Transforming industrial design with AI

Amid the surge of AI innovation ushered in by breakthroughs like ChatGPT, how are Generative AI and Large Language Models shaping real-world business strategies? To answer this, Nordea's Johan Trocmé turned to Hugo Nordell, a seasoned entrepreneur with roots in Silicon Valley's drone and autonomous driving sectors, and a former executive at industry giants like Sandvik and Aker Group. As the founder and CEO of Encube, his latest venture offers a collaborative design engagement platform designed to empower manufacturing firms to overcome R&D hurdles using AI-powered tools. Encube's mission is to streamline the journey from concept to market and cut unnecessary costs for new product development and introduction.

Johan Trocmé (JT): You are comfortable both in the technology and manufacturing industry domains. As a start, would you briefly describe your background? 

Hugo Nordell (HN): I come from a diverse academic background that includes physics, mathematics and economics, and I studied in the US, UK and Sweden. I’m very passionate about creating software that has a tangible impact on our physical world. At the beginning of my career, this curiosity led me to Silicon Valley, where I ended up working on autonomous driving before it was the well-known phenomenon it is today. Back then, it was far less glamorous than it is usually portrayed. For me, it mostly meant spending countless hours in overheated backseats of retrofitted cars in parking lots, perfecting the technology. 

After that, I launched several startups, including one in the drone space, and held leadership roles such as Head of Product and CTO, focusing on incorporating what we now call "machine learning" into autonomous technology and industrial applications. 

I was eventually headhunted to Sandvik, where I built the company's digital transformation for six years, overseeing numerous acquisitions and new software product launches. Before starting Encube, I joined Norwegian industrial group Aker Group as Senior VP of Digitalisation, and repositioned the group’s portfolio companies to better develop, deploy and maintain software and value-added services.

JT: Tell us about the founding of Encube – how did it all start, and what made you do it?

HN: After having experienced the industrial world from both the lens of a startup as well as that of large industrial players, I found that the same set of hardware R&D bottlenecks kept surfacing in both camps. Getting new hardware products out the door to the market is incredibly difficult. Not just in terms of the effort involved – that is largely blood, sweat and tears. It is the way the hardware development process itself runs that is the issue. There is very little automation in place for day-to-day work to help people collaborate better. The software in use isn’t nearly as practical, timesaving and productivity inducing as you’d expect. It’s in the way, more than it is helpful. As a result, incredibly talented engineers waste time on mundane tasks that don’t drive the hardware process forward. This lack of collaborative mechanisms in software tooling makes hardware R&D fraught with human error and causes significant lead times and project cost overruns. The product design that makes it into production is too often not nearly as cost efficient as it could be. Large companies are capable of swallowing these excess costs in the short term, and find productivity in manufacturing over time. But for hardware startups, this fragmented way of working and collaborating is often a death sentence.

It dawned on me that the manufacturing world is trying to address these hardware design problems from the wrong end of the value chain. Making large capex investments to drive manufacturing automation does little to improve the situation if you’re pushing poorly designed hardware requirements through your shop floor. At worst, it will dig you deeper into the hole you’re trying to get out of. The problems have to be addressed at the root, which is upstream, in the hardware design process itself. And the starting point of that spells better design and requires collaboration between engineering, manufacturing and procurement. Using software that is inherently designed to shift what you as engineers spend your time on day-to-day. As it turns out, generative AI, large language models and traditional robust process automation are great tools to help accomplish this.

We founded Encube on the core premise and philosophy that people ultimately drive business processes. So, any software we build must put people at the centre of the experience, deliberately designed to help them make better, faster and more data-driven decisions. Not to replace the human in the loop, but to fundamentally empower people in hardware engineering to spend their time solving the hard problems that brought them into engineering in the first place. Not to waste their time identifying potential manufacturing issues and cost drivers, but what to do about them.

AI, deployed meaningfully, is exceptionally good at doing repetitive work accurately, over and over. Often with significant time savings as a result. By deploying many finely tuned AI models to solve important but mundane and repetitive tasks in the hardware engineering cycle, Encube can free the hardware team to do what it does best: look at the big picture.

This has been a key learning from the very start for us. Instead of trying to build a single big AI model that should do everything, divide-and-conquer the problem into smaller subsets. Identify the smallest common denominator and tailor an AI model to manage that, and nothing else. One AI for one task. Many AIs to accomplish something overall.

This approach to developing solutions is cost-efficient, fast and makes it much easier to quickly understand if something scales beyond a controlled environment. It reduces technology risk and allows everyone involved to continuously see and experience tangible improvements along the way, rather than have their fingers crossed for the occasional big leap. It also makes it much easier to tie development to tangible and measurable business goals, such as lead time reduction or cost per part. Process governance is only possible with measurable metrics, after all.

In manufacturing, this is commonly known as continuous improvement, and it absolutely scales to AI development and deployment, too. I’m convinced that many industrial organisations will find much more consistent and positive results if they decide to adopt this way of engaging with AI, instead of the boil-the-ocean approach I see many pursue in the fear of missing out.

JT: Can you describe the Encube platform? How can it change the process for industrialisation of products?

HN: Encube is designed to help large manufacturers tackle three big hardware development challenges: accelerating talent scarcity, growing sustainability targets and increasing product complexity. We're seeing a shift where the seasoned engineers with decades of production experience are retiring, and new talent familiar with production is harder to come by, especially in North America and Western Europe. This talent gap, along with stagnant productivity growth in manufacturing, is a real issue. How do you remain competitive and take new products to market when the talent who knows what "good" looks like is increasingly missing from the conversation?

Sustainability is also a key focus, driven by new laws like the US Inflation Reduction Act and the need for more resilient supply chains. The world is rapidly moving from a point where sustainability is a differentiator, to where it is a licence to operate. If you cannot show your climate footprint already at the product design phase, you might soon not be allowed to take the product to market. On top of this, companies are actively investing to shift manufacturing capacity away from China, to become more domestic and regional. Which further accelerates the talent gap.

Finally, the product complexity is growing rapidly, particularly in sectors like aerospace and automotive, driven by increasing end-customer performance demands. This complexity growth is already outpacing the ability of current software, making it tougher than ever to go from product idea to market launch.

Encube aims to drastically cut down the time R&D teams need to go from concept to validation, which currently can take years. By using automation and AI, we enable quicker prototyping and a faster evaluation of ideas, while minimising early-stage design and requirements issues that might otherwise result in delayed production start or prohibitive manufacturing costs.

Our software is a collaborative design review and engagement tool. It unites R&D teams around a single source of truth, 3D models and 2D technical drawings. We make it incredibly easy to build a digital red thread of how product development evolves over time, while tracking key decisions and action items close to the designs these relate to. We embed a set of AI capabilities into the design review workflow that help teams focus discussions and decision-making based on automatically identifying possible design and manufacturing issues already at the point of design. Ultimately, this empowers hardware teams to drive product development more effectively, with shorter time to market, lower R&D costs and more competitive cost of manufacturing as a result. With Encube, AI is a tool that empowers data-driven decision-making, not a means unto itself.

JT: Could you explain what role AI plays in your platform? How do you see it evolving, and would it have been possible to launch Encube with the capabilities it offers today five years ago?

HN: AI is foundational to Encube, acting as both a co-pilot for R&D teams and an analyst of complex data. It's like a digital roundtable where AI assists in sifting through extensive conversations, analyses and technical data, streamlining collaborative efforts significantly. Doing the mission critical but menial and tedious work that no one else on the team wants to do. Our AIs examine 3D models and technical drawings to assess manufacturability, comparing designs against ISO standards and flag inconsistencies for the team to review and decide what to do about.

Moreover, we have an AI that simulates actual manufacturing processes, including 3-and 5-axis CNC machining. It generates real manufacturing instructions in seconds and accounts for complex factors such as physical cutting forces and machine capabilities. We call this "physics-informed AI", and it is instrumental in being able to offer customers insights on key questions such as how hard it will be to manufacture a product, and what the cost and climate drivers will be already at the point of design.

We use a range of AI technologies to achieve these use cases, including large language models and computer vision. These are deeply embedded into our offering from day one. They blend seamlessly into the overall user experience and are actively designed to stay out of the way when users aren’t engaging with them. This makes AI more approachable and tangible, because it puts human productivity and ingenuity at the centre of the hardware design process. It doesn’t try to replace it.

Five years ago, launching Encube with our current capabilities wouldn't have been possible. The AI technology we rely on, particularly in terms of generative AI and Large Language Models, weren’t readily available even a year ago. But it has already had a profound impact on how we design and deliver value to manufacturing customers.

JT: What are the key potential benefits for manufacturers from using Encube for their product development, compared with traditional processes? Are you able to give some examples to quantify the upside?

HN: Using Encube offers manufacturers several key benefits, both in the hardware development phase and in manufacturing, compared to traditional product development tools. Particularly regarding cost savings and productivity.

Our AI-powered collaborative design engagement platform has shown to reduce the time-to-market by 20-50% for complex hardware development. These numbers are significant, as R&D processes are often constrained by tight budgets and limited time frames. Delays are costly, reduce competitiveness and lead to excess R&D spending that puts the original business case on the line. Especially when delays threaten start of production.

In the production phase, it's not uncommon for cost per part to exceed initial estimates by 25-30%. Large companies can survive this due to their size and financial resources and are often able to systematically tackle these excess costs over time. Startups, on the other hand, can find such cost overruns fatal. Encube's manufacturability analysis AI addresses this by assessing if and how a component can be made and identifies key manufacturing cost drivers already at product design time, so hardware teams can avoid them entirely.

Moreover, by streamlining the path to market, Encube makes it possible for manufacturers to meaningfully accelerate their product release cycle. This leads to faster product iterations, fewer delays to market and a stronger innovation flywheel. All of which translates into competitive strength.

JT: How do you think deployment of AI will affect manufacturing industries in terms of productivity and human employment?

HN: As a bit of a history enthusiast, I'm quite interested in how new technology paradigms can spark economic growth, create new markets and shape society. AI's impact on productivity is still unfolding; it's a powerful tool, but it doesn't guarantee quality output. While AI's revolutionary potential is clear, especially for generative AI and Large Language Models in particular, we're just beginning to explore its applications. For instance, while generative AI has impressive capabilities, it's also incredibly good at generating incredible amounts of low-quality output. Producing more of something doesn’t necessarily imply productivity growth, even if many people conflate the two. As an executive, I certainly don’t want more PowerPoint slides created by my organisation. I want far fewer, better worded, narratives that help shape business direction.

I think the transformative impact of AI on employment is exaggerated. We will see a period of rapid change, sometimes with disruptive effects on labour markets, but I think it's more likely that AI will create entirely new jobs than that it will eliminate them outright. In the beginning of the 1900s, close to 40% of the American population worked in agriculture. Today, less than 2% do. The tremendous productivity brought forth by automation in agriculture destroyed jobs in the short term. But, over time, this allowed a number of other markets to thrive and grow in a way that wouldn’t have been possible without automation. I think it’s far too early to bet against human ingenuity and our ability to invent new jobs and markets.