Analytics for IoT requires a sound strategy for success

Types of enterprise IoT analytics projects

Kerravala identified four types of analytics for IoT that can help companies get the most value from connected data: Reactive, the most commonly used, is thinking about what happened; diagnostic is thinking about why it happened; predictive is thinking about what could happen; and pre-emptive is figuring out what can be done to ensure that it doesn't happen.

But the insights gleaned from each type of analytics for IoT don't provide equal value.

Zeus Kerravala

"Three quarters of companies that I interviewed do reactive IoT analytics, and only 51% do pre-emptive, and then the other two are kind of in between," Kerravala said, noting that as organizations move from reactive to pre-emptive analytics, business value increases.

This means pre-emptive analytics is more valuable than predictive, which is more valuable than diagnostic, which is more valuable than reactive, he explained. Yet, while reactive analytics can be completed largely with people analytics, he said, companies performing pre-emptive analytics need algorithms based on machine learning because machines can connect the dots faster than people. "And when you connect those dots, you discover new insights and you're able to create a new business model or change operations or improve customer service or whatever -- so that's very important," he said.

Build a winning team to handle analytics for IoT

Another thing companies must consider is who should be involved in determining IoT analytics projects, Kerravala said. The companies that do it best are the ones where there's collaboration between engineers and data scientists, as data scientists often don't understand the technology, while engineers often don't understand data science.

Christian Renaud

Building an IoT analytics strategy also depends on the job role, said Christian Renaud, a research director at 451 Research. "If I'm in an IT organization and I'm thinking about an IoT project, I generally start in the data center," he said. "And I do what we would traditionally have called DCIM -- data center infrastructure management. That's environmental sensors in the data center and possibly some video analytics for security door access control. That's what the IT folks are generally starting with."

The operations department, on the other hand, is instrumenting HVAC systems, industrial equipment on the factory floor, fleet vehicles and delivery vehicles to track and get telemetry data to determine if equipment is functioning properly. "So it's different depending which constituency you speak with," he said. "They're instrumenting their environments. They're putting sensors and connectivity in to be able to extract data and then they'll figure out what's important and what's not on the analytics side. But they don't start with analytics. They start with connectivity and then take a look at the data."

Which IoT analytics tools should you choose?

Maureen Fleming

Looking at the data successfully brings up the topic of which tools are best to perform analytics for IoT. When it comes to determining which tools to use with specific initiatives, some organizations look to the various technologies they already have in-house, said Maureen Fleming, an analyst at IDC. "For instance, a company might have Splunk in the IT organization, so it pairs the Splunk people with the IoT people, and they do an experimental solution to get to success," she said.

In other cases, where it's a tactical rather than a strategic initiative -- for example, the company needs to achieve better performance -- the organization should look at a product that solves the particular problem rather than a big platform, she said.

Don't boil the ocean. Keep the project small and apply the right analytics tool to that one specific area.

However, choosing the best tools for enterprise IoT analytics projects isn't easy in a market filled with what Kerravala estimated to be at least a hundred different products. "There's data virtualization, there are some tools that are integrated into a platform, API management, messaging middleware, replication, extraction tools, data transformation tools, mobile app development platforms," he said. "The reality is, you're not going to have one tool that's going to solve all your needs, so you need to think about what it is that you're trying to do."

That's the hardest part of doing analytics for IoT -- understanding what needs to be analyzed, Kerravala said. "Don't boil the ocean. Keep the project small and apply the right analytics tool to that one specific area," he said. "Companies are connecting everything, literally everything. And some things need to be analyzed and some don't, depending on what it is you're trying to solve. But I would try and boil it down to one or two key performance indicators that are meaningful for a particular IoT project."