Datamation content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
Big Data has had its struggles in recent months as companies experimenting with it have come to an unpleasant realization: they don’t need it. They experimented with it because it was the hot new buzzword but then realized they didn’t have a need or use for it.
However, it could be they just didn’t apply it in the right way. Predictive Analytics, hardly a new concept, gets a big boost when applied to Big Data and can benefit virtually any company or organization in at least one way, argues one analyst.
“Every industry can benefit from predictive analytics, it’s just a question of whether they have the data on hand to make these predictions,” said Rowan Curran, a researcher with Forrester Research.
“There is no company that could not take their customer journey or developer roadmap and not take a [look at a] point where they could take information and predict an outcome to better improve their business,” he added.
Predictive analytics is just what the name implies: looking at existing data to make a prediction on a future outcome. Companies have done this for years to a lesser degree. It’s usually applied to sales, like predicting when people are most likely to buy and where. Your local supermarket’s layout is not random. It has been carefully designed through analytics to maximize purchases in every way possible. And politicians slice data eight ways from Sunday to predict voter turnout.
But Big Data takes things to a whole new level. Curran notes several new types of applications of Big Data-driven Predictive Analytics:
* Software company use it in the quality control process to identify which builds are likely to fail, so the development teams will get an alert saying you did this test in a certain way or missed this step or need this layer of quality control to make sure the build wont fail.
* BMW is using analytics to identify potential defects in car design before they hit the road. Also, they identify trends in their industry to decide which features to include in the next generation of cars. One example: carmakers are mulling the removal of AM/FM radios because no one listens to terrestrial radio in their car any more. They either listen to satellite or streaming services.
* Oil and gas companies use analytics to find out when a well is going to fail based on past data.
* HR departments are using it to identify top performers in their company and apply those traits to hiring, so they attempt to predict who of their candidates will be best performers.
“So folks are using Predictive Analytics all over the place for a plethora of reasons,” said Curran. “And because the barrier to entry is lowering, when one company sees a competitor is using analytics, they often want to get in on it as well.”
How Does It Work?
Machine learning algorithms have been around for 40 or 50 years to identify patterns in data to find correlation and probability of outcomes, said Dave Elkington, CEO of InsideSales.com.
“Some think you are gaming the consumer but it’s the opposite. It’s about personalization. Amazon has built a personalized experience for a buyer,” said Elkington.
InsideSales.com serves 3,000 clients with 60,000 sales people and helps them sell better by anonymizing and aggregating data across all sales people and learning best practices as different sales people interact with different profiles of customers. “None of those 60,000 sales people will make the same mistake twice because there’s the learning from the mass, from everyone working together,” he explained.
He said the reason predictive analytics has become popular is because it’s the next evolution in enterprise software, evolving from mainframe and then client/server and then SaaS.
“SaaS consolidated data into massive data centers and massive distribution on the client side. You could access the data anywhere in the world. This begged a problem: What to do with the data,” said Elkington. Predictive analytics consumes and provides value from that data that was sitting around, unusable in its current form.
And usage of that data is cross-departmental, said Florian Douetteau, CEO and co-founder of Dataiku, which develops a predictive analytics platform called Data Science Studio. Every group in a company can use data from their own group or from others. “It is used in sales, marketing, operations, logistics, and R&D,” he said.
It Starts With Big Data
The basis for good predictive analytics is a good Big Data project, because that will collect all the data points necessary for making accurate predictions. “It’s like going from anecdotal evidence to quantitative evidence,” said Curran.
To take advantage of Predictive Analytics you need a good working basis with strong data management and a very effective data cleaning process. That’s where most data scientists spend most of their time today. That base layer of preparing the data is key to enabling higher level analytics on top of them.
“Predictive analytics is something that does require some human intervention at the beginning. Typically you need between 2 to 6 weeks to setup a predictive model on a significant business problem,” Douetteau said.
The good news is that price has come down, thanks to new technologies and frameworks that simplifies things and enable non-data scientists to apply PA by themselves. “I think that projects that took a few months five years ago, now take a few weeks,” said Douetteau. He noted Stata, SAS and R as three prominent platforms for data preparation.
There are other options as well, according to Curran. These include packaged apps like HR apps to predict turnover of employees or to identify the best among new hires; pre-sales and marketing apps for sales people; and inventories and special offers from Web sites.
“These tend to be point solutions oriented toward one step of the journey. But then you have generalized platforms, like SPSS, SAS, Rapidminer, Knime, Alteryx, Oracle, and Predixion. These are generalized platforms with tools to bring your data sets in, munge them a bit, clean them up and apply a whole host of predictive algorithms,” he said.
There is also a lot of work being done in the pre-processing part. “We now have massive amounts of data and in structures that are unusable, so we’re seeing innovation around algorithms to use that data,” said Elkington. “Our core theme is we believe science holds the key to unlocking human potential. This data science is the science that seems to matter most.”
Fine-grain Big Data
A big trend is to try to make decisions on a very granular level, he added. “That’s where the competitive advantage is today. Granular level means for each customer, and even each customer journey if it’s about your Website. It means street level, hour level if you’re talking transportation. You can’t just take broad decision for a whole population or a whole city anymore in order to be competitive,” Douetteau said.
The fine grain approach is the next natural step you take after having big data. “First you look at the data and establish metrics. This lead to macro-decision and data-driven strategies and that’s already great. Second you try to use this data to operate at the granular level,” he said.
The way PA works is rather complicated, so this data massaging is important. A PA system takes historical data from CRM system, apps, Website logs and merges them together. Then an algorithm is used to “train” the system on that data. The result of this operation is called a “model” that can be used on new data in order to predict future behavior. For that reason, the data needs a lot of preparation beforehand to work with the model generated.
The Future
Douetteau describes the current market for predictive analytics products as “technoslavia,” a reference to Yugoslavia and how it disintegrated into multiple nations in the 1990s. Over the next year, he believes serious contenders for unification will arise.
“In terms of apps, I thing that in the upcoming months, more and more apps will provide to users a feeling of what predictive means, similar to weather forecast that shows rain probabilities, more and more consumers apps will provide insights in this way. As users, we might need to accept a world of uncertainties,” he said.
Curran thinks the next step will include faster results. “The next stage people are focusing on is getting predictions faster and getting them to humans or putting into action to change a customer service faster. So moving from informational to taking real actions and supporting the person understanding the analytics,” he said.
Photo courtesy of Shutterstock.
