Open any artificial intelligence chatbot, such as ChatGPT from openai, or microsoft’s AI chatbot within Bing, and ask how AI technology can help your business. My guess is that you’ll get a lot of ideas about using AI to write marketing copy, or maybe even do research on your competitive position, but you won’t find a lot of advice on how to use artificial general intelligence to help with the core of your business — your manufacturing processes, and how to make those more intelligent for speed and for profit.
The reason is that the latest AI tools can’t help you here, is that AI systems can’t improve your business processes until those processes themselves have been digitized. And process digitization — especially in the manufacturing industry — is still in its early stages. What is process digitization? And how can process digitization help manufacturers use AI models for profit —to lower costs, increase profitability, and mitigate risk? These are the questions we’re asking and answering on the podcast today.
Today we’re talking with AI systems expert Michael Lynch about deep learning for the manufacturing industry. Michael Lynch is the former head of IoT at SAP software, where he developed powerful algorithms to mine huge data sets for general intelligence and for profit. Michale Lynch is now the founder and CEO of Praxie, a startup that helps manufacturers take advantage of the latest AI tools by digitizing their processes. He’s here today to help us sort through the buzzwords in artificial intelligence so you can get the most out of this new technology. Michael Lynch, welcome to the podcast.
What is Process Digitization?
Leah Archibald: I think it would be helpful going into this conversation if we could just start by defining some terms. Let’s start with process digitization. What is process digitization when it comes to manufacturing?
Michael Lynch: A non-digitized process might be one where you’re tracking your KPIs on a whiteboard or on paper. If you go into any manufacturing plant in the world, there are a lot of non-digitized processes. There are still a lot of whiteboards and paper, where the data sets are not accessible to machine learning.
Then people gather up the paper, and what generally happens is they’ll type it into Excel. Now it’s gone digital, but it’s still a flat data set. It’s hard to use that to generate intelligence. We actually call Excel digital paper. It’s better than nothing, but it’s still a document sitting in a particular space.
Process digitization means the entire loop of the whole process is managed digitally. Only then can you take advantage of AI tools to make the process smarter.
Examples of Process Digitization
Let’s take an example of process digitization. In manufacturing, it might be something simple like a Gemba walk.
A Gemba walk is a process where you walk around and the managers see where the manufacturing work is actually happening. It’s often a non-digital process. The managers might take a notepad. If they’re native to computer systems, they might enter some data into Excel when they’re finished. That’s not digitization.
Digitization of the process in this Gemba walk example means that there is a data-collection process in place to capture the key manufacturing issues in a digital way, and to roll up those key issues intelligently to take action on them. Picture the entire Gemba walk process as a full cycle, where a data set is created and actions are taken based on that data, either by AI systems or by human managers.
Digitization of the process basically means turning the Gemba walk in this example into a computerized process that can be automated, reviewed, and mined for intelligence. It gives you full transparency. Before you even get to deep learning using AI tools, you reach a stage in process digitization where you can speed up your key processes, get through your stage gates faster, and automate the steps involved that had previously slowed down your delivery.
Leah Archibald: So just storing information digitally doesn’t mean you’ve digitized any aspect of your process. It’s just means you have a digital record of one particular moment in time.
Michael Lynch: Correct.
Leah Archibald: Digitizing the process means that there’s a flow of data going through one step to the other. Is that correct?
Michael Lynch: Correct. Let’s take another example — a manufacturing work order that gets brought over by paper. A person looks at the order. Then they then build the order. Then they send it on. Then somebody else grabs that and puts it into an Excel sheet that the order was created. That whole process — how long it takes, how many were created, what the issues are, what the scrap is, what quality issues are — none of that’s tracked.
If you put in an MES system, a manufacturing execution system, now you can start tracking time and all those kinds of things. You start to have data sets. Then you put in quality control systems that you can start to use to track your quality outputs — how many bugs, what your scrap is, etc. All of that then rolls up into the other metrics that you’re tracking. It doesn’t have to be all automated data capture. The data capture could be manual. But it’s part of a process flow that’s capturing the entire process
I would say that about half of manufacturers haven’t even digitized their basic processes in those ways. They’re still capturing things in spreadsheets.
Leah Archibald: I’m sad to hear that. Because I was having similar conversations a year ago today with folks working in the digital transformation of manufacturing, and they were saying a year ago that only about 50% of manufacturers are getting on board with digital transformation. Now you’re telling me that in the last year it really hasn’t gotten very much better.
Michael Lynch: I was actually speaking with somebody today about the digital transformation of manufacturing. The biggest problem in digital transformation actually isn’t the digital technology. It’s the transformation. It’s the human part.
Machine Learning versus Human Learning
If you think about any transformation, human learning is much more difficult than machine learning. Let’s take an example of an everyday transformation. Let’s say you want to get in shape
Leah Archibald: Yes, please.
Michael Lynch: What do you have to do? You have to change things. Change is hard. You have to decide what your diet’s going to be. You have to work out every day. And then you have to track yourself. Am I stronger? Am I able to lift more weight? Am I able to walk longer without getting out of breath?
The point is that transformation is a cultural change. It’s a change in the norms and habits of you as an individual. And if you think of digital transformation in a company, it’s a cultural change too. Process improvement doesn’t start with machine learning, it starts with human learning. Everyone in the company is involved in digital transformation. From the executive level, they have to buy into the change. They have to set up metrics that are measured and improved. They have to set up new processes so that process improvement can happen.
So the human aspects of the digital transformation take as much energy as the technology.
More Examples of Process Digitization
Leah Archibald: Let’s take another example of process digitization to talk about how digital transformation works. Let’s say you have one process digitized — the process of checking safety in your manufacturing process.
Michael Lynch: Safety audits, Yes.
Leah Archibald: Safety audits are a good example of a process that’s rife for digitization. You digitize the process, you speed it up, and maybe you get some data sets that you can analyze for safety data. But if that’s the only process that you’ve digitized, you really haven’t done digital transformation in the enterprise. There’s no possibility for deep learning about anything other than safety. There aren’t a lot of next steps you can take to leverage the benefits of that one process.
Michael Lynch: Well, yes. If you think about what digital transformation is supposed to do, it’s supposed to deliver to your customer more effectively, and deliver internally more effectively. The whole purpose of digital transformation is to make the business better. That should be the starting point.
When we work with companies to digitally transform, we often talk about where are the core pain points? I was with a customer recently where they couldn’t ship product because they had no prioritization in their engineering test management. So a really high importance thing for them to do was to digitize an engineering work order system for their test management. If they can digitize that very quickly and get their products out on time, it has a massive impact to their customer perception, to their own sales, and to their bottom line. The place to start is often with the critical metrics that we need to know about as a business. Another manufacturing example is I’ll go into businesses where they have high scrap and they all know it. They know this is a big problem in their productivity. What I like to do is start with a dashboard to look at that key metric. The reality is that the dashboard is only the outcome of digital transformation. It’s not the end point.
Leah Archibald: If the process isn’t digitized then it’s garbage in, garbage out, as they say.
Michael Lynch: A lot of dashboards may be zero process, meaning I can take from an Excel sheet, but there’s no process evaluation. It’s just a metric. And if the metric is low, there’s no management process to address it, except the manager writes down a note or sends an email. So the dashboard of the core issues should evolve out of digitizing the processes and optimizing those processes. And the visibility comes from that, not from just putting a metric on the dashboard.
Leah Archibald: I think something that you’re getting at here is digitizing a process isn’t necessarily smart. It isn’t automatic deep learning. But bringing the data set to the level of decision making — that’s where you can bring intelligence into the process.
Michael Lynch: Yes. Really what digital transformation is about is having a dynamic flow to constantly process and optimize. You just iterate: Okay, we’ve got our time down to this, now we need to lower our metric, and what are we going to do next about it? It’s a constant cycle.
Leah Archibald: This sounds difficult enough for a manufacturing enterprise, half of which are not on board yet with digital transformation. Now I want to layer in this other buzzword we’re hearing a lot about in manufacturing: artificial intelligence. Maybe we could could start by defining AI in terms of what we’re talking about in manufacturing.
What are the Different AI Models?
Michael Lynch: There are a few different AI models that can be used in manufacturing. When I was doing the internet of things (IoT) at SAP, we were taking traditional AI, which was looking at vibration analysis or other analysis of machine data and using algorithms to try to do predictive maintenance. Those traditional AI models still exist, and AI research continues to make them smarter. But a few years ago, the large language models — which were originally designed for predicting the next word in a sentence — were applied to lots of different issues. The large language model itself is generative AI. And generative AI is the big change that is happening now and moving very, very quickly. Now you have ChatGPT from openAI, you have the new Gemini model, and then you have the open source model that came out of Facebook or Meta. These are now out in the world as open artificial intelligence platforms, and people are iterating very rapidly with these AI models.
I think the thing that you can take from this conversation about the different AI models is that there are going to be numerous vendors who will be able to provide you with an AI framework that’s very good. Just like you can get Amazon or Google infrastructure — Azure or AWS — where you can tap into that infrastructure via their APIs, AI tools are going to be available to everybody.
Leah Archibald: And then the next question is, how are we going to use it? Now that we have AGI or machine learning, how do we use it in manufacturing for profit?
How is AGI used today in Manufacturing?
Michael Lynch: Here’s what we are doing today with AGI in manufacturing. Look at any business process. We talked about process digitization, and you brought up the safety audits. For many companies today, a safety audit is done on paper and people type things into an Excel sheet. Then you digitize the process. Now the machine can give you a score across 25 different vectors of how you’re doing on safety audits. Let’s say you have one hundred data records.
Now if I were AI, what would I do with that? Well, I can parse those hundred records and say: Here’s a common theme in your safety problems. Here are six things you might try to do about those common problems. And every single process that can get digitized can get digitized with an AI copilot.
Once you digitize your processes, the information you get back from AGI is really tailored to that process. Instead of just going and typing into ChatGPT, “What should I do about safety audits?” With process digitization you can take your safety audits, your gemba walks, isolate manufacturing issues, and make it super practical. And then the second layer is analyzing all that data that we’re capturing. Keeping that data secure so that your proprietary data isn’t going out into a public large language model, but giving the organization insight into how they can optimize their processes, where the points of most value are, based on AI’s analysis of the data.
Data Security with Large Language Models
Leah Archibald: There’s a fear with large language models, or any kind of open artificial intelligence or open AI system, that if I can’t absolutely tell you how my data is being transformed, I’m not comfortable with sending that data out to that system.
Michael Lynch: It’s a good point. 10 years ago, all kinds of companies had their data on their own servers. They were afraid of cloud. Now thanks to secure APIs, the fear of the cloud is going away, and actually the could is more secure because with your data on site thee are people access problems that cause risks to data security. So I think the same sort of transition will happen at an AI level, where people are afraid in the beginning, but over time, these technology curves adapt to security risks faster and faster.
The fact is that the manufacturing plant of the future is going to be staffed younger people, and they’re going to demand a work environment that isn’t gray paint and old rusty machines, but highly automated and made smarter by AI. But also, the human aspect of manufacturing has to be more modern, to keep those people interested as well.
Leah Archibald: If we’ve seen anything with this new wave of AI technology, is that as much intelligence as a data system can have, what we want in the end is to have that data system talking to us like it’s a person. Like the personal relatability aspect is so strong for us as human beings, that the most technologically advanced system we can come up with is like a really smart person, an artificially intelligent person, that acts like a real person talking to us. So it doesn’t seem farfetched that the human element of creating smart systems is not just like an addendum to the system — it really is at the core of the systems that we’re trying to create.
Michael Lynch: That’s really good insight. You can see a digital transformation project that comes with a digital transformation avatar to talk to all the people about what they’re gonna go through and help them through it.
Leah Archibald: Michael Lynch, this has been such a pleasure. Thank you so much for joining me on the show today.
Michael Lynch: Really enjoyed it, Leah. Thank you.