I interviewed Nikunj Mehta who discussed Why is it hard to find insights in operations & time series data?.
Today we are speaking with Nikunj Mehta, who is the founder and CEO of Falkonry. Falkonry does intelligent pattern recognition for operations effectiveness. We are going to be discussing why it's hard to find insights in operations and time series data. Nikunj, can you first provide a brief background of yourself?
I'm a long-term time series geek, and I have been helping companies gain insights in their operations data. And I'm also an enterprise software architect. For a long time, I've realized that the industrial world is where there is a lot of action in terms of interpreting data that they produce and understanding what it means.
Can you talk more about why you decided to do this kind of work?
I think it's evident, as see industrial IoT and a lot of these machine-learning buzzwords take place in the marketing world, that there is a lot of pent-up demand for extracting insights from operations data. Much of this data that is automatically collected tends to be of a time-series nature. For example, the way temperatures inside rooms change or the way in which electrical wiring varies in terms of the current it conducts over time. And operations is all about behaviors and behaviors that are all happening over time. So in order to understand how we can do better in our industrial and other production operations, we need to have a sense of what's going on as well as what's about to happen. And that requires understanding time series behaviors.
When I was at my previous employer, I saw a very large company struggle with making sense out of their time series data, and I realized then that the software world had done very little to actually assist these operations experts to interpret the data that they were producing. And I saw that there was a very broad opportunity to help them and realize that one of the best ways to help them was through starting a company, Falkonry.
Who cares most about this?
These are operations experts. Many of these go by titles like process engineers or manufacturing engineers. It's their job to gain improvements in operational efficiency, to continuously improve whatever it is that the production is about — whether it is power or widgets or internet service or, for that matter, even healthcare outcomes that people feel happy about.
Because they're constantly striving to improve operations, they're looking for ways to understand where those opportunities lie, and that requires that they gain new insights from wherever. It could be science or it could be data. One of the harder aspects of the improvement business is that after a while, there are no easy low-hanging-fruit type of opportunities left. All that's left is the hard stuff. And we think that in the industrial world, that's where people are at, that everything ahead of them is hard. And at the same time, they know that because it is relatively cheap to collect data, computation is so cheap, software is increasingly seen as an answer for a lot of problems, and most importantly, you pay as you go. So you don't have to make a huge investment up front. These people have been trying to figure out how can we benefit from it. So we're working with people in manufacturing, in industrial operations, transportation, and others, who have a lot of need to gain insights from their time series data.
Can you talk about how things have been changing and whether it's getting easier now?
I [inaudible 00:04:08] some of the general trends that are happening in and around computers and software. So certainly that's happening. Another important thing that has happened is software for analyzing data has become a lot more pervasive. And many techniques for analyzing software have emerged to a point where people see a likely solution for time series data as well.
Now there has also been a substantial crossover effort.People who would have traditionally been focused on selling business process improvement techniques, such as what is common in enterprise software and data, where I come from, have realized that that's a shrinking market and there are large emerging and growing markets in the industrial world that need a lot of software. And so there's been this sudden, new influx of talent going into the industrial world that is looking at time series data, [inaudible 00:05:08].
Secondly, techniques that have in the past been applied to small problems, such as analyzing the cardiac function time series data to understand heart health, are now, through better computational techniques, being applied to solve very complex industrial problems, like understanding what is really going on in a power drive used in a manufacturing plant and be causing it to shut down improperly. Those techniques were just hard to apply to large problems. But software experts, such as those at Falkonry, are able to figure out how to solve these more complex problems using the same type of basic math.So that is one major improvement that has happened in the recent past.
And then, of course, various people are approaching it many different ways. Some try to make it easy for people to express what they are looking for. Others extract insights directly from the data and help people interpret it visually and provide some feedback and subject matter expertise back into the computer so that the system overall gets better over time.
Do you have any final recommendations for operations experts?
I would suggest people, that they take a look at time series techniques, especially those that focus on subject-matter-expert usage. It is very hard to solve problems through developers and data scientists because there are just not enough of them. And it's hard to translate the knowledge of subject-matter experts to a form that data scientists and developers can understand. So when looking for insights in operations data, look for techniques that are subject-matter expert friendly. Number one.
Number two, look for techniques that will surface insights automatically, rather than one where you have to do a lot of setup and provide a lot of input.
And then thirdly, look for techniques where you can solve problems and pay only for the value you're able to get based on the problems that are solved rather than an upfront major investment in new frameworks and technology platforms.
So those are my recommendations. And Falkonry, of course, embodies all three of them. It is designed for subject matter experts. It does not require a lot of upfront work on the part of the subject matter expert. And you only pay for the value you get. So that sort of creates a win-win for these operations experts and their businesses.
Thanks for sharing today, Nikunj.
About Nikunj Mehta
Founder and CEO of Falkonry