Big Data and Business Analytics

Frontiers of Big Data Business Analytics • 63

modalities would be stable over time. In this way, one- time data errors

were diluted, and rule changes long ago were also less weighted.

In the end, we were able to predict customer and revenue numbers for

three to six months within a couple of percent. Time series models do have

the assumption that some level, trend, and periodicity continue over the

time window of prediction. Without the link between student- level infor-

mation and revenue, it would be more dicult to use this approach to cal-

culate the impact of student demographics and lead source information.

As we later found out, one of the issues was that some revenue from

online enrollments was credited to campus, as an incentive to increase ...