Blogs

The top 5 first-time SPSS Modeler implementation mistakes

Independent Consultant

Implementing a predictive analytics solution for the first time can be stressful. Having received approval for a significant investment in software and infrastructure, you want to make sure that you can produce value—and soon. If the investment includes IBM SPSS Modeler, then you have IBM and the IBM partner community ready to help—but even so, you might be unsure about what help you need, what help you want or even which team member to assign to get the help.

Moreover, as when doing anything else for the first time, you might fall prey to some common mistakes. Let’s look at several:

1. Planning no implementation at all

http://www.ibmbigdatahub.com/sites/default/files/noplanning.jpg

 

 

 

 

 

 

 

 

 

You might be tempted to give yourself some time to acclimate to the software—time in which to get your team “up to speed.” But SPSS Modeler models live to be deployed; deployment is their natural state. If you merely query a bit of your data, find an insight or two and leave things at that, you risk turning an incredible profit engine into a cost center. SPSS Modeler should be paying for itself within a year, not becoming just one more tool in your toolkit. If, content to merely look at your models, you fail to insert them into your business, then you aren’t getting full value from predictive analytics.

2. Working on your first project privately

http://www.ibmbigdatahub.com/sites/default/files/privately.jpg

 

 

 

 

 

 

 

 

 

Worrying that you won’t hit the ball out of the park on the first try is an understandable fear, but it brings real risk. Predictive analytics, done well, is transformative. And business transformation must be out in the open. Yes, your failure will have an audience—but you must take the risk nonetheless. By using powerful technology and relying on the right support, you can succeed. If you are paralyzed by fear that your first project will fail on the launch pad, then pause and take stock, making sure that you have chosen the right first project.

3. Rushing to modeling

http://www.ibmbigdatahub.com/sites/default/files/rushing.jpg

 

 

 

 

 

 

 

 

 

We all want to get to the exciting part of a project, but after building the actual model in a predictive analytics project, you are 80 percent finished. If you start running models within a day (or even a week), then you have rushed things. Can you be sure that you are even asking the right question? Have you chosen a suitably strategic first project? Did you consider a dozen or more other projects or project variants before choosing one—or did you race ahead too soon?

4. Not being ambitious enough

http://www.ibmbigdatahub.com/sites/default/files/notambitious.jpg

 

 

 

 

 

 

 

 

 

Have you been enticed by low-hanging fruit? Are you trying to get a quick win within the first few days, even one that brings negligible ROI, so that your colleagues will see the value in what you are trying to do? Are you considering whether an interesting idea, even if it isn’t a high-value one, might win you a little political capital within the organization? Such strategies might sound good, but they don’t usually work. Excellent models change the world. What’s more, excellent models are the ones that really get people’s attention. Keep brainstorming until you come up with a project that lights a fire under your team, and then make that project happen. The tools you are using were designed for nothing less.

5. Being too ambitious

http://www.ibmbigdatahub.com/sites/default/files/tooambitious.jpg

 

 

 

 

 

 

 

 

 

You want to be very ambitious—certainly. But you don’t want to try to boil the ocean. If you lack focus, you will produce a poor model. Imagine, for example, that a company wants to build a model to help identify its best leads. What if the company takes all the data, then tries to do lead generation, inbound sales and outbound sales simultaneously? Wouldn’t that simply mean ranking the best leads and the weakest leads? It sounds good, and it sounds easy. So why build three models? Because a single model won’t work.

Each of those three areas of the business has different data supporting it. If you try to mash all three areas together, the data won’t align well. With your data set looking like Swiss cheese, you’ll fall back to the one common denominator that all cases have—basic demographics—and end up with a bland, obvious model that fails to transform your business, or even to inform you of anything new and actionable.

Rather, focus on one project at a time. Concentrate on just lead generation, or on just existing outbound sales. You have scant information about one—though you have ways of dealing with that—and much more about the other. Regardless, tackling them individually will bring you better results than had you taken them on simultaneously.


Have you noticed what mistakes we haven’t discussed? You don’t see anything about “setting your neural net’s alpha too high” or “checking to see whether you can run nonnormal data through your support vector machine.” Why not? Because these very technical mistakes aren’t the pitfalls you should be avoiding first. Implementing a predictive analytics solution is all about proper planning and strategy.

Come to IBM Insight 2015, scheduled for 25–29 October in Las Vegas, to attend “Best practices: implement IBM SPSS Modeler and overcome shortages of budget, skill and time,” a talk hosted by Keith McCormick and Ram Himmatraopet. Keith is a consultant, author and speaker in the IBM SPSS community, and Ram is founder and CEO of Smarter Data, Inc., an IBM Analytics Business Partner that helps enterprises use data science, predictive analytics and big data software and cloud solutions to create business value and business differentiation. You’ll want to hear what Keith and Ram have to say as they provide a step-by-step solution to the challenges we have discussed.