Uli Palli Of Accella AI: How AI Is Disrupting Our Industry, and What We Can Do About It

An Interview With Cynthia Corsetti

Give AI a chance to prove its value. Start slow, find an easy application — often quality control is a good starting point — and do a proof-of-concept study to demonstrate the value of AI to yourself and your colleagues. Nothing is as convincing as seeing AI in action.

Artificial Intelligence is no longer the future; it is the present. It’s reshaping landscapes, altering industries, and transforming the way we live and work. With its rapid advancement, AI is causing disruption — for better or worse — in every field imaginable. While it promises efficiency and growth, it also brings challenges and uncertainties that professionals and businesses must navigate. What can one do to pivot if AI is disrupting their industry? As part of this series, we had the pleasure of interviewing Uli Palli.

Uli Palli, is the CEO and Chief Data scientist at Accella AI, a company dedicated to develop AI-based solutions for manufacturing companies. Before founding Accella AI, Uli worked as an advisor to high-tech and manufacturing companies, such as NetApp, Duracell, Cisco, Xlinx, HP, focusing on the implementation large-scale initiatives.

Prior to that he was a senior manager at Deloitte Consulting working in their manufacturing industry practice. He has a Bachelor of Business Administration from Karl-Franzens University in Graz, Austria and an MBA from Heriot-Watt University in Scotland.

Thank you so much for joining us in this interview series. Before we dive into our discussion our readers would love to “get to know you” a bit better. Can you share with us the backstory about what brought you to your specific career path?

To say I knew ever since I was a little boy that I wanted to work in technology would be a bit of an overstatement. I grew up in an area of Austria that doesn’t look like your stereotypical “Sound of Music” setting. We had a coal mine, a steel mill and even now, with the coal mine closed for many years, there is still a lot of manufacturing in the area. This wasn’t the place I wanted to get stuck so I studied French and Italian; being a translator seemed like a good way to get out and see the world. My life changed when I bought my first computer — I realized that I love technology, love coding, and love developing solutions that make life easier. My passion for technology eventually led me to Silicon Valley, where I worked for and consulted to a number of the big names in technology with focus on their manufacturing verticals.

I have been interested in AI since the 90s, back when, unfortunately, we did not have the raw computing power needed to implement these models. When this changed in the mid 2010s a couple of friends and myself decided to take the plunge and combine our deep knowledge of technology, familiarity with manufacturing, and our interest in AI to start a company that develops AI-based solutions custom-tailored to the needs of manufacturers.

What do you think makes your company stand out? Can you share a story?

What makes our company, Accella AI, stand out is that we develop AI-powered solutions exclusively for manufacturing companies and even more specifically for the shop floor. We focus on solutions for quality control and predictive maintenance as well as a handful of other very specific applications.

Our solutions, the Accella AI Bot and the Accella Quality Box, were developed in very close collaboration with our biggest customer — a very well-known company, a household name, which I unfortunately cannot disclose. That very close collaboration ensured that we built solutions that really address the pain points of manufacturers producing consumer goods in very high volume. The first application we developed was for quality control in manufacturing plants. Once you can detect a defective product with an accuracy of 99.996% every 50 milliseconds — or 1,200 products per minute — you feel fairly confident that you can handle other industry-specific challenges.

For the proudest moment in our company’s history, I sadly wasn’t present, but our customer relayed the story. Like all well-known companies, they were approached by one of the big AI players (again no names, sorry) who showed them their solution for manufacturing. Our customer looked at it and said: “Let me show you what we got,” and gave them a high-level overview of the solution we developed. Apparently, the salesperson of the big-name company just stared at it and said: “I have never seen anything like that!”

You are a successful business leader. Which three character traits do you think were most instrumental to your success? Can you please share a story or example for each?

The first and probably most important characteristic — and I am sure I am not the only one listing this right at the top — is persistence, or rather dogged determination. We are doing things nobody has done before and there is no how-to manual or expert that can help us deal with challenges we encounter. I still vividly remember the first time I trained a model to differentiate good from defective products. I single-handedly annotated more than 50,000 images. To do so I had to open each image, look at the product, make the determination, whether it is ok or not ok (NOK) and, if it is NOK, decide which of seven different defects it displayed and annotate the image accordingly (this last step was done in collaboration with internal experts). I then sorted the annotated images into folders that were used to train the model. Imagine staring at black and white pictures of days — or rather weeks on end. My family still refers to this time as my zombie period.

I have a lot of experience with technology, different programming languages and experience implementing technology-based solutions, but AI required new skills: I had to pick up Python, the newest popular coding language — and the language of AI. I had to immerse myself in data science and work my way through thick math books full of intimidating formulas. I credit my never-ending desire to learn and my curiosity for the fact that I didn’t quit.

The story I mentioned above about manually annotating 50,000 images is actually also a good example for that desire to learn as well: I honestly never wanted to have to do this again and so we worked hard to develop a much faster, better, and cheaper way of achieving the same results.

The third ingredient is passion for what I do. Not just for technology but also manufacturing. I know this industry, I truly value the contribution manufacturers make to our daily lives, I understand their challenges and I deeply relate to the people doing this work. When I am on the shop floor at one of our customers I am with my kind of people. Without this passion I would be just another vendor, but because of it I am a true partner to our customers.

Let’s now move to the main point of our discussion about AI. Can you explain how AI is disrupting your industry? Is this disruption hurting or helping your bottom line?

I started and currently run a company developing AI solutions for manufacturing operations so I might be a bit biased when I say that AI will have an enormous impact on manufacturing. I wholeheartedly believe that AI is the future of manufacturing. It is a powerful tool that is ideally suited to solve some of the more intractable issues that challenge manufacturers. The question is not whether manufacturers adopt AI, but when; will they do it sooner and get a competitive advantage, or will they wait to adopt until they are forced to in order to avoid their competitors leaving them in the dust?

There are many ways AI can disrupt manufacturing. I’d like to give two key examples from our own experience.

Quality control — the current methods of making sure products meet quality standards are often not up to the task: manual QC is slow, expensive, inaccurate and often completely infeasible. Think of the example I mentioned before: if you make 1,200 products a minute you can’t afford to hire people to check each one so you check samples at the risk of missing defective products. Even at lower volumes, people are generally not a good option: they are not good at focusing on a single task for 8 hours at a time, they get tired, make mistakes and often don’t enjoy their work. It is very hard to hire people for these repetitive jobs.

Traditional computer vision systems have other challenges: they are inflexible, every change in a product or a new product requires rewriting the code, the equipment is very expensive and still doesn’t meet high-speed requirements.

Enter AI; specifically, machine learning. These algorithms learn similarly to humans by seeing examples of good and defective products. Once trained properly they do their job 24/7/52 with consistent high accuracy. They learn to categorize defects, and even find whole new defect categories. They can adjust for environmental factors, such as different lighting conditions over the course of the day, and when you change product specs or add a new product all you need to do is retrain them with more images. In addition, they don’t require high end equipment, reasonably cheap cameras and sensors are good enough. This allows you to add quality stations at several points during production, not just at the end and remove defective products early before more time, material and energy are spent on them.

Predictive maintenance is another perfect use case for AI. If you feed the algorithms enough data about your critical equipment, such as pumps that are used to dispense a liquid and have a tendency to clog, add the information from existing maintenance records and let the AI do their pattern recognition magic, you will learn a lot about what causes unexpected failures. On a very practical note, the AI model can, based on its training, determine which pumps are most in need of maintenance and which ones don’t need attention. So now, instead of the maintenance crew routinely maintaining pumps 1 through 15 because it is the first Wednesday of the month, they can focus on those most likely to fail and not worry about the others.

This really helps streamlining maintenance and saves a lot of money because unplanned line downtimes caused by defective equipment can be very expensive.

Let me quickly mention two additional areas: predictive analytics and HMI simplification.

Predictive analytics is similar to predictive maintenance: it lets you analyze historical and current data and make predictions about the future. In manufacturing predictive analytics can be used to predict characteristics of components based on input materials and external factors. For example, if you manufacture gels the ambient temperature and/or a small change in a raw material can have an effect on the viscosity of the gel. Since you want your gel neither too thick nor too runny, AI can help you figure out how to adjust the process to ensure that the gel has just the right consistency.

AI can also help you simplify complex human machine interfaces (HMIs). HMIs for industrial machinery can be very complex and operators with years of experience might spend many hours setting up a machine for production. This can make small production runs uneconomical. If it takes you 10 hours to set up the machine (which is not unusual for more complex ones) then you’d better run it for a very long time or you will lose money. AI can help here: the algos can simplify setup by identifying the most critical of a large number of processes, materials, and product parameters and make recommendations that cut down the time it takes by 50 or more percent.

These are the applications we focus on, but there are others relevant to manufacturing, e.g. related to forecasting and supply chain optimization.

At our company, Accella AI, AI is what we do, it is in our DNA and we thrive on it. However, manufacturers as a whole are typically rather cautious and price-sensitive. So, we are making a bet that manufacturers realize the transformative power of AI, the fact that AI will save them time, energy, raw materials, allows them to reduce scrap, avoid unplanned downtimes, get more use out for their equipment and are able to deliver high quality products to their customers.

Which specific AI technology has had the most significant impact on your industry?

If I had to single out AI technology with the biggest impact right now I would pick machine learning, especially deep learning algorithms. What makes these algorithms so powerful is the fact that they analyze data and learn to make predictions and decisions without being explicitly programmed. They can learn from past mistakes and are really good at processing images. The way they learn is similar to how a human learns: think of a child who learns to tell cats from dogs. You don’t provide them with lengthy definitions, you just tell them “Look, at the dog over there” and if they make a mistake, you tell them “This is a small dog, not a kitty”. Eventually the child will get it right every single time.

This is how AI learns to tell a good frozen pizza, tire rim, or concrete paver from one with too few slices of pepperoni, a scratch on the surface, or a missing corner.

When picking emerging AI technology that will have a significant impact, I have to acknowledge the current hype around generative AI. We are just barely scratching the surface with regards to what these models can do and already see fascinating examples. Internally we use generative AI to help us write and design marketing materials, a friend recently told me that he used ChatGPT to help him write proposed legislation, and another friend creates amazing artwork.

In manufacturing, I can see many different applications from asking the LLM simple but important questions like “How many defects of category 3 did we have in the last 2 weeks?” or “When was pump 15 last repaired and what was the reason for the repair?” to much more complex tasks such as gathering and cleaning data used to train new AI models.

Can you share a pivotal moment when you recognized the profound impact AI would have on your sector?

When we first started developing AI models with our customer it was a real leap of faith from their end and I cannot give them enough credit for being AI visionaries. Of course, I believed that AI would have a profound impact but we had to prove it.

So, we built the first model, trained it up with the now infamous 50,000 images and put it to the test. The result was astounding: the model worked almost flawlessly, we only found a few instances where the human assessment of the defect differed from the call AI made. To improve the model, we looked at these discrepancies in detail and what we found was even more exciting: in pretty much all cases, the AI had actually made the correct call and the assessment made by the human expert was incorrect. After just one training round the model did better than highly experienced people.

After that I could look everybody who questioned the ability of AI to do QC right in the eye and say “You are wrong, AI can do this, easily.”

How are you preparing your workforce for the integration of AI, and what skills do you believe will be most valuable in an AI-enhanced future?

Instead of answering this question for Accella AI, let me comment on this from the perspective of our manufacturing customers.

This is a hard nut to crack for manufacturers and at the same time it is an opportunity. Manufacturing is a tough business, the price pressure is huge, competition from abroad is strong, and finding qualified personnel is often the proverbial search for the needle in the haystack.

Let’s face it, manufacturing is not particularly sexy, and as a manufacturer your chances of competing with the likes of Microsoft, Facebook, or Apple for data scientists are poor. You simply can’t afford the salaries even fairly junior people command and therefore buying AI expertise is not an option — you need to work with the talent you have, and build on their knowledge and expertise. Nobody knows products and processes better than employees, and therefore these employees need to be part of any AI implementation from day 1.

In addition, new expertise needs to be built and in an ideal world that critical expertise would accumulate in-house and not at some external supplier. Building that expertise also provides exciting upskilling opportunities for existing employees and robust AI initiatives may help to make a company more desirable to job seekers.

Now, of course, you can’t just walk into the shop floor one day, round up five people and tell them that they now are in charge of AI. You have to make it feasible, even easy for people who are interested and ambitious to go up that learning curve. This is exactly where our company comes in, we have built the platform, wrote all the software needed to integrate AI with existing systems and have made it easy even for AI novices to learn how to train a model, deploy it to many lines and multiple plants and to coordinate what we believe eventually will be hundreds or even thousands of models.

This gives manufacturers control over their AI-based solutions, helps them build the necessary expertise in house so that in the future they don’t depend on expensive service providers every time they want to add a new AI solution.

What are the biggest challenges in upskilling your workforce for an AI-centric future?

Again, I’ll provide the answer from the perspective of our customers in the manufacturing industry. For them, hiring and retaining employees already is a steep uphill battle. Estimates vary, but I have seen reports saying that more than 600,000 job openings are unfilled in manufacturing in the US, and Europe isn’t doing much better.

Upskilling is a tough challenge especially when you start with too few people anyway. On the other hand, offering these new opportunities can make you more attractive especially to younger workers and therefore help with hiring. Younger people — I see this with my son — are generally completely unafraid when it comes to new technology. While I still contemplate what to do, my son has already grabbed the mouse, clicked here and there, typed a few words at breakneck speed and has gotten an answer (mind you, not always the correct one). It is this intuitive understanding of how to work with technology that manufacturers can tap into and offer young employees and new recruits the opportunity to be part of the next big thing: AI.

What ethical considerations does AI introduce into your industry, and how are you tackling these concerns?

Ethical considerations related to AI are very important and I do not want to downplay the need to have robust discussions about the topic. In manufacturing as opposed to many other industries, however, I think the ethical ramifications are less serious.

There is the risk of job losses but as I mentioned before, having to lay off people is not the problem most manufacturers face. In fact, AI can help to address staffing shortages by doing jobs nobody wants to do. Does that mean that nobody will lose their job? Of course not, but at least short- to mid-term, the risk is much lower than in many other industries.

There are also few concerns around potential pitfalls of AI such as biased models These issues should certainly not be taken lightly in any field AI is applied to. However, compared to industries, e.g. pharmaceuticals, the consequences are generally less dire in manufacturing. While it’s not great if a biased model overlooks a defect, it’s a lot less harmful than if you exclude a group of people from receiving a novel anti-cancer drug. At least for right now, ethical concerns around AI in manufacturing do not cause me to lose any sleep.

What are your “Five Things You Need To Do, If AI Is Disrupting Your Industry”?

My top five things leaders in manufacturing, QC and operations need to do is:

1. Familiarize yourself with AI. There is a lot of hype out there and it is getting really difficult to follow everything that’s going on. The good thing is that you don’t have to, just focus on learning what AI can do for manufacturing, how it can benefit your company, your department, and how it can help you solve concrete problems. It’s intimidating at first but it’s actually not so difficult to understand the basic principles. I have seen the transformation myself: people who used to look puzzled and uncomfortable when AI was mentioned now talk about training data sets and the latest validation as if they had done this all their lives.

2. Give AI a chance to prove its value. Start slow, find an easy application — often quality control is a good starting point — and do a proof-of-concept study to demonstrate the value of AI to yourself and your colleagues. Nothing is as convincing as seeing AI in action.

3. Plan ahead to ensure success. You will need data, technology and people to make AI work. Most companies do not have the data they need so it needs to be collected. This is a key reason we recommend starting with quality control, if you manufacture at reasonable throughput, you can collect enough data in a few days to train the first model. Technology is another area: you need the infrastructure to transmit images, enough disc space to store them, etc. People are possibly the hardest challenge; for example, you need IT/OT willing to work with you. I remember a project last year where IT dragged their feet on giving us remote access to their internal system for more than three months. When we finally got it, we had one week left to collect data before the seasonal project came to an end.

4. Don’t reinvent the wheel. While you want to develop AI skills in house and eventually be able to train and deploy your own models you don’t want to start at zero. Companies like ours have developed platforms that make integration with your internal systems, e.g. your PLCs and MES, very easy, and allow for quick deployment on many lines and across plants. I have had several initial conversations with companies that took a quick look at our capabilities and said “we’ll just hire a few data science students and build a solution ourselves.” To this day I have yet to see one of them actually deploy an AI solution.

5. Don’t delay. I keep saying and I wholeheartedly believe that the question is not whether manufacturers will adopt AI but when they will do so. They can start now and eke out a competitive advantage or they can wait and adopt late to avoid a competitive disadvantage. From where I stand it makes a lot more sense to get the process started now.

What are the most common misconceptions about AI within your industry, and how do you address them?

The industry right now seems to oscillate between two extremes: on the one hand there is the notion that AI is complicated, newfangled stuff that we didn’t need so far and can do without and on the other hand there is real FOMO which leads to companies shooting from the hip to “do something with AI”.

Neither of these extremes is helpful. What we need is a measured, logical, step-by-step approach to AI implementation in manufacturing. I think we will see a balance as more implementations happen, more use cases are publicized, more manufacturing leaders talk about their experiences. We try to make a contribution by talking about our approach and the success stories as well as the challenges we see to paint a realistic picture.

Can you please give us your favorite “Life Lesson Quote”? Do you have a story about how that was relevant in your life?

I am reaching way back to the year 1314 in quoting Robert the Bruce, King of Scotland: “If at first you don’t succeed try, try and try again.” I think this works very well with the persistence I talked about before but also reflects optimism and a belief in ultimate success.

Off-topic, but I’m curious. As someone steering the ship, what thoughts or concerns often keep you awake at night? How do those thoughts influence your daily decision-making process?

It’s not the big things that keep me up at night, but the little annoyances that come with still being deeply involved in developing our solutions: the bug we can’t find, the failure code we can’t explain, finding a way to access the data at a customer site without literally driving there and copying them, a way to squeeze a little more performance out of a model to meet a customer need. In the end I am still the geek I always was and my mind can’t stop looking for solutions to technical problems.

If you could start a movement that would bring the most amount of good to the most amount of people, what would that be? You never know what your idea could trigger.

I am passionate about conservation and one of the important aspects in that context is personal responsibility. My movement would be one that is centered around adopting a sustainable lifestyle, teaching people how to avoid waste and excess, and show that smaller is better, less is more.

Of course, the question how we can use AI to create a more sustainable world is always on my mind and hopefully what we are doing at Accella AI helps by allowing companies to use less energy and reduce waste.

I try and walk the talk and you’ll be surprised to find that as a self-declared geek, I have an iPhone 6, a second-generation iPad, one small TV, a 2011 car we bought used and a cheap pad to take on hikes and to the beach so my good iPad doesn’t get damaged or stolen. I have also started shopping for clothing second hand to do my part to keep clothing out of the landfill.

How can our readers further follow you online?

You can find me on LinkedIn (https://www.linkedin.com/in/uli-palli/) and you can also find our company there (https://www.linkedin.com/company/accella-ai), our webpage (https://www.accella.ai/) has good information and use cases for those who want to start looking into using AI in manufacturing.

Thank you for the time you spent sharing these fantastic insights. We wish you only continued success in your great work!

About the Interviewer: Cynthia Corsetti is an esteemed executive coach with over two decades in corporate leadership and 11 years in executive coaching. Author of the upcoming book, “Dark Drivers,” she guides high-performing professionals and Fortune 500 firms to recognize and manage underlying influences affecting their leadership. Beyond individual coaching, Cynthia offers a 6-month executive transition program and partners with organizations to nurture the next wave of leadership excellence.