Wednesday, July 11, 2012

Everybody LOVES statistics!

At least everyone I know loves statistics. Since my social life pretty much revolves around the people I work with -- people whose livelihoods revolve around what the numbers are doing -- "everyone" may be a small dataset, but still...

One of those guys is Jon "Denver," a Truck Buddy who stops by to "work" in the slow season every year. During the long winter downtime, we drank a lot of beer and made spreadsheets and databases. He was phenomenal at teasing useful, actionable data out of the seemingly infinite number of spreadsheet rows and columns. And one of those megaprojects he took on was a worksheet that gives us job time averages.

The immediate use for those was obvious: when we had a big enough dataset, we could start basing our estimates on that rather than do it "artistically," i.e. applying experience to every move request that comes in. While we still have an actual human (Dave) looking at each and every form and providing an estimate, we now have an extremely powerful tool to base our estimates on. The result is greater accuracy and faster response times.

With these statistics we learned a few things, and a few things we suspected were confirmed. For example, the top three types of jobs we do, by far, since we started aggressively tracking this information last December are:

1 BD to 1BD apartments. We've done this 210 times. The average time from start to finish is 3.05 hours.

1BD to 2BD apartments comes in second. We've done 121 of these since we started tracking the data. The average time: 3.32.

2BD to 2BD apartments. Instances: 101. Average completion time: 3.87 hours.

The stats we've been keeping also tell us how well the application of those stats is being done. For example, this shows the process could use a little work:

The breakdown:

In general, this is good. When we give estimates, we give a low and a high number. For example, a typical one-bedroom apartment might be bid at 2-4 hours. This chart shows, based on 535 examples JUST from what we call a "Standard Local Move," (which doesn't include POD/truck loads, deliveries, multiple-stop moves...), that:
  • We are under the HIGH end of the estimate 82 percent of the time. 
  • We are under the LOW end of the estimate 20 percent of the time.
  • We are right in between the HIGH and the LOW 62 percent of the time. (Ideally, this is where we want to be.)
  • We are over the HIGH end 18 percent of the time. (This has gotten higher due in part to an ever-widening dataset. When we first started collecting this information, our Over High statistic was awesome -- under 4 percent...)

I think this is respectable, but obviously we want to be in between the high and the low 100 PERCENT of the time. Ideally, we'd be able to plug all the factors into the MTB Estimatron 3000 and get a precise estimate every single time. However, that's impossible. Even if we could quantify all the variables, we'd still be off because there are always surprise factors (other movers hog the freight elevator, traffic patterns change, etc.)

Nonetheless, we can tighten it up a bit, which is what we're always working on.

Continuing trends:

In general, people "trade up" more often than not in the DC area. That is, they move from a smaller place to a larger place in one year.

If people don't trade up, more often than not, they move into a similar place. Not only do they more frequently move into the same STYLE of place (1BD, 2BD, etc.), they move into a place with similar features, such as three-floor walk-ups, similar views, basement apartments, etc.

This is just a small slice of the delicious, juicy data pie we've been cooking up. With our custom-made MTB Estimatron 3000, while still a work in progress, we can tell you, for example, how long a typical one bedroom apartment job with three floor walk-ups takes, or how many jobs we did in a certain zip code, etc., etc. Personally, I'm looking forward to the end of the year when we will have one full year of data collection completed. That's THOUSANDS of records just bursting with sweet, sweet data.

Damn. I'm hungry. Must be time for lunch.

No comments: