# Measuring Sales by Kinds

Another day, another set of data... I've been investigating what I call "peripheral" data sets in order to get a different perspective on how previously unseen or unmeasured activity affects the overall operation of our company. In my last post, I looked at how our busyness could be represented by the volume of communication over our internal project management system from one month to the next. In looking at that picture, I realized that the volume of activity is much more drastically affected by maintenance work for our clients than by new projects. I classify "maintenance" as any work done for an existing client- it's a pretty broad spectrum, but since our new project process is so regimented, the split in categories is pretty realistic as far as our company's day to day experience is concerned. When I noticed that October of 2008 had the highest volume of communication, I wondered what our maintenance sales were that month and how they related to new project sales. Sales data is the easiest information for me to dig up, but I wasn't interested in the particular sales totals as much as the *relationship* between the numbers.

This brings me to the graph you see above. As I said, I wasn't so much interested in how much we sold from one month to the next as I was about the breakdown of sales- how much of it was new business and how much of it was maintenance. So, I determined the *percentage* of each month's sales total for the past few years that came from new projects and maintenance. For example, the graph above shows that in October, 2008, 37% of the month's sales total came from new project sales while 63% came from maintenance. No wonder we had so many posts to our project management system that month! As you can tell from glancing at the graph, this is a relatively infrequent occurrence; more often than not, the new project sales account for the majority of the total. When I first plotted the data, I didn't add the percentage values because I was more interested in the general relationship, as well as any trends that might be perceivable from visualizing the data. Again, glancing at the graph seems sufficient to conclude that there are no obvious patterns, nor an obvious trend in any direction (i.e. maintenance percentages trending upward or downward).

**Averages and Average Averages**

Then I wondered about averages. The data set covers three years, but it isn't three *full* years. Additionally, the current year has a couple of extreme cases (January, in which maintenance accounted for only 19% of the sales total, and September, in which new projects accounted for only 2%), so I decided to look only at 2008's average.

In the chart above, I plotted out a spectrum displaying only the percentages of total sales accounted for by maintenance sales-- the lowest, 24%, came in July of 2008, while the highest, 71%, came in August. Two concurrent months bookending the spectrum seems to clearly show that there isn't a seasonal correlation. But back to averages, the average maintenance sales percentage for 2008 was 41%. What's interesting about this is that 8 months out of 12 were less than or equal to the average, leaving only 4 months in 2008 that exceeded it. If I isolate 2007, the average maintenance percentage for the 7 months plotted is 42%. If I isolate 2010, the average maintenance percentage for the 10 months plotted so far is 38%. These numbers are pretty close together. In fact, only 11 months out of the plotted 29 had maintenance percentages that exceeded 41%, which is a "score" of 40%. Maybe there is some significance to 40%...?

Ultimately, I'd love to see the percentage of maintenance account for more consistently higher amount. I think doing more work for fewer clients is to our and our clients' advantage- it fits in with my motto of what we want to do: Serve fewer clients at a higher level. I believe we'll get there.

One last thing: The graph above doesn't show the *number* of new projects sold on a month to month basis. In 2007, the average was 4.1. In 2008, the average was 4. This year, the average so far is 2.6. To me, that's the kind of decrease I want to see. It means that we're selling fewer projects on a monthly basis this year, but at higher costs each (fewer at a higher level). So, all in all, one more piece of the puzzle...

Chris Butler| November 20, 2009 2:07 PMAgreed!

Nolan| November 20, 2009 1:56 PMNice. I think that would be an interesting data.

Looking at data on the whole is a great idea, especially in the depth. The nice thing about data is that it can either confirm or completely demolish gut feelings. I'm more of an analytical guy myself, so I love seeing how the numbers stack up.

And like we were discussing earlier, this kind of analysis will actually lead you to questions that you really want answered that may not be evident at the first. Gut feelings get you asking the questions, but analysis leads you to the right questions, and hopefully the right conclusions.

Chris Butler| November 20, 2009 1:18 PMNolan,

Agreed. One thing that became apparent to me during this exercise was not the percentages over time but the number of projects themselves. As I mentioned in the concluding paragraph, that is the trend that I didn't expect to see but did. This is the positive thing that comes out of explorative analysis like this- that what you expect to see (in this case, I was honestly looking for an upward trend in the percentage of sales from maintenance) is not what becomes the really important thing. In the end, what was important to me was seeing the number of projects that made up new project and maintenance sales decrease without taking the sales totals (in dollars) with it. That means serving fewer at a higher level.

Chris

Nolan| November 20, 2009 11:39 AMI think a useful metric from your data would to be to compare each of the month's percentage to the percentage to the overall year percentage, and not any averages.

There is a catch in taking the average of each of the percentages, since each month has a different statistical weight (like different number of projects, different prices, etc). For example, let's say January had 10x the projects as any other month and it also had ~40% maintenace sales. Looking at the year as a whole, January *should* dominate the other values. Even if every other month had 60% maintenance sales, if the number of projects was skewed enough towards a couple of months with a lower percentage, you still might have something like 45-50% maintenance sales for the entire year.

This concept is called Simpson's Paradox (http://en.wikipedia.org/wiki/Simpson's_paradox). There is a good example on that link--I'll summarize it.

Say you are accepting applications for a job. The first month you 500 applications and you accept 250 of them--that's 50% for month one. The second month you only get 10 applications and you accept one of them--that's 10%. Your average acceptance rate is not 30% (the average of 50 and 10), but a shade above 50% (251 accepted out of 500 applications).

So to sum up, I think a more useful metric would be the see how a month compares to a whole, and not as a comparison to each of their own individual, but there could be a unique situation where that might be useful.

Chris Butler| November 20, 2009 9:42 AMMaggie,

Glad you're enjoying the measurement series. As for help, what did you have in mind?

Chris

Maggie B| November 19, 2009 10:54 PMOooh. Should have withheld my comment for this post. Fascinating. We need this at our firm. Want to help?