The field of advanced analytics, once the proprietary domain of companies willing to deploy costly enterprise software from the likes of SAS Institute and SPSS, now has an alternative from a startup, Apollo Data Technologies, which is offering predictive analytics solutions as a service.
The Apollo service offers its data warehousing expertise to join data sets and to clean data for the creation of data models.
Cleaning up data for purposes of designing predictive models is an undervalued expertise, according to Jeff Kaplan, principal and a co-founder at Apollo.
"We've worked with Fortune 10 companies and they all say their data is clean when actually it is a mess," Kaplan said.
Apollo uses Microsoft data mining algorithms to build predictive models based on a company's business goals.
The models Apollo builds, using standard industry algorithms plus seven new algorithms from Microsoft, can be used to help a company forecast everything from customer churn to inventory levels.
For example, in order to recapture revenue lost to online classified advertising, the Seattle Times newspaper engaged Apollo to uncover the subscribers and nonsubscribers most likely to advertise in the Times. Using the results from the analytics, the Times deployed a targeted marketing campaign to those identified as most likely to advertise.
According to Martin Schneider, an enterprise software analyst at The 451 Group, Apollo reaches a wider audience than companies such as SAS by commoditizing their code for reuse to keep costs down.
"If they have experience with the Seattle Times, for example, and they go to another news media company, they can reuse a lot of the code and lower the cost for both themselves and the customer," Schneider said.
The service initially creates a predictive model that is used only as a test to be sure that the design works and is scoring properly. Once the evaluation has shown itself to be successful, it is applied to operational systems.
"There's a transition where we turn over the solution to the client," Kaplan said.
At this point, the model will update itself as new data comes in. It also automatically scores new data and learns over time. At the point at which behavior in the population no longer fits the model, it needs to be "tweaked" again, Kaplan said.
Although pricing depends on the size of deployment, the company is heading toward value-based pricing.
"We will do a project for cost and for every dollar we save the customer we will share in that success," Kaplan said.