IBM SPSS has been supporting in-database analytic modeling for a while now. We are trying to get some floating licenses out there to expand the use of it to a bunch of other people. Right now, we just have the fixed user licenses that we deployed. Sometimes it gets a little bit convoluted with the licensing. It has been fine on-premise, because we host a whole lot of IBM products. We have been using it for about six months, and we have been just playing with getting our models up and going, so we actually have the whole pure data and Hortonworks analytics products that we are going to be deploying in the analytics environment, that's where our server product will go, then we will have all of the governance pieces in place to start doing production deployment. In terms of models, we are getting it off the ground. We're doing real-time right now, but we are doing batch once we get the server product up and going. However, so far, it has worked pretty well. Also, we are getting licenses to do an SPSS server on the back-end, so as to offload some of the work off the desktop. We are a bit limited because we are using it on a desktop, but we are moving it into a server architecture so we can have a little bit more horsepower for it. We are trying to get into doing some identity resolution with it, so we are using the professional version (the higher version) with it. We have gone through some SPSS training, so primarily we take the data and figure out what they need to try to predict or what they are trying to figure out, then we use the tool to normalize the data, maybe doing some text analytics. I have a team of people that are working with the tool right now. We use it to try to do predictive modeling and data exploration. I don't like going to things where I can't get help, if I get stuck. I think knowing that there are a variety of partners out there with expertise in the product is a very strong selling point for me. Knowing that the product has had success for numerous other customers in the past for similar use cases, for similar types of customers. What's most important when selecting a vendor is the proven practice of the product. So that deal we got on the Gold package really sealed the deal for us. I think without Collaboration and Deployment Services it wouldn't have been a worthwhile investment for us and it would have failed on the deployment. The reasons we eventually chose this solution were that we were made a very good deal on the Gold package, which gave us more capability. And, as a brand new BI practice at a company that never had one before, I was just trying to build my practice from the ground up, and I didn't want to limit it to just BI reporting, so I took on the challenge of bringing in this new software, and staking my reputation on it, and it's paying off. I came to the World of Watson Conference in 2015, and when I saw SPSS Modeler and what it could do, I just sampled it, and it really, to me, spoke volumes about some of the inefficiencies in the way we were doing business. We're on-prem. I find the on-prem to be a pretty seamless experience, it flows directly from our data warehouse into the Analytics Server, and then we're able to deploy it back to the data warehouse for deployment into our CRM system. The association model tells the sales rep what product that customer should be buying, based on their sales purchase history. It generates pretty good sales.Īnd then we have a second model that does item recommendations, based on some association modeling. It gives the salesmen a ranking of which customers are their biggest opportunity on that day, and they just go down that list and call them. And then it scores the opportunity with that customer, based on how much money they spend with us. Each one is designed to do a customer recommendation, where it determines which customers should be ready to buy today, based on the recency of their last purchase, how frequently they purchase. We have separate models for our US call center and our UK call center. We're running batch, overnight, and I believe we have three machine-learning models in production at the moment. In a recent sample that I pulled, it successfully predicted two-thirds of our sales in a given week. The primary use case is to augment our sales processes, to help our call center determine which customers to call, which products to push to those customers.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |