Analytics itself must be adaptive

Analytics itself must be adaptive

Summary: Business analytics models need constant refinement from data inputs to make it more reliable, as equations can easily go out of date, IBM researcher says.


Business analytics should itself be adaptive and regularly refined by new data that users feed back into the system as that is the whole purpose of predictive modeling, says IBM researcher, who adds that this will help companies better understand the data and anticipate future probabilities.

Using whatever model they formulate, business organizations have been doing forecasting for years, said Brenda Dietrich, vice president and CTO for business analytics at IBM. But the problem, she pointed out, was that people would build a forecast or predictive model and "use the same model and equations to analyze [whatever] new data over and over again, forever and ever"--until something changes unexpectedly or goes wrong, and by then, that model is no longer useful.

In today's era of big data, companies must regularly rebuild and refine their equations, because new history--and data--is created every day, Dietrich said in an interview here Monday.

"The things [businesses] are predicting are not physical constants. We're predicting price elasticity and customer sentiment. So why should we expect the forecast models that enable us to predict these things to not change?"

The IBM-er acknowledged that enterprises, like people, want definite numbers, and are uncomfortable with probability distribution. In other words, they get frustrated when they have to make a decision where they must "reason with uncertainty".

Business analytics will not give a specific answer but it can give the "variability of future outcomes" or an informed prediction of probable scenarios, she said.

Dietrich explained that data analysis allows companies to extrapolate outcomes linearly and decide what appropriate action to take next. Those actions also generate new data, which should be fed back into the analytics model so it is continuously refined, improved, and accurate, she said. For instance, analytics can show that there is a 95 percent probability that revenue earned will be between X and Y, and there is a 50 percent probability that the figure will be in a narrower range.

This notion of constantly "learning from the data" is a new and exciting development in the analytics space, because it means a company can see, as time progresses in reality, whether it is moving toward X or Y, and decide the next step it should take, she said.

In short, the constant feedback of new data ensures that the analytics system is not only "refreshed" constantly, but also keep tracks of how well business decisions--based on analyses--have worked out, Dietrich highlighted. And it is this second aspect that helps provide "confidence" for businesses to deal with uncertainty about the future, rather than make futile attempts to eliminate uncertainty altogether, she added.

Topics: Enterprise Software, Big Data, Enterprise 2.0

Jamie Yap

About Jamie Yap

Jamie writes about technology, business and the most obvious intersection of the two that is software. Other variegated topics include--in one form or other--cloud, Web 2.0, apps, data, analytics, mobile, services, and the three Es: enterprises, executives and entrepreneurs. In a previous life, she was a writer covering a different but equally serious business called show business.

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.


1 comment
Log in or register to join the discussion
  • Adaptive Business Analytics

    Models are not useful within several months after being released. New marketing products are now a must in nearly every industry and their main characteristic is that their structure is something new for Information Systems. Otherwise, they are not high quality marketing products... Information changes are exploding nowadays and model builders must calculate with consequences of changes in advance.
    If model is built without features for maintenance (data administration) then its value is highly questionable... what's the value of one time shot? There is a permanent need for continuous comparison of results...
    Article highlighted just one side of adaption, adaption in data flow from source systems for data analytic models. What happens understanding of changes in processes? Big data might or might not represent process changes correctly. Knowing what happens within production systems from the aspect of processes is the key to understand big data changes influence on models.
    At the end, models are becoming more and more just another bunch of beneficial data. We are lost in the sea of beneficial facts not understanding what is going on. Business analytics and their respective models should be factors of knowledge integration, not raw knowledge multiplication. Business analytics should be adaptive for knowledge integration: