As you’re probably aware, in my spare time I’m one of the people who help organise the SQLBits SQL Server conference in the UK. For the last few conferences I’ve worked on marketing, which basically means writing all those SQLBits emails that get sent out every week and thinking of ways to get ever more people to come along to the conference. Given that we handle all our registrations via RegOnline and can download all of the data we capture during the registration process I thought I ought to practice what I preach and do something BI-ish with it.
The first thing I did, of course, was to load all the data into PowerPivot. That worked well but, let’s be honest, you’ve already read 500 blog posts about loading data into PowerPivot and don’t want to read another one. Then I remembered the emails I’d been getting from Predixion, the data mining startup founded by amongst others Jamie MacLennan (who used to be on the SSAS data mining dev team), about their beta and thought I’d see what I could do with it.
Predixion have got a number of videos on their site to help you get started:
… and Kaspar has already blogged about what he’s done with Predixion twice (here and here), as has Mark Tabladillo, so I won’t repeat any of what they say and just concentrate on my results; suffice to say that anyone that used the free SSAS data mining addin will feel very at home with using Predixion Insight – in many respects it’s a much improved version 2.0 of that tool.
The most useful results I got were using the ‘Detect Categories’ tool on a table containing all my registration data to identify groups of attendees with similar profiles. For obvious reasons I can’t show you the raw data or anything derived from it that might identify individual attendees, but here’s a screenshot of the Excel report that got generated for me:
This report also shows data on the first cluster or category that the tool identified, and which I named ‘Hardcore SQLBits Fans’. You can see that these people are extremely likely to have registered very early: we opened registrations at the end of July, the ‘RegDate’ column shows the date they registered (DaysToConference shows the number of days from RegDate to the conference date), and this group was highly likely to have registered within two weeks of registration opening. Similarly they are very likely to be attending the full conference and not just individual days, and as a result are spending a lot of money with us. We’ve always known there’s a core group of people who come to SQLBits regardless of where or when it is, and this is that group – they’re our biggest fans and therefore we’ve got a duty to keep them happy!
Here’s another group, equally important for us, which I named ‘European Big Spenders’:
As you can see, the people in this group also spend a lot of money with us (perhaps more than the hardcore fans); registered slightly later, in mid-August; and are coming outside the UK. We made a special effort to attract more attendees from Europe this time so this is all good to see, but interestingly this group is very likely not to be using a discount code to get money off the price of the paid parts of the conference. I spent hours setting up discount codes for various European user groups and it doesn’t look like this strategy has paid off; maybe the fact that these people are coming is totally unconnected to any of the efforts we made?
Last of all, let’s look at the ‘Delayed Friday’ group:
The people here are only coming for the Friday and not attending the training day; consequently they’re spending less money overall. They registered in mid-to-late August and the discount codes used (which I’ve had to black out, sorry) suggest they heard about us from user groups, emails that our sponsors sent out to their customers, or places like SQL Server Central. They’re not such rabid fans of SQLBits that they would come automatically, but they’ve been successfully marketed to, seen what we’ve got to offer and registered. These are the attendees we’ve had to work hard to attract, and so the ones we need to understand best to ensure they turn into the hardcore SQLBits fans of the future!
So, as you can see, with one click of a button Predixion Insight came up with loads of interesting, useful information for me – information I probably wouldn’t have found if I was browsing through the data myself using SSAS or PowerPivot. And this is only a taste of what it can do; I didn’t even try out much of the more advanced functionality. I can imagine Predixion’s cloud-based data mining offering is going to generate a lot of interest, and maybe after many years the time has come for data mining for the masses?