Progress, a provider of application development and deployment technologies, recently acquired DataRPM, a privately-held provider of cognitive predictive maintenance software for the industrial IoT (IIoT) market. Mark Troester, vice president of strategy at Progress, discussed how the addition of DataRPM’s predictive analytics and meta learning capabilities help to round out the Progress platform, enabling the company to embrace a “cognitive first” strategy.
What does “cognitive first” mean?
Mark Troester: When we look at the future of the business application we like to say that they need to be cognitive first, meaning that we need to break out of this role of having a transactional or operational system that is separate from where we actually do analytics and business intelligence. There has been a move to take analytics results and surface those within the application, but this goes beyond that.
How is this put into practice?
MT: When we are thinking about building applications, we leverage the outcomes of predictive analytics to drive the behavior of the application. Instead of inserting a dashboard, we have a predictive result that might drive the workflow. If you are expecting a failure in a machine and you can react to that, you can build processes around what you need to do to fix that machine, to notify people, and other things. For us, it is taking the predictive nature that you can benefit from with machine learning and integrating that into application development lifecycle, as well.
When you think of what Progress has been known for, what does DataRPM add that expands its mission?
MT: If you look at our portfolio and what we need to bring together to deliver on this strategy, there are a number of different things. When you build an application, you need to have some type of front end and today that is usually graphical and it is omni-channel in terms of being able to engage a user across multiple devices which extend beyond web and mobile to chatbots, artificial reality, and virtual reality. We have been in the market for a long time in terms of helping to build those front ends. Then the back end is also important where you might want to run business logic, integrate with different systems, and manage security. And we also have significant back-end technologies based on some of our mobile technologies which will be enhanced to support micro services and server-less technologies.
While we have had those two components and strength in terms of data integration with our DataDirect products where we can connect to any data source regardless of location, what we didn’t have was the cognitive component. We had this rich history and capabilities around AppDev but we were missing the machine learning, predictive analytics capability. The choice was to build it ourselves knowing that we were not experts in that field or to bring it to market more quickly by finding the right asset to buy. That led us on this journey to find the right technology to either buy or partner with, and ultimately led us to DataRPM.
Why did you select DataRPM?
MT: We wanted to buy a company that fit into our revenue structure in terms of what we could afford, that had made it through some of the initial hurdles in terms of going to market, had name-brand customers, and had proven technology. And, then we wanted to specifically look for a technology that made it much easier to manage the data science lifecycle or to manage the building of the analytical models. And, then we also wanted to look for a company that was a good match for our customers. As a development shop, we sell to both corporate accounts as well as ISVs who use our technology to build their applications and a large number of our ISVs are in the manufacturing space. While we wanted a horizontal predictive analytics capability, we also wanted something that would specifically help the manufacturing base and that took us down the path of predictive maintenance.
Which verticals are you targeting?
MT: The first will definitely be the predictive maintenance or industrial IoT—large organizations that are on the manufacturing side that have machines that they need to predict failure for and ensure that they do not go down. That is the vertical that DataRPM is in, as well.
At one point, they were trying to be more horizontal but they found that if they tailored their solution to a specific vertical that that was a better way for them to go to market, and that is again why they were good fit for us in terms of our ISVs selling into manufacturing. From there, it can be used in a wide variety of ways—whether that is healthcare, financial services, or others.
Can you explain the DataRPM concept of meta learning?
MT: When someone goes to use machine learning, the first thing they need to do is determine what they are trying to answer or predict, such as when a machine is going to fail or how to improve a patient outcome. They start by looking at the data that they have and the typical process involves a data scientist who is manually going through the data and creating different models. Then, they build and test the models and, over time, figure out which model, or model combinations, yields the best results in terms of prediction. That is a time- and labor-intensive process and requires data scientists that are in short supply. What DataRPM can do is take a meta learning approach where the algorithms are learning themselves. And, so instead of having a data scientist that is manually doing all this work, the models themselves are learning from each other.
How so?
MT: Let’s say they are trying to predict whether a washing machine is going to fail. Obviously, the washing machine failure is going to be based on a lot of different things—not just the machine but the environment, as well. And if they have machines running in different environments, they need to have the right predictive model and that is going to vary based on the environmental factors. DataRPM has this concept of a digital twin where there is a digital model that represents every individual washing machine. The secret sauce that DataRPM has is the ability to learn from one entity under control which increases the speed in building models and also increases the accuracy of the models themselves.
What other trends are you seeing in this space?
MT: There are macro trends that everyone is talking about: the accessibility of compute power is so much greater than it used to be, there is a huge amount of data which is easier to get to, and there are advancements in machine learning. Those three have come together. And, in terms of micro trends, there is first of all, the meta learning approach which helps to democratize machine learning because there are not that many organizations that can use it effectively because it is very expensive. There are the digital goliaths like eBay and GE that can do it, but it needs to be democratized. And then, there is the notion of cognitive first. It is great to have cognitive computing and machine learning, but I like to think of cognitive first in terms of application development to put that into practice so you don’t have analytics on the side. It should be integrated into everything which is just going to make business a lot more productive and drive more revenue.