Podcast #74: Toward Intelligent Manufacturing's Tipping Point

Good day everyone, and welcome to another episode of the Momenta Edge Podcast, this is Ed Maguire, insights partner here, and with us today we have Jon Sobel who is CEO and Co-founder of Sight Machine. Sight Machine is a really interesting company, we’ll be getting into that in a little bit, but they are really at the vanguard of applying analytics to help manufacturing clients in particular optimize their operations and output, and reduce risks, through the application of their technologies. Jon had been a participant in one of our prior webinars, but now we get a chance to really dive into things and explore a number of topics. So, Jon it’s great to have you on the podcast.

Thank you, Ed, it’s a pleasure to join you.

I always like to start things off by getting a bit of a sense of your contacts, and what has brought you to where you are today, particularly in terms of your relationship with technology; I know you’ve got an interesting background having worked for some very prominent companies. I would love to hear about what’s brought you to where you are today.

You bet, and there are a lot of lefts and rights of how I gone toward what was interesting at the time; my technology career started in the mid-nineties, I started working with a semi-conductor company that was doing graphics chips in the PC era. That very quickly led to a stint at Yahoo! where I didn’t look up for about six years, I joined in the late nineties and when I left, I was on the management team there. It was a really interesting time because the first chapter of the internet was coming to a close, and you could tell where Google was headed and what was about to happen next. That whole era was a lot of times spent thinking about the application of internet to a whole range of industries, but many of them at the time were virtual world industries, so a lot of information, a lot of work around the adoption of internet technology.

I then went to work on the other side of the street, I went to a large media company, I went to CBS when online video and mobile was happening, I wanted to see what it was like to be on the incumbent side of the technology dynamic. I was there for a couple of years, I missed being on the disruptor side, and that’s where I met Nathan Oostendorp, the CTO of our company and the founder, the guy who started it all off. He and I got to know each other at a pre-cursor to GitHub, an open source distribution platform that housed a storage website called Slashdot, and a large open source distribution platform called SourceForge. Nate and I worked together for a couple of years there, the whole focus was on modernizing a 10-year old platform, and Nate built this very impressive big data backend, just as big data was becoming big data. We were generating 30 terabytes of data a day at that time on the site and pushing around about 2 million software files a day all over the world. So, that was a really good foundation in where technology was headed 10-years ago.

I then got very interested in energy, and I spent some time in the energy field. I was briefly on the management team of Tesla, I also worked for a really interesting carbon capture company, and the takeaway from those experiences was, the opportunity for the intersection of high technology in traditional industry. It was very evident even then that Silicon Valley was very focused on stuff like social media, and still virtual world things, but that there was going to be a lot of opportunity for companies that could talk to both sides of the street. At that point around 2011 Nate took the work we had done around data, processing huge amounts of it, and started to think about where to throw that. He contacted me when he had one other partner, one of our founders, Anthony Oliver, and some help from another one of our founders, Kurt DeMaagd, we all started thinking about manufacturing in 2011. So, that’s how we got to here!

That’s a pretty diverse background. I’d love to get your thoughts on some of the experiences that shape your views working at SourceForge, because some of the interesting discussion threads that we’ve had in the podcast have focused a lot on the role of open source, and the open source development model, but also maybe some of the challenges involved in building businesses based on open source technologies. I would love to get some perspective on what you saw there before GitHub had really emerged.

 There were a couple of things Ed, one was there was definitely a moment in time when open source was really almost a belief system. When I joined, I was a businessperson who was asked to manage a bunch of open source engineers, who were very mission-driven by the ethos of open source, as people who have studied that know there’s definitely a value system underneath it all. But what was happening, and GitHub really epitomized this, is it was becoming a very practical way of developing. Today people use open source all the time, and it’s no longer so much mission driven as just a really good way to develop.

There were a couple of things at the time that seemed self-evident, I’ve since realized that they weren’t self-evident to everybody, but one was there’s a difference between building a business model around, ‘Being an open source company’, versus just using open source in your stack. There are a lot of great companies and most software companies today use a lot of open source in their stack, but they don’t self-identify as an open source company. And we’re that way, we have over 100 libraries in our stack, and if you have good hygiene and you’re just careful about making sure you use software with the right license structures, it’s an incredibly scalable, cost-effective way to build a complicated piece of technology. So, that was just starting then.

Another thing that I took away of course was part of the ethos of the open source community, and this has gotten very mixed into how technology has developed all around the world now, open source traditionally was an extremely non-hierarchical meritocratic environment, people worked on distributed teams, it didn’t really matter what your job title was or who you were, your contributions to the project, or, your ability to help foster collaboration were really the currencies in that community. I had the good fortune to work with a lot of really smart engineers who thought that way, and I believe those kinds of approaches to software development are how you build a really great piece of technology. So, the experience of being around a bunch of really good engineers who have been steeped in this way of working, definitely influenced what I brought to the party when I got together with Nate, to start working on Sight Machine.

It’s interesting because as I think we’ve mentioned before in our offline conversation, I have met Nate and interviewed him for a report that I had done on innovation, at my prior job at CLSA. It’s amazing how small a world is, he was talking about the open source innovation landscape, and it is really fascinating how in 25 years we’d gone from a sort of fringe technology movement, spearheaded by engineers and do-it-yourselfers, to now with open source technology being foundational to the largest corporations in the world, and the biggest internet companies, and the most scalable infrastructure. It really is remarkable what a shift that’s been.

I did want to touch on your experience at Tesla, because the Tesla experience as you alluded has really shaped your views of the application of technology in industry. I would love to get your thoughts and insights on what you saw initially working with Tesla in the early days, but also some of the problems, challenges and opportunities that you saw in the energy industry.

Absolutely, and similar lessons from both experiences. In its earliest days a lot of Tesla’s uniqueness and advantage was around the skillful application of software to classic power management problems. So, in many ways it really had a software component, and of course over time as people talked about the experience with the vehicle, and the way a software is used to enhance the driver’s experience, or to continually update vehicle software, everybody has come to appreciate that one way we can look at cars is a rolling computer. Yeah, there are very significant considerations around it being a car, and cars being something expensive and dangerous that have to be taken into account, and so both at Tesla, and as I had the good fortune to get into some really interesting energy problems.

I came to feel that there are mindsets & world views in traditional industry that need to be taken account of and understood. There’s a whole view of risk that is necessarily different for life safety issues than we typically have Silicon Valley. There is also a speed and a spirit of innovation in the high-tech world that is very different and very useful, that can be applied to traditional industry. I just remember thinking, at both companies I was doing transactional work, regulatory things, communicating on behalf of those companies to a whole range of people who are not from the world here, and I got very intrigued by opportunities where you can bring both worlds together. I don’t remember exactly when it was, I think it was 2012-2013 Marc Andreessen wrote about ‘software eating the world’, it’s true that software eats the world, but you’ve got to understand the world, you can’t just do it from a garage especially, and  as the stakes and the span of it get bigger we’re actually bringing this all together. This is the future of technology is doing real stuff of more and more physical world consequence.

I got very intrigued by, how do people from both worlds talk to each other, and build something together? That’s what’s been so gratifying about this is, you really can’t think of a more traditional industry than manufacturing, but like every other industry on the planet there’s a ton of data, there’s a lot to do, and if you can figure out how to skillfully apply what we do here in Silicon Valley to that world, there’s a lot of opportunity for gain. And as I think we will probably get into, the cultural and human being part of this are far more challenging than the tech! So, it’s an endlessly interesting journey.

It is. I’d love to hear a bit about the backdrop and the origin story behind the genesis of Site Machine; what was it, whether it be manufacturing or the pain points that you identified, where you saw that there was a need which hadn’t been addressed yet?

Here’s the hacker mindset, we were talking about opensource a moment ago, and there’s a bunch of things that people who work with that community will immediately recognize; hackers really don’t like to do unnecessary work, there’s a lot of motivation in software development around avoiding waste and wasted time and coming out with elegant solutions. One of the things I so admire about Nate, Anthony, and Kurt is, in contrast to many technologists who fall in love with what they’ve built, and then go look for what to do with it; we were always motivated by, ‘What’s the problem?’ and then, ‘Let’s build what we need to build to solve the problem.

So, the origin of the company is, we decided that manufacturing was a really interesting domain because on first principles its appealing in a number of ways that people now appreciate but weren’t widely understood then. If we think of data as fuel for insight, how much data is there? Manufacturing has more data than any other category by a fact of two. So, it’s a domain with a lot of data, if we think about the economic value of the data, what is the percentage point of improvement in manufacturing work, in any large manufacturer’s worth tens of millions of dollars, maybe more? So, there’s good initial conditions from a technology point of view, a lot of data, there’s very compelling economic impact that’s quantifiable and meaningful. It’s a really hard area, and in 2011 there hadn’t been a lot of progress in 20 or 30 years in using this data.

So, here’s what we knew at the outset, we thought okay this is a cool category, it’s hard but we can make this data useful. Then it was, let’s go figure this out, and here’s where I just feel so appreciative of how our leadership approached this; we spent about two years, most of us were in SE Michigan and there’s a lot of automotive manufacturing in SE Michigan, most of us were there and those of us like me who want to spend a lot of time there anyway, we went to a bunch of factories. We did what you should do when you start a company, we asked a lot of questions, we spent about two years going to a bunch of plants and asking, ‘Where is the pain?’ ‘What is the pain?’ And what we quickly realized was, plants were awash in data, but they couldn’t use it.

Like many start-ups we began with a very specific focus on a certain type of data. We started working with image data from machine vision systems, that’s not a widely known area, but what happens is there’s a lot of cameras in factories that take pictures apart that are being produced, the cameras generate a bunch of images, and we started to apply very sophisticated AI techniques to understand the images and identify variation. We got hired by a couple of really big companies, and what immediately happened was, they said, ‘It’s cool that you understand this, but we want you to understand everything at once, we’ve got all different kinds of data, and we’ve got a bunch of point solutions, we want something that understands it all’. So, about two to three years into this, that’s when we really locked into our opportunity. Ed, I can’t tell you, it came from maybe going to 40-50 plants and just talking to people, we heard the same thing again and again.

What are some of the unique challenges about the different types of data that you find in manufacturing plants? I’m assuming you’ve got Plc’s, you’ve mentioned machine imaging, but also there’s a lot of proprietary data sources and protocols etc. I would love to get your sense as you’ve mapped out some of the challenges were, and how you prioritize the work that you did to at least identify where you could add the most value as a young company.

You nailed it Ed. In very simplistic terms, if we think of what we used to call big data as three V’s, volume, velocity, and variety. As of 2011-2012 the big data world had come up with really good solutions for volume and velocity, Hadoop was all the rage and people were very intrigued by the needs and the opportunities around computing huge amounts of data at once, but most of what was being done was for reasonably well-structured consistent datasets. If we think about going into the physical world, and we say this light-heartedly but its really true, manufacturing is the NFL of data variety for all the reasons you just alluded to. You find more different types of data generated by different things with less consistency, in manufacturing than you do anywhere.

This is true inside of any single plant, you and I could both be plant manager for the same company, we could be making the same stuff, and the data environment of Ed’s plant would be completely different than the data environment of Jon’s plant. It’s different in every aspect, its siloed so it’s in different places, its structured differently, it’s on different rhythms, different timeframes, and so the great challenge which is more difficult in manufacturing than almost any other environment, is to make sense of this data holistically, and to put it together in an automated way, not by hand, so that you can get useful information out of it.

What we ended up doing, we were challenged very early-on, there were maybe ten of us, and one client came to us, and it was a really cool moment in the company’s life in retrospect, they said, ‘Can you all continuously stream data out of a bunch of different assets, from a number of different plants, owned by different contract manufacturers? So, every facet of a problem showed up here, ‘Can you stream this data, it’s all different, and can you integrate and blend and make sense of it on an ongoing basis, and if you can do that here’s a sizeable cheque. Tell us how much time you need and how much money you need’. So, we were really lucky to find a company that wanted to roll a dice on solving that problem. We bit it off, this was 2014, and we decided let’s see if we can solve this, if we can’t well it will be over, but if we can we will really have advanced our capability in the field.

We did solve the problem, and what we ended up realizing, and this wasn’t planned, but what we realized was the way to attack that problem is to abstract it at a very deep level, to come up with ways of handling the data that are essentially independent of the machines or processes you are starting with. Abstract the problem in such a way so that you can basically take in data from any source, any process, and make sense of it. That required in manufacturing what companies have found again, and again, and again, is that they just use standard analytical techniques, or get a bunch of data scientists to work on stuff; everybody invariably ends up doing bespoke models for a local problem, nobody gets any scale, so, we were forced by taking a big bet to bite the scale bullet right away. It goes right at the heart of what’s so difficult about this, billions of dollars have been spent in industrial internet trying to make sense of manufacturing, and the reason most people hit the wall is because of the variety problem.

So, that’s how it all got started for us, and of course that solution is not a manpower problem, it’s not a matter of having a hundred engineers, it’s a time and experience problem. Once we got into it and started to refine the models and go to more and more plants and get smarter about data, and tuning the models and so-on, you get a really nice accelerating advantage.

I’d like to zoom in on a point that you highlighted, which was this idea that every plant is unique, and there’s so much I guess heterogeneity among the different types of machines, and different types of data, yet in building that abstraction layer how did you work with your clients to attach context, whether it be semantic context or business context, to the different types of data, so that you could build this framework that you could use to be able to replicate it?

I really appreciate the question, it’s a great question because you’re right. If we think of the flow of data in a left to right way, if in our minds we put a picture of all this crazy heterogeneity on the left, and then we think about pushing that data through some sort of abstraction layer, we have to get to a small consistent set of things on the right, otherwise we’re just taking chaos and turning it into more chaos. Here’s where again the thoughtfulness of our technology leaders were so just spot-on, we didn’t invent any framework, we didn’t ourselves come up with how to organize the insights from this data; that structure, that set of principles about how to organize it happens to come out of manufacturing.

Here’s what we realized after working at five or ten companies, it doesn’t matter what you make, if you’re making cars, drugs, food, or oil, there are the same set of questions and needs that you have as a manufacturer. The discipline is manufacturing itself, and every manufacturer wants visibility in the production, they want to know how much stuff they’re making, they want to know how fast or well they’re making it as compared to an ideal, and they want to know why they’re having problems. The problems generally fall into only a couple of buckets; your machines may be going down in ways that are unplanned, so you want to understand why that is happening, your quality may have issues, and there’s actually a family of matrix in manufacturing that many companies use called OEE, many of your listeners will undoubtedly be familiar with it. But it turns out that there is this megastructure around manufacturing that already exists.

So, what Sight Machine is helping companies do is to take all this heterogenous data and organize it around deep structures, ways of thinking about manufacturing that they already have, and it turns out that if you do that part well, what you referred to a moment ago is abstraction, you can then associate all of this data with these concepts, and then the thing that everybody thinks is hard, the analysis, the math, the techniques that we apply to data, that goes lightning fast because you’ve got the data in the right building blocks. It turns out that no matter what you make the building blocks are basically the same, that’s true for discrete and continuous, it’s true for really whatever product you make.

I’d love to understand a bit more about how Sight Machine has expressed this in technology. Could you talk a bit about your technology, and then how that gets applied in the context of your customers manufacturing environments.

So, we’re a software company, the first thing I should say at the outset is we’re completely agnostic as to company’s edge situation of cloud strategy, we work on whatever edge networks they have and however they’re moving data. If they are moving data or however, we might extract it and aggregate it, all that is quite flexible. We’re a subscription software product, and what our product does, and this is a really important thing for me to emphasize, it’s a product, it’s not handcrafted, it’s not bespoke, it’s not a consultant-heavy thing, it’s a product that’s streams data as its being generated, through a transformation layer, and then takes what you can think of as a manufacturing data warehouse, and either pushes that date in the client systems that the client has, a factory information system or wants to use Tableau, or SaaS, or whatever, great tools are out there for BI visualization, power BI like you can push it into all that stuff.

We also have a browser-based visualization and analysis layer, and all this is very open. So, we’ve got APIs on this, we’ve got an FDK, we’ve got data discovery tools, and either we or our clients, or advisers of theirs can start doing very sophisticated things with the data. We have a number out of the box analytics that come with our product, so we can show clients on a continuous basis; here’s this chronology production, here’s this closing down time, here us causing quality issues, we can offer predicable analytics on top of it. Given everything that’s happened in the past, what’s likely to happen in the future, and obviously the more data we can put into this the better, because you get more and more robust insights. What we’re finding is, a number of companies go through a very similar progression, believe it or not reliable visibility in the production is a big deal, it’s amazing how well companies do with Excel and daily roll-ups and reporting, but having real visibility is helpful.

So, a lot of times companies will start with visibility, they’ll be somewhat surprised by what surfaces as problems, they’ll dig into it, they’ll figure out what’s causing them, they’ll fix those problems and move onto harder ones. So, what we really end up doing with our client is, we offer this product, it brings a lot of new insight, and it ends up helping companies develop a workflow around data. So, we will work with them on developing that workflow, and really bringing this into production.

What are some of the organizational dynamics involved with a successful implementation? You’ve had a chance to develop solutions from the bottom-up as it were, but how do you best help your clients identify their pain points, and are there best practices or challenges that need to be overcome when you’re combining your information technology-based solution, with traditional industrial or manufacturing machines and technologies?

That’s a great question, and as you might expect, the organizational challenge ends up being the most complex and important. We’re learning all the time about this, and here’s what we understand so far, it’s absolutely essential at the beginning of engagement, so McKenzie said something like 98 percent of companies are digitally transforming, if you ask most companies, ‘Why are you doing this?’ a lot of times you don’t get the same answer from everybody, it’s not clear. So, this involves some work in some pain, and the first comment is, ‘Somebody’s got to know why it’s doing it and what it’s for, or it’s going to flounder’. It doesn’t have everything figured out, but it’s got to have very a very strong committed leadership that has some sort of North Star in mind. And you know, being a better manufacturer is a great place to start!

And that leads to the second thing which has been really interesting to work through with companies. Transparency is something that every company says it wants, but the incentive structures and the way people are measured is often at war with the idea of transparency. It’s very hard in some cultures to suddenly reveal all the problems in a plant and have everybody be okay with it. Now, ideally that’s exactly what would happen, organizations would embrace a plant manager for saying, ‘Hey, I thought we were here, but really we’re a little bit behind and here’s our challenges’. So, one of the things we worked really hard to do is make sure companies are ready for this, I can’t tell you how many times the data has come out, and the first moment of truth when it’s all been put together and people have said, ‘That can’t be right, those numbers are wrong’. Obviously, you adjust, refine, and validate but almost always there are big surprises at the beginning because companies have never really been able to shine a light on what’s going on.

We find that just about every company that says they want transparency, not all of them do, and it’s really hard to coach. Think about it, if you’re a plant manager and you’re going to get punished for revealing previously unknown problems, that’s not easy to do. So, we work really hard to guide the management teams to think something like, ‘This is actually good, we now know stuff that we didn’t know before, and now we can get better’, and of course, there’s a great tradition of continuous improvement in the manufacturing world, so we find that almost always the manufacturing types are pretty cool with this. The business types may/may not be on the same page! This is where organizational politics, and, ‘Are we as good as we thought we were?’ and all that stuff comes in.

The other thing is of course, this involves a lot of change for IT. IT is becoming a much more strategic function in companies, and it’s got so much opportunity in front of it, but there’s vulnerability on all sides here. IT leaders have to learn a lot of stuff about manufacturing they might not have known, and it’s not easy for an IT leader to go, ‘I have no idea of what good production really looks like. Is this good or bad, how can I help?’ So, we see a whole range of behaviors and dynamics between IT and OT, that’s where this is won or lost, and ultimately, ultimately, that’s about company leadership; if you’ve got a good leadership team that’s telling everybody we’re going to win, we’re going to get good at this, we’re going to find out some bad stuff we don’t like, but we’re going to work through it’, then you’re going to be fine. If you’ve got the usual politics and people covering their own turf and all that, then no matter how good the tech is it’s not going to work.

Yeah, that’s a great point that you make, that transparency is not necessarily… people get what they didn’t expect a lot of times, and that can certainly create some issues.

What about some of the common benefits that you’ve seen? I think what’s so interesting that you’ve expressed here in your experience is, manufacturing analytics start out from this really disparate heterogenous many sources of data, but then when you’re able to roll this into a consistent framework and apply some of the experience into more of a packaged approach, or I’d say more of a standardized approach; what are some of the benefits that you see, the real advantages of successful implementation of your types of solutions?

You see very concrete gains, sometimes shockingly fast. There was one company that we started working with about a year and a half ago, and as soon as we went live with them in their first plant, the data suggested that their cycle time was significantly longer than they thought for the automated process. So, anyone who knows manufacturing, well that’s totally bad, you think it takes three hours to make something, it takes four and a half, it takes four and a half hours, that’s terrible. What was fascinating about this company is they held a hackathon, they literally got a bunch of operators and said, ‘Let’s ask everybody to look at the data. Let’s try to find out what’s making the cycle time longer than we think’, and in a matter of weeks they improved their cycle time by seven percent. That’s an enormous leap for a highly automated company, and it highlights another dynamic that’s very challenging. This has taken me some time to appreciate.

You will go into plants where people have been working literally for decades to improve their process, they’d got it up to a certain level and they just keep getting it a little bit better, a little bit better, and it feels and seems like any sort of big gain would be impossible, but they’re doing this without the benefit of what their data can really tell them. When you open it up this way you can support big step function leaps in improvement that nobody expects. Once that happens of course people get really excited. So, we’ve seen that type of thing, we’ve seen very complex judgements that are made by people get reduced to a recipe. So, in a lot of the process industries, for example if you’re making paper, or believe it or not, milk, cheese, there’s all kinds of processes out there where there may literally be a thousand parameters that are influencing the decision you have to make about whether to turn something on, or off, of adjust it. Literally every time you run the process, you’re generating data, and if you think about the golden run, the perfect run for that stew under specified conditions, we know from years and years of running the plant when it worked and when it didn’t, and why, and if we can crunch all that data you can now take human intuition, which is very good, and you can quantify it and put it down into a recipe.

So, things like that come up again and again. I read a great study the other day that talks about people’s ability to process information, it was a study about people betting on horses and horse racing, up to a certain number of parameters, something like 15, we get better and better if we have more information. But the human mind can’t process more than 15 things at once, and if you think about a typical manufacturing process that has 40, 50, 500 parameters, there’s no way we can make sense of that. But if we can let the data and the math tell us what the right recipe is, or what the right things to attack are, you can drive very significant gains.

The last thing I’ll say about this Ed is, I’m giving you production percentages, what’s really cool about manufacturing is, it’s such a fixed-cost industry that every incremental unit that is produced is highly profitable, because you’ve already paid for all your machines and people to do stuff. So, you can drive a five or ten percent improvement in productivity, and these are the numbers that everybody’s reporting. McKenzie has got some great work on this, it’s what we’re seeing too, that’s significantly more impact on profitability than five or ten percent, sometimes its three or four times that impact. So, this is no joke if you can really get it down in.

You mentioned earlier that you use machine learning and some other techniques as well to accelerate the identification of value creation, but I’d love to get your perspective on the role that AI and machine learning can play, and whether there are some misperceptions in the market in terms of the potential outcomes or power, that we’ll just call it AI for now, can really bring to the table, and whether people need to be worried about losing their jobs because of this technology, particularly in manufacturing, right?

Great question. I was thinking when you asked me about my background how many technologies cycles I’ve personally been through, it’s more than three or four now, and every time something like this happens everybody seizes on it, so this is the thing, so it’s partly true and it’s partly not. AI and machine learning are great techniques, I believe I saw a presentation from him, he’s a great evangelist and communicator about AI, and he said, ‘AI is kind of good today for things that a human being can do in a second or less, that’s about it, that’s about where it’s at’. There’s a huge amount of misperception that my friends and colleagues in Silicon Valley have fomented about just what AI can do relative to what it actually can do. What we cannot do today and what no-one should expect is, just take a whole big blob of factory data and put AI on it, it doesn’t work there’s no underlying structure that AI can work on, it can’t make sense of a bunch of out-of-order sensor data. But what you can do is, once the data has been correctly structured, you can use a lot of these techniques, and machine learning is a really good one, to start eliciting meaningful signal from noise.

Your question about job displacement I think is really important, so much of what’s been driving manufacturing for the last 10 or 20 years is a combination of automation and offshoring. We see a lot of plants that are short of people, they don’t have enough, there’s something like a million to two million unfilled stem jobs in the United States, and there’s no shortage of need for people who can work with and understand information. The jobs of the future increasingly are going to involve judgement, because I tell you, even though we give a lot of really good insight to our clients, we’re not running the machines, were a long-long way from that! Flying cars is maybe somewhere in the future, but not anytime soon, and that’s where lights-out factories are too, there’s still the need for people to look at the data, interpret it, and figure out what to do.

So, there’s no question that a lot of work has been automated and routinised, but there’s a huge need for people who can work with processes and information, and companies that figure that out are going to have the edge.

How would you compare the state of adoption of either advanced technology such as yours, across manufacturing? I realize it’s been some time since you worked in media and at Slash.dot, but I’d love to get your perspective of where we are in the adoption cycle across manufacturing, whether there are any differences that you see in terms of who are early adopters. Ultimately how you start to see the next generation of manufacturing playing out?

Absolutely, and we think about that a lot. My personal view is that it’s still really early, there’s been a palpable shift in the market in the last two years, most companies are spending money and trying stuff. If I hazard a guess and put a rough cut on it, maybe 1 in 10, or 1 in 7 or 8, are really serious, experienced, and thinking about scale. Scale is the big divide, just about every manufacturer is doing a bunch of Plc, but the problem with Plc’s is, you can test the technology, you can test it on the bench, but that’s not the same as scaling the technology. We don’t get any of the operational issues you mentioned a few minutes ago, we don’t get the really stress testing whether technology can work together, we don’t get any of the real benefit until we start scaling.

Now, that number seems to be going up fast, and I read a fascinating report last week from Morgan Stanley, and they said in their experience when 20 percent of an industry gets serious, that’s a tipping point. I don’t think we’re 20 percent, but we will be in a year or two, and there’s now fear in the market, it’s really interesting! When we started out it was all about hope, there were a couple of companies that thought, ‘This is really cool, I want to be a good visionary, an early adopter, I want to learn about this’. There is now palpable fear from companies that they’re going to be left behind their competitors, and that’s driving a lot of spending and engagement.

So, if we compare it to the internet or open source, where are we? First innings, maybe second innings, I think we’re just getting started. There are so many companies that are dabbling in this, and a meaningful but small percentage of them are getting serious, but that number is starting to grow fast.

What do you see as key risks going forward, in realizing successful adoption of the advanced analytic systems and manufacturing, and I guess put another way, what keeps you and what keeps your clients up at night?

From the client point of view, probably the biggest risks are organizational more than technical. If we go down the list of standard things that people talk about that are really important; security is very important, but the data is generally a way from control. So, if you want to analyze data, that’s very different than getting into control systems and monkeying around with how you actually run the plant. So, from a purely analytics point of view, security is very important but it’s not determinative of whether or not you do this, or how well you do it. The organizational issues I think are keeping up a lot of our clients at night, they know they want to do this, they know it’s important, they’re trying to figure out how.

Thinking of ourselves, we’re seeing a lot of growth, things are starting to happen really fast for us; this may not be what you were expecting. As a company our biggest challenge now is just ourselves, we’re in execution mode and it took us five or six years to build our technology, the stuff was no joke, it took a long time and it was hard. We’re now entering the realm of being a growing business and company, these problems are tried and true and every company has to deal with it, that’s where a lot of great companies hit the wall. So, we’re spending a lot of time making sure that we’re built out to scale, and that we’ve got good leadership, and above all that we give our client results, that’s the acid test. Manufacturers are skeptical smart quantitative people, and if you show them the numbers, they’re incredibly loyal. So, it’s a really interesting moment because for so long Ed this was evangelism, and now companies have proof, we have proof as a technology partner with them, and now it’s all about partnering well, and executing well.

That’s great, and it sounds like you guys have made a real significant amount of progress in the past few years, to really realize that vision. As you look forward, are there some technologies or approaches that you’re particularly excited about? Or, new and untapped opportunities that you guys are focused on?

I don’t know that any of this is particularly new, it’s seeping into the world, but I’ll tell you some of the things that my colleagues identified a year or two ago that we’re really excited about. Container technology for cloud applications, being able to work on a variety of clouds and by extension on the edge is a really powerful theme in the market right now, because so many great analytics companies grow up in one cloud. Then of course clients say, ‘Well I’m on another cloud’, or, ‘I want to be multi-cloud’, or, ‘I want to be private cloud’, hybrid… or whatever, you’ve got to be able to apply this across a bunch of different storing computer environments. The boundaries between edge and cloud and all that stuff if blurring, there are these great technologies out there now that puts huge amounts of computing power at the edge. So, that whole theme, and there are a lot of really cool announcements this past summer around that from the cloud companies, that whole theme is really pervasive and important.

The other thing we’ve seen which we’re really excited about is, the development of strong horizontal layers in stream processing technology. Forgive me for geeking out here, if you really want to understand what’s happening in a plant, it’s one thing to go back and play with yesterday’s data, which is what most solutions in the past have enabled us to do, but if you want to see what’s happening right now and which of the five or six things that you think might be the problem, are actually a problem, you need to be stream-processing that date, you need to have that data flowing through a pipeline, kind of subject to continuous revision and analysis. There are now some really cool technologies around stream processing, that are making that whole platform level of doing that more robust and scalable. We built a bunch of that stuff ourselves four or five years ago, and we’re now reliant on some great open source technologies to do that.

And I guess the other thing I would say, which is not per se a technology answer, but is really important, is we’re seeing great companies like Microsoft lean into this. There are a couple of leaders out there in the market that are really getting behind manufacturing as a vertical, and I’ll just be really candid; Sight Machine could have the greatest technology in the world, if it goes to X, Y, Z large enterprise manufacturing and says, ‘Hey, let’s do analytics across all your plants’, the god’s honest truth is, it’s very hard for a large enterprise manufacturer to rely on a young company, not matter how cool the technology is. But if Microsoft, Google, or Amazon goes to them and says, ‘This is on, it’s happening, we’re supporting it’, then it’s a different ball game. We’ve been really impressed with some of the leadership we’re seeing out there, from the incumbent technology companies, they’re taking manufacturing very seriously, Microsoft in particular is a leader in the category, it’s making a huge difference in the marketplace.

That’s great insight because companies that are focused on ensuring that they mitigate risk for their critical systems, I think they do look to the big established companies to validate new ideas, and as you mentioned, it really becomes very helpful for the industry overall to move forward and adopt new ideas.

Just to put a fine point on this, this is something I’m increasingly aware of, if you’re a CIO at a large company, or a couple of levels down from the CIO, you’re going to put your name behind the initiative and recommend that your company try something like this; many of these companies are not the most forgiving environment, and you feel a lot better making that recommendation when somebody like Microsoft is standing next to you, and let you know they’re going to make it work no matter what. So, there’s an analytical set of risks, and then there’s the human being risk of, ‘Am I going to put my career on the line, put my name behind this?’ All of that is starting to happen now. So, that’s what’s got to happen for an industry to move forward, and I think that’s why that 20 percent tipping point is so important, because when you see enough people around you doing it, the cost of sitting out outweighs the cost of taking a risk.

Right, you get the FOMO, the Fear of Missing Out kicks in as a driver of urgency, which benefits everybody who’s in the industry. Yeah, that’s good to hear that we’re headed that way.

This has been a terrific conversation, and I think we’ve really explored a lot of great insights here. One question I always like to ask is for a book recommendation that you could share with our listeners, or other resource for anybody that’s interested in something that you might be able to share.

There are two books making the rounds in our company right now, one’s called Creative Construction by a Harvard Business School professor, Gary Pisano. It’s a cool book because it takes conventional wisdom, and it turns it around a little bit. The premise of the book is, we all think large companies can’t innovate, but maybe they can, and, how can they?

Another interesting book that we’ve been reading and talking about on our leadership team is called B4B, it was written 2013 I think, some time ago, and it’s a really interesting book because it talks about the evolution of the software industry business model, from pure product model to product with a lot of sources. But the next generation of company, that book argues will be an outcome-oriented company, a company that uses a scalable product and sources to deliver a specific outcome for a customer. We find that very compelling because in manufacturing the ROI is so concrete and so important, and if we set up the discussion from the beginning of, ‘Hey, let’s lift your production two percentage points, use our product, let’s come up with a programme together’, both companies now are committed to the outcome. So, a very different mindset than, ‘Hey, I’m going to ship you a piece of software off my dock, and good luck. And if you need any help, call my services arm’, it’s subtle but it’s important, so we like that book a lot.

I’ll recommend one more, you invited me to do one and I’m going to do three, because I’ve just finished a book that isn’t really about work but …

Oh, no that’s great, we love it.

Something I’ve been thinking about lately, it’s a biography of a guy named Jon Boyd who was a really interesting theorist of doubt, military strategy but ultimately changing innovation. In the startup community people have been talking for years about what’s called OODA loops, and even trying to explain it dumbs it down too much and robs the concept of what it’s really all about. The story of the man and the idea that he came up with are inextricably linked, and one of our leaders, our engineer leader Jon Merrill and I were talking about OODA loops, and I thought I’m going to go figure out really how he came up with this. It’s just an intriguing book, it’s a book about a very brave soul trying to bring change to the world, and what that’s like.

So, for any of your listeners who are trying to do that in their companies, and just thinking about moving something forward in the world, I was very inspired by this guy’s story.  It’s a great book because he’s not a perfect human being, and it tells what he went through and how he figured out a lot of cool stuff, and there is a lot of applicability from that to everything that we all do every day. It’s called Jon Boyd, I believe it’s by Robert Coram, ‘The Fighter Pilot Who Changed the Art of War’. I had a long plane-ride last week and I thought, ‘I’ve got to turn on the brain candy here and just get away from the Excel, and it was pretty cool!

That’s great, that was my understanding of it, that it had come from fighter pilots who need to make these decisions, these life or death decisions with very limited information, and he had to navigate in quite uncertain environments.

Also this B4B, you kind of hit on I think a really important transformation, that I haven’t heard a lot of other people talking about, but it struck me that when software moved from on premise to as a service, not only does it change the nature of the delivery of the technology and the architecture, but it also changes the relationship between the client and the vendor.

I think you’ve hit on the point there, which is that first of all if you’re subscription-based, you have to ensure that your clients are happy enough so that they’ll keep coming back and renewing. But secondly, I think this goes back to a point that you made at the top of our conversation which is, it isn’t as much about the technology anymore, that’s really just a means to an end, but it really is more about the outcomes and the value that you bring to your clients. Those are wonderful recommendations.

Thank you, I appreciate the kind words. Our clients really have impressed on us that just as you said, technology is a means to an end. Let’s really focus on why we’re doing this, and what’s the benefit, let’s really focus on value. If you line up a new customer that way and you have something cool, you’ll never go wrong.

Better words have never been said about business! I totally agree. Well, this has been a great conversation, we’ve been speaking with Jon Sobel, CEO and Co-Founder of Sight Machine. Again, this is Ed Maguire with another episode of our Momenta Podcast. And Jon, thank you so much for the time.

Thank you, Ed, it’s a pleasure.

[End]