Wednesday, June 8, 2011

Three Things You Need On Every KPI Graph

The frighteningly vast majority of performance reports and dashboards are filled with graphs that fail to communicate the real signals our performance measures and KPIs are desperate to show us.

example of control chartThese reports encourage us to react to trends that aren't there at all, and to miss important signals about future problems. We need a revolution in the performance reporting space, and a big part of that revolution should be the use of control charts, charts that include 3 very powerful features that dramatically improve our interpretation of and response to our performance measures and KPIs.

FEATURE 1: Your performance measure in time series

This month compared to last month, this month compared to the same month last year, this month compared to target – all are limited and risky, misleading comparisons to make. Why? Because you have no context within which to assess if the difference you're looking at is typical or not. There will always be differences, but the only way to see which ones are worthy of your response (or explanation) is to see the data in the full context of a time series.

Plot your performance measures in time series, with as much history as you can reasonably get your hands on. Around 20 values is good, such as the last two years' worth of monthly values.

FEATURE 2: The measure's mean line

The signals about performance lie in the patterns, not the points of data. Two points is not a big enough sample to determine if there's a change. Adding a mean line to your time series, calculated from the first 10 points or so and then kept constant, will give you a visual benchmark to determine when future points of your measure are behaving differently from the past.

You're starting to build what's called a control chart, and there are statistical rules that help you know when a signal is there and when to recalculate a new mean line. For example, one of the signals of change is a run of 7 points of your performance measure on one side of the mean line. You can then calculate a new mean line using those 7 points.

FEATURE 3: The measure's limits of natural of variation

Everything varies. Everything. And performance measure values tend to vary quite a bit, from month to month or week to week or quarter to quarter. Just because one point varies more than 10% from your average, doesn't mean you have a problem. It depends on how much natural variation is in your measure to start with.

Control charts include a measure of the natural variation. For example, if a point falls outside the limits of natural variation on the chart, then you know something has happened out of the ordinary and it's worth following up. Limits of natural variation also help you pick up signals from only 3 or 4 points of data in a row (you don't always have to wait for 7 points to behave differently).

BUT: You do need to learn how to use these control charts.

It doesn't come naturally, and people who don't take the time to learn make mistakes in constructing and interpreting them. But they're really easy to learn and VERY worthwhile learning.

The most convincing and clearly articulated resource that I've ever seen about using charts like these for management reporting is Donald Wheeler's book, Understanding Variation: The Key to Managing Chaos. If you consider yourself interested in performance measurement and performance improvement, this book MUST be in your personal toolkit.

Ratchet up your knowledge on control charts for management reporting. Get a copy of "Understanding Variation: The Key to Managing Chaos" and then go find some performance data, and have a go at using a control chart to reveal the true signals.

Thanks to Stacey Barr / Stacey Barr - Measure Up

No comments: