The Apollo Data Technologies' principal describes how to use predictive analytics to create cross-sell and up-sell opportunities.

Marketers rack their brains every day to come up with innovative ways to expose their customers to targeted messages that ask them to buy their products and services. Now, along comes predictive analytics and predictive modeling, which is demonstrating, with great success, a new way to create and offer attractive cross-sell and up-sell recommendations to customers based on past purchases. 

For example, you're ready to checkout on Amazon.com when you are queued to a screen that says, "Customers who bought "The World is Flat" also purchased the book, "The Tipping Point." Another powerful retail application predicts what products and quantities to stock at each store across the chain to prevent out-of-stock inventory and maximize sales opportunities.

According to analyst firm IDC, "predictive analytic projects yield a median ROI of 145 percent." The great advantage of predictive analytics -- and what helps make its return on investment relatively high -- is that it can unlock value from the data companies have collected over the years, most of which has previously remained backlogged and of little use.

You may be asking, what exactly is predictive analytics and predictive modeling?

Predictive analytics and predictive modeling are automated ways to sift through massive amounts of product, customer, geo-demographic, behavioral and transactional data to identify vital information like which individuals have a propensity to purchase, what is the optimal price point, how much inventory to stock, what customers are at risk of churning, what is the next best offer and much more. 

Predictive models are built from past transactions, such as purchases, and factor in consumer behavior collected from demographic data, website traffic and keywords searched. Integrating these data sources provides a richer set of variables to model and determines the best way to target market to each segment and/or individual.

Today's predictive analytic solutions go well beyond the traditional difficult-to-use, off-the-shelf statistical software packages-- which have fallen short on delivering the analysis marketers need to deliver highly effective target marketing, sales and inventory forecasting, retention modeling and more. 

Moreover, traditional statistical software packages cost hundreds of thousands to millions of dollars and months of time to see results. Add in additional fees for internal resources, outside consultants to build and deliver the analysis, as well as annual recurring software maintenance fees and you can see why predictive analytic solution providers are gaining large mindshare.

Have I got you interested? Good!

Now that you know what predictive analytics and predictive modeling can do, here are five things you need to be aware of to implement a successful predictive analytics and modeling solution:

1. Secure commitment across the organization
Get on the same page. Key departments -- marketing, merchandising, IT -- must collaborate to ensure consistency and quality while gathering data, interpreting models, executing marketing campaigns and integrating predictive models with the marketing database.

2. Build a marketing database
Buy-in complete? Now you need a place to store your customer data, which includes both existing and new customers and products. A marketing database is essential to update and rescore the predictive models for new customers and products.

3. Schedule iterative model reviews
Iterative model reviews are crucial. Too often, analysis projects are a recipe for unused shelf-ware: your in-house or outsourced tech guru is dispatched to build a complex statistical model and deliver it "complete" to the marketing team. Marketers should be in at the start to define the sales and marketing requirements, contribute company specific business logic and help determine what variables should drive the model for the most useful predictions. Weekly or bi-weekly model reviews will keep everyone focused and the project on track.

4. Test the model
Before you go any further, test the predictive models on a sample set from your marketing database. Testing lets you evaluate the model's performance and identify ways to tweak it for greater accuracy before "going live" with your entire customer base.

5. Automate model updating
Through the marketing database (see Step 2), instruct the predictive models to run on a pre-established schedule. The benefits? Minimal maintenance, regularly refreshed customer and sales data, and updated predictors so that each time you target new prospects you are taking advantage of the most recent data-- and best intelligence.

Jeff Kaplan is principal, client services at Apollo Data Technologies, the first company delivering true predictive analytic solutions for key vertical markets. Kaplan leads sales and marketing at Apollo, helping clients use data mining as a strategic discipline for growing revenue and maximizing profits. Prior to co-founding Apollo Data Technologies, he managed product marketing and planning for Revenue Science, formerly digiMine, where he incubated new product entries, including web analytic and CRM applications. In addition, Kaplan led software implementations globally for Fortune 100 companies. Prior to Revenue Science, he was a global financial and planning analyst for several divisions of GE Capital, responsible for profit and loss reporting.