“If you have a data scientist in your company, survey them and ask them what their chief pain point is or what prevents them from going faster,” Erik said.
“A very very high percentage will tell you that they’re spending a lot of time collecting, filtering, massaging, cleaning data, getting data ready for modeling, basically.”
“They’re spending 95 percent of their time getting the data, and only 5 percent of their time able to explore modeling on that data.”
“Our strategy is to really aim to automate that first part of the process — not just to make it more accessible for a process engineer who doesn’t necessarily have this deep data science expertise, but also to automate the drudgery that the data scientist has to go through to pre-process all that data,” Erik explained.
“We’ve developed a library of capabilities from 25 years of experience of things you need to be able to do to prepare data for modeling. We’re seeing that make a pretty big impact.”
“It’s usually this question of, ‘what inputs do I use in order to model this? Can you tell us what we need to include as the data that gets modeled?’” Erik said.
“It’s not just about wrangling the data. It’s about making the process of modeling robust enough so that if you don’t understand what the inputs should be, throw everything at it. The model should be robust enough to pick through what’s important and what’s not important. That is one of the most important enablers.”
GrayMatter’s Advanced Industrial Analytics Briefing decodes how companies can eliminate production variabilities that cost time and money.
Schedule a 90 minute one-on-one briefing to learn how to use historical data you already have to spot ways to improve product quality, optimize energy usage and enable predictive maintenance.