Predictive Analytics: The Futurists' Formula

Uncovering Unseen Information

Predictive analytics can also be used to determine pricing thresholds for specific segments of customers, affecting a sales team's ability to cross-sell or upsell. Although there is no one perfect business environment that supports predictive analytics, some of the more compelling deployments come from companies with a wide product range, says Gareth Herschel, research director at Gartner.

Pete Eppele, vice president of product management for price optimization and sales effectiveness solution Zilliant, agrees. Within a high-volume sales environment, a wide assortment of products, the sheer number of sales reps, and variables in customer needs can put additional strain on a company. "There are so many decisions to be made and a number of selling circumstances for any given product—this is where science plays a big role, so people don't get lost in the complexities," Eppele says.

For example, a major manufacturer turned to Zilliant to increase its sales volume and to sell additional product categories to its existing customer base to improve customer retention rates. Using SalesMax, a proprietary purchase pattern profiling solution that identifies, quantifies, and then prioritizes leads in a CRM or sales force automation system, the company could analyze customer purchase data over the course of 12 months and receive alerts when account-specific opportunities arose.

Within 90 days, SalesMax unearthed $55 million worth of customer opportunities, and the company has since closed more than $6 million of that untapped revenue opportunity.

Similarly, VersionOne, a project management software company, started using Birst analytics to enrich its CRM system. Using the data it had gathered by running Birst on top of Salesforce.com, the company's sales team could better understand which deals were moving in its pipeline, which deals were marked as "closed-won" or "closed-loss," and which deals were getting adjusted, according to Richard Fuller, vice president of sales for VersionOne.

This year, the sales team will roll out an ideal sales pipeline based on the data analysis from Birst. Because Birst can uncover where there are gaps or where sales reps are stuck and which actions need to be applied to move them further down the sales pipeline, VersionOne will be able to determine the next best step during every stage of the pipeline.

No Perfect Model

Although there are clear benefits to running predictive models to improve business processes, plenty of companies are still dealing with data manually, observes Pankaj Kulshreshtha, senior vice president of analytics and research at Genpact Limited, a business process and technology management services firm.

For companies that have yet to deploy predictive analytics software, it's important to remember that predictive is built on probability. "Almost all models have a margin of error, but significant impact can be driven even with models that are eighty to ninety percent accurate," Kulshreshtha explains. The key, he says, is to create a test-and-learn environment to continuously measure the impact of a model and then to introduce changes to improve the model until a very low margin of error is attained.

Forrester's Gualtieri concurs. Because a predictive model is designed to determine the probability of an outcome, it's up to a business to decide which of those probabilities—such as altering customer messaging or tweaking a sales model a certain way—it will invest in. Because the models rely heavily on historical data, a company must "continually refine and rerun" test models to account for external forces like the economy, markets, and competitive movement. "The good thing about [predictive] software tools is they tell you how good the model is—in that snapshot of time," Gualtieri points out.

Experts offer a word of caution before running predictive models—know your business before you try to automate it, Trident Marketing's Brown says. Another important factor is good, old-fashioned instinct, Herschel says.

If a company has some hesitancy about whether or not it's acting on the proper indicators, it's best to use analytical models to double-check a company's instincts, he says. If the two are in agreement, "it is probably safe to go ahead."

Because the success of predictive analytics depends on the level of integration into a company's systems, as well as its richness of data, "some of the leading CRM companies will start to increase the amount of customer information they capture and embed predictive and descriptive analytics to find that nonobvious information," Gualtieri maintains. "There are all kinds of scenarios about a person's aspirations and their mood" that companies will want to tap into.

Associate Editor Kelly Liyakasa can be reached at kliyakasa@infotoday.com.