Tel: (01) 7671786

Mon → Sat : 6am-10pm

Email: [email protected]






OKR Anti-pattern: Using measures of system behavior

Photo by Pixabay on

As we’ve discussed multiple times here, your key results should be outcomes. Outcomes are measures of human behavior that deliver business value. This is what sets OKR’s apart from other goal-setting frameworks. We’re no longer managing to features, or output. We’re managing to outcomes, or measures of behaviors our users and customers do in our products and services. The metrics you choose as your key results help you define done and need to be specific and verifiable with evidence. 

It’s a percentage so it’s a good metric, right?

One anti-pattern that emerges regularly is teams adding in any metric they can come up with as long as it has a percentage. The rationalization is often as simple as that. It has a “%” sign in it so it’s a good key result. Teams often fail to dig deeper into exactly who is performing the behavior that is represented in the number immediately to the left of the % sign.

We’ll see metrics like, “reduce home-page load time by 80%.” A supermarket company I worked with once offered metrics like, “50% of products on our shelves will be store brand products.” In both examples the teams were confident these made for good key results. After all, they got specific with a number. It was related to the usage of their “system.” And, well, it was a ratio or rate – i.e., it had a percentage sign after it. It turns out though, these are not good key results. 

Is it the product or the customer?

System behavior can look a lot like human behavior when you reduce both down to metrics. However, system behavior (what the product/service/software/system does) is not the same as human behavior. System behavior is a feature of your product. The metric associated with it is also a feature of your service. It helps you determine to what extent you’ll implement that feature. It doesn’t, however, tell you whether or not you’ve delivered value. 

Let’s take another look at the two examples from above:

  1. “Reduce home-page load time by 80%” – This is a measure of how fast your home-page loads. Improving this behavior of your system can be an infinite process. You can always make the home-page load faster. At what point is it fast enough? At what point does it not make sense to spend any more effort improving this aspect of your user experience? The resulting human behavior to your site loading faster is a reduction in bounce rate. Reducing the bounce rate is a measure of human behavior. Bounce rate tells you whether you’re delivering value. When it reaches an acceptable level (you decide what that is) you can stop optimizing home-page load time. This is a good key result.
  2. “50% of products on our shelves will be store brand products” – This one, too, feels like a human behavior at first. In fact, it will likely take humans to make sure the shelves are stocked with at least 50% of our own products. However, this too is a feature of your service. The desired user behavior is an increase in the frequency of same-store visits and in average order value. Depending on the products you stock (and a myriad of other factors) the likelihood of someone coming back to your store and spending more money than last time will vary. The former is a feature of your store (what products you choose to stock and which brands they come from). The latter is an outcome. It’s a measure of human behavior and is therefore a good key result. 

Who does what by how much?

To help avoid this anti-pattern follow this easy validation technique for every key result your team generates. Ensure each metric can easily fit into this template:

[Who] does [what] by [how much]

Let’s try it:

[Our website] [loads faster] by [80%] – not a human action. 

[Our daily users] [don’t bounce off the home-page] by at least [90%] – a human action. 

Remember, as we improve each part of our system’s functionality, we must ask, “What will people be doing differently if we do a great job?” That’s our measure of success. That’s a good key result.

Source link