Look at the chart above. From a cursory view, it tells the sales trend of a product for a certain year. We can notice that the sales go up and down at various months throughout the year. If I were a member of the management team, I would make the interpretation that the sale of this particular product is somewhat stable for that year. However, I would not be able to tell whether how good the product’s sales numbers are in relation to its monthly (or yearly) sales targets, or how I am faring compared to my competitors for similar products. This is where using context can help enhance the quality of your data presentation to your audience.
Some options for providing context…
I have pointed out several examples of contextual data above. Let us take a look at some of them in more detail below.
- Organizational Target
- Industry Standard
- Common Average
- Nearest Competitor
This contextual data is very useful in measuring internal performance indicators, such as sales in the chart above. Many corporations use organizational target as their default performance measurement benchmark in various indicators, such as sales, revenue, number of customers, and others.
The good thing about this type of contextual data is that it largely remains fixed throughout a particular time frame, say a year. For example, you hardly hear a sales target to change over a fiscal year, once it is set at the beginning of the fiscal year. Due to this, these organizational measurements can be easily tied to employee performance – their individual targets are largely fixed, and they can effectively plan towards achieving them. They are also easy to set, driven either by management or historical data from the organizations themselves.
However, with the exception of financial targets, organizational target is a largely internal measure. It says little about how a company performs with regards to its larger community. This is where the other two types may help.
Standards exist as part of regulatory requirements or driven by specific groups in an industry to foster common measurements of performance. Those in heavily regulated industries, such as finance, legal, and telecommunications, would often have to adhere to measurements specified by the government.
This type of measure can be beyond the control of a particular company. Depending on the measurements, they may change quite frequently, making planning difficult and discouraging corporations from using them.
On the other hand, some standard measurements are not driven to force regulations, but is a way for people to gauge the average performance by others in the same measurement. Example includes unemployment rate – there is no industry standard, but published average unemployment rate for each country may help a country to gauge how well they perform compared to other countries. I call this the common average.
They can be used freely as organizational targets themselves, as baselines. Like the industry standard, corporations may have very few control on this type of measurements. However, both industry standard and common average are great for marketing and advertisement purposes, since most of them are public knowledge, and thus would put more credibility for corporations which meet those numbers.
Sometimes, corporations, especially established ones, would not want to compare with the standard averages for targets, since they are probably way ahead of the curve in terms of meeting standard measurements. In these cases, they would likely compare their performance to those most competitive to them.
Doing so would be arguably difficult, since company-specific data would likely to be confidential, and the public ones are far and few at best. However, knowing these numbers may help a corporation to be even more competitive.
…back to our example…
The sales numbers above can be improved by adding contextual data. In this example, the most viable type would an organizational target, and this is illustrated below as a before versus after comparison.
Enhancing contextual data further…
Just having targets and knowing if we meet them or not is often not enough as performance measures. Even if we do not meet a target, we probably would like to know how good or bad we did anyway. There are several ways to visualize this, and I discuss them below.
- Performance Band
- Quartile Band
I have discussed performance band several times before, notably in the post on my KPI dashboard, and the post on communicating time-series data. Typically, a performance band is divided into several different hues that depict different performance ranges. A typical example is the red, amber, and green bands of a key performance indicator (KPI), as illustrated in the bullet graph below.
In this instance, we can note that there is a target, represented by the grey bar near the amount $500,000. We also can note that the current performance has not meet the target and it is in the red band. Thus we can conclude that this measurement’s progress is very bad, and some major intervention may be required to improve it. The simple line chart provides some time-context to the measurement.
The quartile band is not necessarily used for gauging performance (although it can). It is used fairly often in social studies to represent distribution of a population. For instance, a baby’s weight can be tracked across time and plotted on a quartile band. This will let us know if which band the baby’s weight fall into, for example, the 25th quartile or 50th quartile. Either way, there is no good or bad performance, the baby’s weight just happen to fall within a particular quartile. The same example can be illustrated with salary scales. There is no good or bad salary, it is just what a company is willing to pay and it is up to individuals to gauge on a quartile band whether that amount commensurate with the job expectations.
…back to our example…
The sales numbers in our example way above can be improved by adding a band. In this example, the most viable type would a performance band, and this is illustrated below.
When not to use context…
There might be times when contextual data may not be required at all, but these are very few and far. When you deal with end users, contextual data is almost always needed to be used for the users to be able to make sense and relate to the data.
When you are just newly exploring a dataset, then maybe it is not yet a good idea to think contextually. I call this the data exploratory stage. In this stage, you are just exploring the patterns and trends of the data, slicing and dicing to get some meaning from it, and generating new views that would offer more perspective into the data. At this stage, contextual data is hardly required.
Even in the statistical analysis stage, which often precedes exploratory stage where hypothesis are formed, contextual data is hardly required. At this stage, you are concerned with the viability of the data as you test out the hypothesis.
Only when you have solid result would normally context come into the picture. Here is where you would link the hypothesis with real-world contextual data to form sense-making in the data.