SAN DIEGO -- According to one speaker at a recent analytics event, the industry is advancing past the problem of defining “big data” and on to an even bigger question: How can all this this fast-paced, multi-structured stuff be analyzed to produce useful information?
According to some, big data analytics will require
More on big data analytics
‘Big data’ analytics weighs a mix of the old and new
‘Big data’ analytics programs require tech savvy, business know-how
Worst practices in ‘big data’ analytics programs
“We’re going to go through a phase of analytics. What that’s going to teach us is the kind of discovery and routinization metaphors employed today -- data scientists playing in sandboxes and engaging in exploration -- start to collapse quickly,” he said. “We’re going to move on and we’re going to talk about decisions.”
Demarest admitted he regards big data through the lens of the machine and sensor data that inform a majority of his work, a perspective he said has been characterized as “skewed” and even “jaundiced.” Still, he recognizes that big data has complicated the business environment. Before integrating big data into an analytics or business intelligence program, Demarest suggests looking to the sensor market as a potential big data analytics guide.
The sensor industry, for example, tends to look before it leaps, Demarest said, by first asking what outcomes it's trying to achieve or change and then working backward to figure out the data needed to meet those goals.
Most businesses, on the other hand, tend to spend time and money building infrastructure to support routine decisions rather than complex or highly coordinated ones, Demarest said. That means pouring money into keeping the business running without much innovation. Big data, though, could mean big innovation.
“One of the beauties of the big data revolution is that the data is almost always available somewhere,” he said. “But when we don’t start by asking about outcomes, we don’t know what data we want.”
The sensor market is also more advanced when it comes to the step after analytics -- decision making.
For businesses today, Demarest said, making a decision is highly variant and subjective: Give five people the same data, tools and dashboards and ask them to make a decision; they’ll provide five different solutions.
Those five people will tend to follow a generic, cyclical model that includes data analysis and a projection of what the company may look like in the future. But when the process enters the collaboration phase, the model tends to break down. That erodes the rest of the cycle, which includes steps like monitoring the implementation of the decision and determining how the final outcome compares with the initial projection.
“The fact about this industry today is that the making of a decision is largely uninstrumented,” he said. “Propagation occurs in an uninstrumented environment.”
To understand how the decision-making process works, businesses need to look at email traffic, text messages and parse through PowerPoint presentations, Demarest said. Or they will need to take a chapter out of the sensor market’s book.
“As time compresses and complexity increases, what we’re going to have to do is figure out how to replace that smartest person, in many cases, with software,” he said.
In other words, automation. Sensors are built with this decision-making cycle, and, according to Demarast, businesses will eventually follow suit.
“Decision management -- how we use technology to make decisions -- is of course where our industry began,” Demarest said. “In the 1960s, smart guys at MIT said, ‘We have computers; we ought to be able to use them not to automate routine business processes, but to improve the quality of decision making.’ And it’s coming back.”
Doing so may mean embracing something businesses have tried to expunge from the enterprise for years : Decision management is not a layer like, for example, infrastructure or BI; it is a silo.
To get started, Demarest advised seminar attendees to scrutinize the data they’re already collecting and find decisions that are routine or algorithmic, take a long time to complete or are performed by only one person.
“And you want to find the ones that are closely connected to top-level financial metrics or KPIs [key performance indicators],” he said, “because this is all about determining outcomes.”
That way, businesses can monitor how the process is working and make the necessary adjustments.