Anna Katrina Shedletsky is co-founder and CEO of Instrumental. Her problem: How do you make electronics manufacturing more efficient and less wasteful?
Anna started her career as a design engineer at Apple. It was her job to visit the factory when a new device was about to go into production and try to figure out all of the potential manufacturing problems that might arise.
She realized this was an almost impossible task that relied on hope and luck -- and that it led to an incredibly inefficient and wasteful manufacturing process.
So she started a new company, Instrumental, to try to come up with a better way to figure out what's likely to go wrong, and how to fix it.
If you’d like to keep up with the most recent news from this and other Pushkin podcasts, be sure to subscribe to our email list.
Pushkin. I'm Jacob Goldstein and this is What's Your Problem, the show where entrepreneurs and engineers talk about how they're going to change the world once they solve a few problems. My guest today is Anna Katrina d Letski. She's a mechanical engineer who started her career at Apple, where she worked on the Apple Watch that was responsible for we call it packing the suitcase. So the industrial design team makes beautiful surfaces of like what it's going to look like on the outside, and the product design engineers figure out how to get all the parts inside that suitcase. If you will, you got to get all the parts in, and then you know that product needs to be robust. When Anna was working on the watch, she saw a massive problem, not just with the watch or even just with Apple, but with all of electronics manufacturing. The problem is this electronics manufacturing is incredibly inefficient, incredibly wasteful. And I'm not talking about consumers throwing devices away. I'm talking about the manufacturing process itself, what happens before the product even leaves the factory. So Anna left Apple and started a company called Instrumental to try to solve that problem in our conversation. Anna laid out the problem really clearly when she explained what happens when a new device is about to go into production. It starts with a sort of practice build, when a few hundred or a few thousand devices are assembled as a test. If you're part of a team of engineers working in your cushy office in Kupertino, this is the moment when you and your team have to go and actually see this thing you've designed being built. You'd all get on a flight and you'd fly in earlatest China. It's where a lot of stuff is built, although today maybe it's somewhere else in Greater Asia. And then you land in a city, and then you'd get into a factory van and you might drive for an hour and a half or two hours out into a very rural part of China, and that's where you'd have a massive manufacturing center. And then we would be there for about ten or fourteen days. What is the point of you being there. I'm there trying to find issues, find problems, find parts that aren't fitting, find quality issues with parts. So like we expect there to be problems and quality issues with the parts, Anna said, she's not allowed to talk about specific preproduction problems they had with the Apple Watch, but she did give me a hypothetical source of problems the antenna. So, okay, picture of factory with hundreds of workers packing all these tiny little parts into the watch. One of those parts is the antenna, this tiny little thing the sizes of like a one inch piece of dried spaghetti, and it has to go in there just right so that the watch can talk to the world. If the antenna isn't just right, the watch doesn't work. Yeah, So let's say hypothetically, you at the end of the line, you have a pile of units that the antennas don't work. The antenna's really challenging. It's like near a lot of metal and antenna's a metal don't don't mix, and so the antennas don't work. What you have as an engineer is test station results that say this signal is low. It could be an eighty percent tissue, like eighty eight out of ten don't work. It could be a you know, point one percent issue. You only have one that doesn't work. It made a thousand and nine hundred and ninety nine of them work, But what's wrong with this one? Because if you make a million, what's wrong with that one? If you make a million, one in a thousand is a big problem, right, yes, exactly exactly, And so in development you care about every failure. So, okay, so you have this problem the antenna doesn't work whatever often or once in a while, what do you do? We would gather up those physical units, which is non trivial to get all those units in one place in this big factory, and then we would very carefully start to rip them apart to try to get more data about what's wrong with these units? Why don't they work? Now. One of the things that's interesting about antenna specifically and Watch, if you've ever seen a teardown of the watch, is when you're taking it apart, you could very easily be destroying the evidence of the thing you're trying to find. Oh, like you just budge the antenna a tiny little millimeter or whatever, and that matter, yes, and it matters. And so you might try to take some X ray images before you disassemble, so you have some idea literally there you have an X ray camera, yes, which is like a three dimensional X ray, And then you take them apart, and you'd very carefully try to figure out what was going wrong. And sometimes those things are obvious. You open it up, it has no antenna in it, So that's why the antenna doesn't where. Sometimes it's very obvious. Sometimes it's not obvious, and like, as a skilled engineer, you're looking at it, You're like, I have no idea, And then you start looking at the process. You go, you walk, you know, fifty meters up the line to look at where they're putting the antenna's in, and you're watching that process and trying to see is there something happening here? Are there like are there screwdrivers swinging across the line that are like going to hit the parts and knock the antenna? I mean, I feel like there's sort of two layers of problems here. Right. There's the fundamental problem of, oh, we manufactured a million of these things and we know ten thousand of them are going to break, right. That is just a straight up problem. There is also the like engineers like you have to go fly to China, you know, stand on an assembly line and just hope they catch things, hope they figure out where the problem is. Not even they're looking for problems, they don't even know their problems are yet, Like I would go to a build and it would be a successful build if we left the build with one hundred issues in a spreadsheet of different things that we needed to fix. So that was the success. Like, the more problems you find, the better. I mean, it's a success because you found it because you know the problems are there, and you're just worried that you're not going to find. Yes, that's like your real nightmare problem if you're like a design engineer, right, the one that like you don't know is a problem until you've shipped a million pairs of earbuds or something. The technical term for that is an escape. So that's a failure that has escaped the problem escaped the factory. Yes, yes, and so escapes are what essentially then cause returns and if we go back to that waste, yeah, which is hugely costly. Right, it's bad for the reputation of the manufacturer because it's like, oh, it's a crappy product, bad reviews, not a reliable problem Amazon, bad reviews, and you've got to like refund the money, send the person a new one. So it's costly on many levels. Yes, yes, and so returns should be avoided, and you avoid returns by reducing escapes, and you reduce escapes not only by figuring out all the things that you can possibly test for. And so that's why there are testations on the line, but the testation can't test for everything. Did you just live in fear all the time of missing some terrible problem? I don't even know how to answer that. I think that a good engineer is a paranoid engineer. So for me, as an engineer, I thought it was kind of silly that there is so much luck involved. You just have to get lucky and catch the problem before it goes into production. Luck like that, yes, and so you're trying to find you're relying on luck to find things that you can improve in development. And then there's also like the scramble and the heroism that happens when you do have a problem that everybody learns about. Oh, we have this problem. So then you have this fired drill where everybody flies in tries to figure out what's going on. There's tons of freneticism and activity, and this just seemed like a waste, Like why is it that we're so reactive versus proactive? Why is it that like I don't have access to the data that I need as an engineer to actually proactively solve these problems. I have to go and like hope versus No. After the break, Annah launches Instrumental, a company that's trying to help engineers solve the problem of finding problems. Help him go from hope to know. That's the end of the ads. Now we're going back to the show, So just a reset here. Anna has landed a job at Apple, the company where engineers dream of working, this amazing technology company, and yet she decides to leave. So I asked her in the end, what made you quit your job and go start this new company, Instrumental. I can share some thing that that you can use, but it's maybe it might be a little sideways to what you're talking about, but it's the true reason I started the company. So about a year before Apple Watch shipped and the world knew that it even existed, I actually had a personal tragedy happened in my life. My husband was killed by a drunk driver. Oh my god, and I was twenty seven, he was twenty seven. And I realized that life is precious and limited, and you don't know when your last day is going to be, and so you want to be proud of the days that you have and how you spent them and the impact you could have on others, and also reevaluate the justification for my existence on this planet. And I decided that Apple is great. I'm glad I was there, but it can't be my life's work for me. I felt like I needed to do something bigger. And the big problem that we all need to solve is how we build stuff is so wasteful. That is a more meaningful direction. And there is a big problem that I know that I have some intuition about how to go solve, And so I took what I understood as like, wow, it's like really hard to find and fix these issues. We're wasting so much money, so much physical stuff. We are pumping so much chemicals into our rivers, we are burning so much energy. We are wasting human lifetimes of time doing things that don't matter. It's just wasted, and we need to figure that out. We need to change how we build, We need to change how we think about building products. So this is like a little maybe sappy for your audience, but this is the true reason that like really made the change of thinking about away from thinking about, oh, my job as an engineer, I make stuff too. Oh, my job as a human is to figure out how to do the things we need to do better. So Anna decides she's going to try to make electronics manufacturing better, more efficient, less wasteful. And to do this, she and an engineer she worked with at Apple decide to start a company. They call it Instrumental, And when it started, they had one big idea. Instead of forcing engineers like Anna to frantically run up and down the assembly line hoping they'd spotted every potential problem, they would put dozens of cameras up and down the line to capture the entire assembly process for every single device. It's the idea that there is value in this data that today does not get collected, and so the actual core pieces, Let's go get all the data. That's the that's the core innovation here. Let's go get all the data. The data in this case is visual data. What does everything look like as this phone or this watch or whatever is getting put together? Is that right? That was the first data set that we went after was the image data set, because there isn't one. So we're creating a data set that didn't exist before. And as you know, as they say, a picture is worth a thousand words. When you take a picture, you don't need to know in advance what you're going to be looking for in the picture. It's a very high resolution data set, and then you can come back with twenty twenty hindsight and look at the picture and be like, oh, like, here's a problem in that picture. And so that was the initial concept of like, we need to find where these issues are, which means we need to see them. We work in the physical world, so we need to see them. Remember I was talking about the antennas, where you had to carefully take them apart, and as you took them apart, you'd sometimes damage the evidence you're looking for. Well, if you have images of them as they went together, then you don't maybe have to take them apart, or you maybe have additional data about what they looked like as they were going together. But you don't know which ones are going to be failures in advance, so you take pictures of all of them and then you have the ones after the fact that ended up being failures. You have those images. So that was like the first idea. So the basic ideas. You take pictures of the whole thing, and you identify, you know, whatever, the serial number on each say, watch each device that's going down the line, and then if there's one device that doesn't work, you say, show me all the pictures of that one, and then you say, how is that one different than all the ones that do work in the pictures? Yes, yes, But it's not just about figuring out what's wrong once you find a problem, right is It also about getting better at finding problems in the first place. So discovery, we're able to solve the discovery problem for our customer. We built algorithms that look for anomalies, and so what's an anomaly. An anomaly is something that's different from the other ones. So if you build a population of units, they should all look the same, Okay, a bunch of devices, right, The ones that look different are probably interesting, and so we build algorithms to find the ones that look different. And that was kind of the first offering that we provided, And so our customers were able to use that combination of the photo record with the algorithms that would highlight automatically the things that were different and interesting, and they were able to use that to find issues. Basically, one of these things is not like the other, yes, exactly, So like, oh, this one's bent a little bit, this one's the parts supposed to be black and it's white. Is that a problem? Maybe not? Maybe we highlight those things and then the next piece is okay, well what's causing like problems? Okay, so you've built a system that is better at identifying for finding problems. So now even if a device pass all the tests, your system will say, well, you want to check out this one missing screws because it just looks a little different missing screws. Okay, so yeah, like that missing screws, there's no screw tester. You're right, like missing screws. Devices will actually miss screws. I'm so naive that I'm like, well, surely a device wouldn't be missing a screw, but one hundred percent, it's like one of the top ten defects in production is missing or extra parts, extra screw just to make up for the other one that's missing a screw. On average, they have the right number of screws. Yes, on average they have the right number of scares. So we need to be able to find issues. We need to be able to help engineers fix them, and then we need to help them make sure they don't come back. Those are kind of the three things that we have to do. So okay, those are the three key things Anna's company does to solve this big problem and to tile those pieces together. She gave me a case study, a case study from a company that was trying to build a webcam and the company was having trouble with the antenna. The antenna, which, as Anna told me about earlier in the interview, is this crucial part that is surprisingly hard to get right. They always fail. There are very sensitive parts, and all of our connected devices really rely on their antenna performance. So you don't think about it, but you know it's an important part. And so we had a customer who had failing antennas and they were able to take that group of that population of units and they were able to actively, like root cause it to three different things. And so this is actually the next piece of the puzzle, which is we need to find root cause. And so in this case, the customer was able to see a couple of different things. The first thing they found is a group of those failures was actually had to do with the alignment of a connector that was connecting the antenna. So the connector, if it was like shifted a little bit this way, it was good, and it was shifted a little bit the other way, it was bad. And this was very subtle. This is not something they had like a specification. This is a discovery. Okay, so the angle mattered and they didn't know that in advance. Okay, Now this is also maybe too much in the weeds. But there are a lot of reasons, and antenna can fail in a lot of different ways. So it's not just like, oh, if they're failing, they're all the same root costs. Actually, there could be a group of different causes that cause failures. So you have two different phones from the same build, both of which have the antenna that doesn't work. It might be a different reason for each phone that the antenna doesn't work, which makes it harder to solve. So you solve one problem, but then some of the phones still don't work. Is that what you're saying, Yes, yes, that's very common. They also found that the software version that they were using on the tester was also correlated to the failures, so most of the failures were from one particular tester, So in fact, it wasn't a bad antenna. It was a test for some of those units, not all of them. Some of them still had the connector coming out at a wrong angle. Yeah yeah. And then the last one on the same group of failures, in the same group of failures, there was another group of units where what the image based algorithms found as a high correlation to the color of the circuit board. And so color doesn't affect performance, but it was a different vendor. One circuit board was like a slightly different shade of a color than another, and so that means like the vendor of that circuit board was for one of the vendors that made that circuit board was more likely to have failing in it. So that gives engineers then three different things. They can go chase down. They can go chase down a process is year around the angle of the connector. They can go chase down like the drifting test station and work on calibrating the testations. And then they can go and investigate if there's something meaningfully different between these two vendors of circuit boards. So they're able to continue building, but they have now three specific things they can go do, whereas without instrumental they would kind of be guessing. So, like, what's the sort of frontier for you? What's the thing you're trying to do that you haven't quite figured out yet, Like, what's the next problem to solve? Yeah, I mean so it goes back to the reason for being for the company, the reason for being being that twenty cents of every dollar that's wasted in manufacturing, we haven't solved that problem yet. What does that require? We need to change how we think about how we build things, how we design things, the process of iterating through that development process, the process of what happens when it goes into production, what happens when you return units? And what do we do with that information from your return as a consumer. And today we're really not doing that as an industry. Not gathering data in a meaningful way is that we're not gathering data, we're not thinking proactively. As an engineer. If I'm designing something, I'm thinking about, you know, the next build and production, but I'm not I'm not necessarily thinking about, like how can I develop this product's that I can capture the most data from it the fastest, the earlier in the process versus later in the product. Like everybody knows their data is valuable and there's stuff in there that could be valuable, but they don't know how to value it because the problems haven't happened yet. When they're on fire quote unquote and they are spending a million dollars a month on returns, they know exactly what the value is in the data that they don't have, Yes, but when they haven't started the program yet and they don't know what fires they're going to have, they don't know how to value having the data. And like preventing those problems in the first place, the vision is that that data is enough and then if you figured out how to harvest that data, you can actually build lines that improve themselves and then you eliminate the waste. So this new different way of doing things, like can you just tell me sort of specifically, like a few details, what would it be like, what would it look like, what would be happening in this world? Everything gets built cheaper, everything gets built with less waste. And when that happens, we have maybe we have cheaper products, we have maybe more profitable companies, we have less waste going into the world, less physical waste, chemicals and rivers, less energy use, less human lifetimes wasted, And it's just thinking completely differently about like that manufacturing actually is a machine itself, and that we need to optimize that whole process as a machine itself versus a means to justify the ends of like we need we need to get products out the other end of the line. In a minute, the Lightning Round, Anna tells us what mechanical engineers know about the world and the inefficiencies out in the wild that grind her tears. Now, let's get back to what's your problem? Great, let's do the Lightning Round. Are you ready? What's one piece of advice you'd give to someone trying to solve a hard problem. I love that question. I love it. I had a science teacher in high school taught me how to solve hard problems because I was working on science research. I'm a science fair kid, and he taught me to break those problems down into manageable pieces. So like, if you can take smaller steps and understanding a problem and the small steps you can do, then it becomes easier to solve. So as somebody whose job is sort of to find errors and inefficiencies in the world. Is there some some domain something you encounter in your daily life that you just you really want to optimize inefficiencies in things that cause lines really irk me, like at the airport or in like a restaurant, like anything that feels like a non optimal like kind of scheduling and line that causes a line of humans. That is something that I that I noticed, and I just like really grinds my gears. Of all your patents, which one's your favorite, I'm gonna go with my first patent more because of what it means. So I share my first patent with actually who is now my co founder, Sam Weiss, and we met in two thousand and nine at Apple, and that's actually the summer that we invented our mutual. It was our mutual first pattern, and we had to build essentially a switch very similar to you know on the side of an iPhone there's like a little ringer switch that you can you can kind of down. So we had to build a switch like that that had a very small ford back there, like it needed to fit in a weird shape. And so that was our first pattern. That pattern will never see the light of day. There will never be a switch that has made in that design, but it was cool. That's awesome. What do mechanical engineers know about the world that nobody else really gets? That everything is imperfect and different from every other thing. Is there a particular piece in say my iPhone that is the one that is like has caused the most manufacturing problems, the most problems on the assembly line? Is there some piece that's like, oh, that piece is the one. I never worked on phones, so I don't know specifically, but in general in products, the antenna, anna, anything with glue, and anything having to do with water, So those are the three things. It's like. It's like the antenna's heart displays are hard too, but like, so basically everything. Is there some trick that mechanical engineers use when something isn't working, like a remote control or whatever. I always think about it, is it a hardware problem or a software problem. If it's a software problem, reboot it. If it's a hardware problem, I don't know, check the batteries. I make sure it's plugged in. Anna Katrina shad Letsky is the founder and CEO of Instrumental. I have a request for you this week, and it is this. Please let me know who you want to hear on this show. You can email me at problem at Pushkin dot fm. That's problem at Pushkin dot Fm. Or you can tweet at me at Jacob Goldstein. Today's show was edited by Robert Smith, produced by Edith Russolo, and engineered by Amanda k Wong. I'm Jacob Goldstein, and I'll be back next week with another episode of What's Your Problem