality reduction.


Credit scores??A project's red/yellow/green status? Business valuations??All of them squish several months' or years' worth of fine-grained records into something simple.


And this is fine, so long as you remind yourself that a metric is -- by definition -- an oversimplification.

It becomes a problem when you forget how much detail you lose on the journey to that simple number or status.


The take-away: It's time to take a look at your metrics. Which ones do you actually use? When did you last inspect them, to reminder yourself how they came to be, and what messy details they hide?

David Trollope

Finding insights in surveys

2 年

Can’t agree more. So often, I hear “Be data driven”, “collect data to support your decision”, but often metrics are collected and used beyond their purpose. It’s important to regularly review why you are using a metric as well as what the metric is telling you, discarding if it no longer makes sense. I like to think of metrics as ice cream. They are tasty when you first get them, but if you hold on to too long it melts and all you have left is a mess on your hands. ??

要查看或添加评论,请登录

Q McCallum的更多文章

  • My favorite writing from 2024

    My favorite writing from 2024

    I'm sharing a list of pieces I really enjoyed writing in 2024. (The list of what I enjoyed reading would run for ages…

  • When generated images take on a life of their own

    When generated images take on a life of their own

    (Photo by Cullan Smith on Unsplash.) (This LinkedIn article is mirrored from the post on my website.

  • Measuring the wrong thing

    Measuring the wrong thing

    (Image credit: patricia serna on Unsplash) I never thought I'd have quite this much to say about metrics, but after…

  • Same name, new face for AI

    Same name, new face for AI

    (Image credit: Dawn Low.) Lately I've seen several articles like this one, about this hot "AI" field that's getting all…

    2 条评论
  • Congratulations, you are now a data company

    Congratulations, you are now a data company

    (Photo by Danist Soh on Unsplash) I wrote a lot of software for companies earlier in my career. One lesson I learned…

    1 条评论
  • When your metrics are fooling you

    When your metrics are fooling you

    Photo by Dillon Wanner on Unsplash Following my posts on metrics and companies drifting into autopilot, today I have a…

  • Is your company on autopilot?

    Is your company on autopilot?

    (Photo by Cédric Dhaenens on Unsplash) Do you have a minute? Look around your company and ask yourself: "Why are we…

  • The top failure modes of an ML/AI modeling project (Part 2)

    The top failure modes of an ML/AI modeling project (Part 2)

    Someone once told me that risk management is a matter of asking "What are you worried about? And what are you going to…

  • The top failure modes of an ML/AI modeling project (Part 1)

    The top failure modes of an ML/AI modeling project (Part 1)

    The good part about machine learning (ML): you can build a model to automate document classification, pricing…

    3 条评论
  • When your ML model is living in the past

    When your ML model is living in the past

    This screen cap is from a newsletter I was working on a couple of days ago (January 2023): The Google Docs grammar…

    2 条评论

社区洞察

其他会员也浏览了