It Must All Equal Zero

A considerable part of my job is reading the tea leaves to determine why our market share is what it is and what it is likely to do next. We calculate our market share by submitting our sales and getting aggregated totals back. There are various details to this information, but to maintain confidentiality of contributing members’ specific information a lot of the details are deliberately hard to connect. As part of this deliberate ambiguity, we can tell what was produced at the factories and what was sold to customers, but we cannot link the two directly. For some purposes we report factory share and for some purposes we report customer share.

Throughout this year a gap has opened up between our factory share and our customer share. Ordinarily the difference is not hard to explain; units produced but not sold are generally in stock in someone’s inventory. It’s a simple concept and it more or less has to be true (although there is always the possibility of reporting errors, and there are several different kinds of inventory that may not easily follow the reporting guidelines).

I am aware of this straightforward explanation and made use of it in November and December. It’s getting somewhat threadbare now. It’s not really adequate to explain what’s been going on. So today I started checking into our numbers to see if there was something going wrong on our side.

The actual collection of our numbers and the hand-off with the trade association that compiles the numbers is not part of my responsibility. I generally make use of the numbers as provided. When I went diving into the numbers looking for problems I was excited to quickly find a discrepancy. I was almost sure I had found the explanation!

Problem was the discrepancy I found was a bit too large. When I checked with some other number-keepers they confidently told me that my numbers did not represent reality. But now I had to decide whether I had gathered my numbers incorrectly or the numbers were simply wrong. I thought of a way to test the numbers and just like that, I had my evidence that the numbers were wrong. Good stuff! I love finding an explanation!

But with acquired caution I sent my results to be verified by the official number-maker, and he promptly replied, explaining that the reason I was getting unbalanced results was that half the time our system stores sales orders as numbers and the other half as text. It’s bizarre, it’s pointless, it’s bad data management, but it is the way it is. Correcting for this oddity would reduce my imbalance.

Only it didn’t. After I transformed all the numbers to text to permit matching I increased the discrepancy, quite substantially. The discrepancy I now “uncovered” was larger than the original problem I started with. An investigation begun in the morning was now extending well into the afternoon.

I found that several other fields I thought were stable could actually change during the life of the order, including the identity of the customer. These changes were all reflected progressively in my record set, so that one order could appear several times. As anyone who works with databases knows, once a record appears more than once it tends to multiply like a rabbit. I had more than one rabbit. It took a while to hunt them all down.

By the time I got done sorting out all the incorrect relationships it was nearly quitting time. I checked my number one last time, noted it, and realized, like one waking from a dream, that I had lost all sense of context. I didn’t know what I was supposed to do next; I forgot what I was doing before I got sidetracked making sure I was doing it right.

It was one of the more interesting days I’ve had of late. I did some thinking, and then improved on the thinking. A satisfying day.