What happens when a metric becomes a target?
Reflections on Goodhart's Law
I’ve been a runner since my early teens.
My dad got me into the sport, initially taking me and my sister on easy jogs around the block, gradually building up to more challenging routes and distances.
For a long time, my relationship with running had nothing to do with competition.
I ran because I enjoyed it, because it helped clear my mind, and because it was something I did with my family.
To the extent I was interested in getting faster, the metrics I used to track my fitness were rudimentary. I had a regular, uphill route I ran at least once a week, and a basic Casio wristwatch to measure my time.
I had no concept of the distance I was running, nor of my split pace, nor of the total elevation. All I had was a general sense that if I could complete the route in under 30 minutes, I was in pretty good shape.
Over the last few years, I’ve noticed my relationship with running change: a change that started when I downloaded Strava.
With Strava, I can now see detailed metrics for every run I complete, including distance, heart-rate zones, and pace per kilometre. Through the app, others can see these stats too.
On the one hand, seeing my average pace per kilometre has arguably made me a better runner, motivating me to build fitness and bring that number down. At a subconscious level, the public nature of this stat likely pushes me to try a little harder than I otherwise would.
But what happens when 00:00/KM stops being just a metric and starts to become a target?
If maximizing my pace per kilometre becomes my primary goal, this incentivizes me to avoid more challenging uphill routes. It also disincentivizes me to complete longer runs at an easy pace: the kind of running that is vital to building endurance.
This is an example of Goodhart’s Law, which is commonly phrased as: ‘When a measure becomes a target, it ceases to be a good measure.’
Once you notice Goodhart’s Law, you start to see it everywhere.
You see it in the education system, where prioritizing attainment incentivizes schools to ‘teach to the test’.
You see it in customer service, where rewarding the number of calls handled leads agents to rush to resolution, instead of focusing on solving problems.
And you see it in L&D, where targeting satisfaction scores leads teams to design interventions that are more ‘fun’ than they are effective.
This is not to say that test scores, call-handling volumes, satisfaction rates, or kilometre pace are poor metrics. But they become less useful when they start to be seen as targets.
For L&D practitioners, there is no getting round Goodhart’s Law, but there are steps you can take to mitigate its impact:
📊 Use multiple metrics, not just one — Designing an evaluation strategy that includes metrics at different tiers of the LTEM or Kirkpatrick models makes it less likely that one metric will shape behavior in unintended or undesirable ways.
❤️ Combine quantitative data with qualitative insights — Using focus groups, interviews, or free-text survey questions adds human color to quantitative measures, providing complementary data that can’t be easily gamed.
⚡ Connect metrics to behavioral and performance outcomes — Measuring outcomes tells you if all those stats you’ve been trying to optimise actually lead to better results.
Want to share your thoughts on this week’s newsletter? Need help with your next project? Get in touch by emailing custom@mindtools.com or reply to this newsletter from your inbox.
🎧 On the podcast
Completion stats are a vanity metric. Happy sheets are pointless. Revenue impact will get you a seat at the table.
We’ve all heard these statements from conference stages. Some of us, your hosts included, might even have said them. But in last week’s episode of The Mindtools L&D Podcast, we wanted to dig a little deeper.
We invited Evolve L&D’s Tom McDowall to join Ross G and Claire, to discuss:
why the context behind completion metrics is crucial to understanding them
how to design effective learner surveys
how ‘metric chains’ help broaden our understanding of learning impact.
Check out the episode below. 👇
You can subscribe to the podcast on iTunes, Spotify or the podcast page of our website.
📖 Deep dive
It is axiomatic that in today’s fast-changing, technologically disrupted business environment, workers need to be able to adapt quickly and learn new skills.
But what role do different leadership styles play in enabling adaptive performance (AP)?
In a 2024 meta-analysis bringing together over 30 studies, Bonini et al. set out to answer this question, hypothesizing that styles emphasizing worker-involvement (e.g. transformational) would be more strongly related to adaptive performance than control-based approaches.
What they found was that although there is a moderate, statistically significant relationship between leadership and adaptive performance (r ≈ .37), no single leadership style conveys a particular advantage:
‘[…]contrary to expectations, there was no evidence supporting the existence of a stronger relationship between one or more leadership styles and AP(H2) and there was no difference between more or less top-down styles.’
Instead, leadership appears to support adaptive performance by creating a trusting, psychologically safe environment for people to engage with change:
‘The opportunity for members to share leadership behaviors created a supportive climate that promoted proactivity and, in turn, stimulated adaptivity. Furthermore, by encouraging mutual trust, members developed a psychological safety net that was ideal for individual initiative and it encouraged group discussion on goals, strategies, processes and how to cope with new, unpredictable or paradoxical situations.’
Bonini et al. (2024). ‘The relationship between leadership and adaptive performance: A systematic review and meta-analysis’. PLOS ONE
👹 Missing links
🤖 The human skill that eludes AI
The latest versions of ChatGPT, Claude, and Gemini make fewer mistakes than their predecessors. But has something been lost in the process? According to Jasmine Sun, early models like GPT-2 were far weirder and more prone to hallucinations, but these qualities arguably made them better at creative writing. In a recent interview, Sam Altman conceded that even future models may only be able to produce something equivalent to ‘a real poet’s okay poem’.
🫠 Your boss just did your job over the weekend
If you’re the person who likes to show up to meetings with the cool thing you ‘vibe coded’ on Sunday night, you may want to think about the message that sends to your colleagues. Even if AI can help you achieve things that would previously have taken days or weeks in a matter of hours, your flashy demo might get people’s backs up if it sounds like you’re undermining their work. That doesn’t mean you should ignore reality, or that you shouldn’t be having these conversations as a team. But how you approach these conversations matters if your goal is to bring people with you.
😑 Burnout isn’t a badge of honor
As David Kelly points out in this edition of his newsletter, there’s a difference between working late because you’re engrossed in what you’re doing and working late because your ‘workload has backed you into a corner’. Yet we often treat the latter as a badge of honor, proof of our commitment to our organizations, to our stakeholders, and to the noble work of learning. To help spot the warning signs he’s on the path to burnout, David has a Post-it note on the wall behind his desk that reads: ‘Are you swimming or avoiding drowning?’ The two things might look the same, but they feel very different.
👋 And finally…
As a card-carrying nerd and a fan of the TV show Scrubs, this video felt like it was made just for me:
👍 Thanks!
Thanks for reading The L&D Dispatch from Mindtools Kineo! If you’d like to speak to us, work with us, or make a suggestion, you can email custom@mindtools.com.
Or just hit reply to this email!
Hey here’s a thing! If you’ve reached all the way to the end of this newsletter, then you must really love it!
Why not share that love by hitting the button below, or just forward it to a friend?