What is hierarchy of metrics?

2 States of Metric Tracking

When it comes to using data, businesses often oscillate between two states.

One state is where they track some very high level metrics: revenue, marketing cost, returns, e-commerce conversion rate, etc. But it does not include any relationship of those high level metrics with the input levers and drivers. So, if a high level metric is trending up or down, you won't immediately be able to find the reason for it, or the changes that are needed.

The other state is when they come across a listicle that says '50 useful e-commerce metrics you should be tracking'. And so they start tracking all of them, without establishing any cause-and-effect relationship between them. But nobody in the team really goes through all those 50 metrics after a month or so, after it gets automated.

The Solution: Metric hierarchy

So, what's the solution? What's the sweet spot? A business doesn't need a 'list' of metrics, it needs a 'hierarchy of metrics'.

Let's start working backwards. In terms of outcome, there will probably just be a couple of metrics that matter. At the CEO level, it might be revenue and cost. At the marketing head level it might be traffic and quality of traffic. Essentially a couple of metrics -- one of which measure volume, and the other measures quality, or is a guardrail. So, identify that pair of metrics which make up the first level of the hierarchy of metrics for that report.

Then, define the input metrics for that pair of metrics. For example, revenue is traffic multiplied by e-commerce conversion rate multiplied by average order value. Nothing else outside of these three metrics in a mutually-exclusive-collectively-exhaustive sense. So, the respective MECE input metrics for the level-1 metrics make the level-2 of hierarchy of metrics.

But, here is the important part: if you are satisfied with level-1 metrics, because they seem to be in line with the past numbers, or the targets, you can stop going through the hierarchy of metrics right there.

And, if you are not, you can then go to level 2 and find out the input metric that's causing the deviation, instead of requiring an ad-hoc root cause analysis every time.