The Center for Data Innovation spoke with Anna-Katrina Shedletsky, CEO of Instrumental, a U.S. firm which uses machine learning to help manufacturers improve their products and processes. Shedletsky discussed how mechanical engineers and manufacturers impacted by COVID-19 can work remotely and the obstacles to AI adoption in the U.S. manufacturing industry.
This interview has been edited.
Hodan Omaar: You had the idea for Instrumental whilst working on factory floors in China as a product designer for Apple. What problems were you trying to fix and how did machine learning and artificial intelligence help solve them?
Anna-Katrina Shedletsky: During development, engineers race to find and fix as many issues as possible to ensure a quality product is delivered on schedule. But the biggest bottleneck is discovering critical issues in the first place. The task of finding and fixing issues is so critical that companies typically fly engineers to factories around the world. I thought it was ridiculous that part of the engineering and development process was so reliant on luck: being in the right place at the right time. During my time as an engineer, I saw a lot of inefficiencies that seemed very solvable, if only we were more proactive about collecting and using data. I started Instrumental with the desire to lift up the discrete manufacturing industry to enable teams to build better with technology. Instrumental’s combination of real-time images and AI offer a perfect solution to this problem because it can give eyes into the factory from thousands of miles away and intelligently process data to surface defects and shifts in quality.
Omaar: In a recent Economist article about reopening factories after COVID-19 you were quoted saying that in electronics manufacturing “We’re going to do five years of innovating in the next 18 months.” With borders shut and engineers sheltering at home, how are you helping your clients work remotely and overcome the need to physically find problems on the line?
Shedletsky: For decades electronics brands have solved the problem of not knowing where the issues are going to be on the line by sending engineers to the factories in person. That is no longer possible and isn’t expected to be possible for some time. Our existing customers were the first to realize that Instrumental was going to be a core strategic asset for them in finishing their ongoing development processes and supporting production ramps, without onsite engineers, for the first time ever.
How are they doing that? Instrumental has always been about putting actionable data in the hands of the engineers and technicians who are most able to do something about it. Since 2015, a core part of Instrumental’s concept has been to aggregate this data in the cloud enabling our customers to log in from anywhere to see what is happening on their line. Before COVID-19, Instrumental’s core selling features was the technology’s ability to discover new issues and support rapid failure analysis in development and to reduce rework and improve yields in production. David, an engineer at Motorola Mobility, could sit in his office in Chicago and discover new quality issues or design issues on assembly lines 6,000 miles away.
Today, the ability to have remote oversight through images of every unit at key states of assembly and rich real-time analytics on defects and defect rates is enabling our customers to do much more than if they were just in the factory. Our algorithms use machine learning methods to discover new issues, pushing luck and travel out of the equation. They’re seeing significant efficiency increases in the acceleration of product maturity, reduction of rework, and increase of overall yield while also having a tool that enables them to communicate about issues easily with a globally distributed team all working on the same problems.
Omaar: What obstacles do you think the U.S. specifically faces in embracing AI in manufacturing and how important is overcoming these in staying competitive?
Shedletsky: AI as it relates to manufacturing and efficiency gains is nascent. No one country is particularly further ahead than any other, although all recognize that this kind of technology could be such a game changer that to not adopt it might lead to an inability to compete.
What’s taking so long in this adoption? People in the manufacturing industry are practical; when a technology has a positive return on investment, saves money, and enables something new it will be widely adopted. The reason that hasn’t happened yet for AI is twofold. First, a few years ago there was a big marketing bubble for “Industry 4.0” and “Industrial IoT”, powered by AI, data lakes, and digital twins—a buzzword soup that overpromised and underdelivered. Many companies rushed to invest but then felt burned when the technology did not deliver. This has had a chilling effect for the adoption of technologies that came after. Second, the first-generation of technologies that companies rushed to try didn’t work that well. But with continued innovation and development they have come a long way and can make a significant impact when deployed the right way. Instrumental is in this category of new technologies that can have a significant impact. We’ve gotten where we are by always focusing on our customers most painful problems and providing solid evidence that we’re having a significant and positive impact.
My hope is that COVID-19 wakes the industry up. There are now new challenges that need to be solved which will open the doors to innovations like cloud data, AI, and others which will make the industry better.
Omaar: There is a perception that full automation of the supply chain is on the horizon. How far along is automation in manufacturing in reality and what are your thoughts on how automation impacts society?
Shedletsky: Manufacturing makes up half of the world’s GDP, so it’s really quite a big and diverse collection of verticals which are all at various stages of automation. There are many industries that are incredibly far along; process manufacturing like chemicals, paint, food, and pharmaceuticals make use of a lot of automation. My home industry, electronics, is still extremely manual; hundreds of human hands make your phone or laptop or home IoT device.
Electronics products in general have really short lifecycles. Development is 6-8 months, followed by 6-12 months of production, and then replaced by the next generation product. Automation is not currently flexible enough to be easily retrained from one product to the next—definitely not in the way that humans can—and so there’s a long way to go before I expect full automation in electronics.
My personal perspective is that automation is a necessary part of the evolution of manufacturing. You wouldn’t know it from media in the U.S., but there is a labor shortage for manufacturing worldwide. We do not have enough people who want to do the jobs required to produce the things we all want and need every day. Automation is not about replacing people, it’s about elevating people from menial and repetitive tasks to positions of oversight, maintenance, and design where our brains and talents are powerful.
Omaar: What are some of the cutting-edge developments in the field of applied AI to manufacturing, and what are some of the leaders in the manufacturing space actually doing to build better?
Shedletsky: Frankly, most of the cutting-edge developments of technology in manufacturing are much more basic than AI. It’s things like leveraging cloud databases to aggregate data from multiple places in the supply chain and using basic math to transform that data into useful and actionable pieces that engineers can act on. We are seeing multiple companies tackle this across manufacturing for supply chain data, process data, and product data. It’s really an exciting time to be part of the industry.
As for what the most admired companies are doing to build better, they’re investing in smart capabilities and tooling that amplifies their teams. As an example, one Fortune 100 customer of ours is ramping a new product for the first time with no engineers on the ground and seeing multi-percentage point yield improvements from use of AI on their lines. Another Fortune 100 customer is partnering with Instrumental to aggregate performance test station data throughout their supply chain (including returns from customers) to answer challenging questions about how they can improve overall quality and customer experience. Both companies got to where they were by leveraging data in ways no one else in their spaces had considered before and they are going to keep their lead by doing the same for their manufacturing.