There are two interesting things about Predixion. First, it runs in the cloud, though it doesn’t have to. Secondly, it uses Excel though, again, it doesn’t have to.
But what does it do? Predixion is a data mining (and predictive analytics—see later) vendor. Currently the product supports association rules, decision and Gaussian trees, k-means clustering, linear and logistic regression, Naïve Bayes, neural networks, probabilistic clustering, regress trees and two different types of time series analysis. No doubt more algorithms will be added over the course of time but this looks a decent enough start.
I have to say that I looked askance at Predixion when I first heard about it. Excel for data mining? Mining in the cloud?
Let me take the second question first. To begin with there is an on-premise version of the product so if you don’t want to load your data into the cloud (and lots don’t) then you don’t have to. Actually, it wasn’t the question of wanting to that I was worried about, it was actually getting the data loaded but the company assures me that one of its early adopters loaded 9 million rows of retail transaction records in 1 minute.
Of course, these may be very short records (Predixion couldn’t tell me) but the numbers seem respectable enough, though they would not be adequate for, say, Wal-Mart but, hey, there are plenty of smaller retailers (well, all of them actually).
As far as Excel is concerned, the truth is that you couldn’t have done data mining with Excel before Microsoft introduced PowerPivot. Predixion is tightly integrated with this and it is, if you like, the secret sauce behind the product.
Now, if you think about it, what is valuable about employing Excel (for anything) is that it is amenable to use by business users. However, data mining isn’t something that you’d normally think of business users doing, so Predixion is building functionality on top of its algorithms that business users can sensibly deploy without understanding the guts of how data mining works. Things like shopping basket analysis or key influencer analysis.
Going beyond that, Predixion is also building a partner channel to deliver analytic applications. It is focusing initially on four main areas: retail (market basket analysis and customer profitability), healthcare (insurance/clinical fraud and disease prevention), financial services (customer retention and marketing) and hi-tech manufacturing (inventory optimisation and product quality).
It also has a couple of interesting technology partners (apart from Microsoft, which sees this as an opportunity to sell more SQL Server, Excel and SharePoint [which is used for publishing]) notably Composite Software, which allows it to get at a wide range of data sources (in combination) and Lyzasoft, which uses Predixion’s API to use Lyzasoft at the front-end rather than Excel, and Zementis, which enables Predixion customers to import and score PMML-based models created in tools such as those of SAS and SPSS (IBM)..
However, the real kicker, if you’re not interested enough already, is the price: $99 per user per month. Now that has serious disruptive potential.