HR can be obsessed by metrics or KPIs (a KPI is a metric that you deem to be important). However the way these metrics are used often inhibits decision making and often creates undesirable incentives.
A metric as an average, and averages obscure
Let’s take an example; turnover rate. The usual calculation is that:
Turnover=number of leavers during a time period / average number of employees during the same time period.
What the calculation does is reduces the full data to a single number. It doesn’t in itself tell you why it happened, whether it is a good thing, what could have been done to change it etc.
There are two ways of increasing the turnover ratio. You can increase the number of leavers or decrease the number of employees. If you’re hiring aggressively you’ll see a drop in the turnover. As soon as you stop the turnover rate is likely to go up.
As discussed earlier, every individual has a likelihood of leaving in the next time period conditional on a wide range of factors from how long they’ve had their new job or boss to their age, gender and education level. Typically you want to know whether, given these conditional probabilities, variables which the organization influences is causing a change in the turnover. You want to know where there is an issue and what you can do to fix it.
A single turnover metric doesn’t help you do that. It’s an average. After calculating you have less information than you had at the start.
In 1973, Francis J. Anscombe published a paper titled, Graphs in Statistical Analysis. In it he highlights why using graphs (visualisation) to understand the data is essential. He uses 4 datasets of 11 paired variables (x & y) which all have the following properties:
Mean y=7.5 (to 1 decimal place)
Variance y=4.1 (to 1 decimal place)
Correlation=0.816 (3 decimal places)
Linear regression line y=3+.5x (to 2 & 3 decimal places)
We can see that by most usual data reduction method (which can be seen as similar to metrics) the 4 data sets are effectively the same. However when they are plotted you can clearly see they have completely different properties:
A similar thing can quite easily happen when looking at metrics: the true picture is hidden. The solution is the same – visualise the data points to see the real relationships. It’s why we always try and show distributions rather than just COUNTs or averages.
Make sure you’re comparing like with like.
Let’s consider an HR department. It has a much higher ratio of HR staff to employees than the benchmark. That would obviously show inefficiency, right?
Not so easy. Let’s say that company has highly paid staff and that as a business the most efficient way of delivering employee services is to shift work away from expensive line managers to (relatively) inefficent HR managers. That would increase the ratio of HR to employees but could be the most efficient solution. Alternatively the firm might believe in offering a high level of service from HR and prioritises this over the cost. When comparing yourself against a benchmark you’re unlikely to see the full picture.
A better single point metric.
Let’s say you need to use one figure but don’t know what to use; what would we suggest?
If you have to pick one figure then the one that you should use is:
the marginal value.
Cost per hire could be brought down by lower cost (the top) or higher number of hires (the bottom). A declining cost per hire can easily obscure a rising cost of hiring the next employee. That cost of hiring the additional employee is the marginal cost of hiring. You can plot this easily to show how it changes as the number of hires changes and it’s much more likely to show where and when you need to address an issue.
The best solution would be to visualise the data in combination with using multiple figures, including both ‘metrics’ and marginal values.