Generative AI for Product Managers: Superpower for Industrial Design

But with great power comes great responsibility

Abstract

Generative AI is revolutionizing product management, empowering teams to conceive and develop groundbreaking products. However, to fully harness AI's potential and mitigate risks, human expertise remains indispensable throughout the product development journey. This is particularly crucial in the concept and design phases, where AI-generated ideas must be evaluated against real-world constraints. Ultimately, it's human wisdom and sentience, interacting with AI's computational power that leads to innovative, safe, compliant, and appealing products.

Introduction

Product managers (PMs) constantly navigate a complex landscape of challenges, from accelerating innovation and ensuring quality to driving revenue growth and outpacing competitors. The pressure to do more with less, while deciphering vast amounts of data, can be overwhelming.

Generative AI emerges as a powerful ally for PMs, offering solutions to these challenges across the product lifecycle. Recent data suggests a growing adoption of AI among PMs, with a 2022 IBM study revealing that 21% were already utilizing AI daily.(1) It's likely this trend has accelerated since then.

However, to truly harness the potential of AI and achieve real business value safely, it is crucial to recognize its role as an enablement tool. Its power is most effective and least risky when managed by human experts throughout the product management process. This principle will be underscored in a series of posts highlighting the usage of AI in different phases and tasks within the product management lifecycle. This post will focus on the design phase of product management, specifically hardware design, to explore how generative AI is transforming outcomes and how human experts are using its power to deliver startling results while mitigating risks.

Note - I speak of both product managers and product designers in this discussion, recognizing the need for design expertise as well as the need for the product manager to articulate the product vision and serve as the voice of the customer in the design process, expressing their pain and needs, yet providing guardrails by keeping an eye on profitability and growth goals.

Gen AI and Industrial Design: A Creative Spark

Product Managers can influence great designs by identifying customer needs and clearly communicating product performance requirements to the design team. The design of a part or product typically undergoes several iterations of feedback and tradeoff analysis. This iterative process of concept development, refinement, and testing is time-consuming. Additionally, it can suffer from a ‘lowest common denominator’ effect that may stifle innovation as preferences from powerful voices are added to the mix.

The beauty of incorporating generative AI tools into this process lies in their ability to quickly produce new concepts from fresh market and user insights. These AI-generated concepts may be free from conventional restrictions, such as the mindset of ‘we always do it this way,’ leading to startling results that capture customers' attention. The traditional approach often results in incremental improvements in visual appeal, user experience, performance, and cost, while the generative AI-infused approach might introduce the kind of market impact product managers dream about.

An emerging design practice blends generative AI outputs with human expertise by feeding user research to the model to generate novel concepts early in the design process. Research data in multiple formats and modes—such as verbal recordings from sales calls, chats from Customer Support bots, survey responses, and social media comments—can be input into the AI model to generate concept images. Tools like Midjourney, OpenAI’s ChatGPT and DALL-E 3 ,and Stable Diffusion can be used in various situations and to varying degrees for these tasks. The Board of Innovation, a consulting firm specializing in product innovation, provides a comparison of leading AI-image generators in this article.

However, AI image generators often produce ‘hallucinations’—concepts that may lack aesthetic appeal, usability, safety requirements, or manufacturability within certain cost constraints. Another issue is that early AI-generated concepts often lack standard components, resulting in performance issues and non-compliance with industry, regulatory, and environmental standards. For example, in the realm of AI-enabled 3D-printed medical devices such as implants or stents, dramatic improvements in both fit and cycle time are occurring (2), yet early stages of designing patient-specific implants can prioritize unique geometries for factors like bone integration or weight reduction. These designs sometimes can overlook crucial considerations such as biocompatibility, mechanical strength, and regulatory compliance.(3)

In addition to medical devices, AI-driven design practices are appearing across a broad span of industries. For instance, Jaguar Land Rover (JLR) uses generative AI tools that leverage the digital twin capability in NVIDIA’s Omniverse platform to render high-fidelity images of its vehicles in various environments.(4) This has been effective for the rapid creation of digital assets used in advertising. The next application may involve designing a new vehicle using AI-conceived designs, visualizing them in various environments, conditions, and user behaviors. JLR is also exploring having customers use configuration tools to create customized designs and interact with generative AI via voice prompts to find the right personalized mix of vehicle features. Similarly, BMW North America is using their generative AI platform, EKHO (Enterprise Knowledge Harmonizer and Orchestrator), to provide feasible sets of options for vehicle customizations in minutes, streamlining the customer experience.(5)

In the footwear industry, companies like Adidas have used generative design to create innovative shoe designs.(6) By inputting performance data and aesthetic preferences, the AI generates unique lattice structures for midsoles that provide optimal cushioning and support. In the medical device industry, generative design is also being used to create customized medical implants and prosthetics. By analyzing user-specific data, AI can generate designs that perfectly fit an individual's anatomy, improving comfort and functionality.

In aerospace, Airbus has collaborated with Autodesk and their Dreamcatcher software to develop generative design tools for aircraft components. In one notable example, they used AI to design a bionic partition for the A320 aircraft.(7) This partition, inspired by the structure of bones, is significantly lighter than traditional designs while maintaining the required strength and stiffness. This not only reduces the overall weight of the aircraft, leading to fuel savings and reduced emissions, but also demonstrates the potential of generative AI to create more efficient and sustainable designs.

In another case, a recent McKinsey report cited a situation where industrial designers at an automotive OEM created 25 variations of a new car dashboard in two hours using a touch screen interface.(8) These designs then required refining by a designer employing iterative prompts with image-editing software to produce more presentable images with varying form factors, materials, and colors so customers could see a more realistic look and feel before providing feedback.

Good AI hygiene in product design, therefore, requires that human experts review AI-generated concepts before incorporating them into CAD systems. These experts must judiciously adjust outcomes through prompts about required specifications and standards compliance. Once satisfied, the leading concepts are incorporated into the customer feedback process to fine-tune or eliminate certain features or aesthetics. This cycle not only ramps up innovation but also gains market affirmation in a fraction of the time typical for these cycles. McKinsey reports seeing more than a 70% improvement in product development cycle times through these practices.(9)

From Dream to Reality: Manufacturability of AI-inspired Designs

While Gen AI tools that enable text-to-image creation and iterative voice prompting can rapidly produce fascinating and novel designs, getting to manufacturability takes more technology and more expertise. CAD (Computer-Aided Design) and CAE (Computer-Aided Engineering) systems are increasingly incorporating AI and machine learning algorithms to streamline design processes, optimize geometries, and even suggest design improvements based on performance simulations. These systems are also becoming more aware of manufacturing constraints, such as material properties, fabrication limitations, and industry standards. Some CAD software, like Fusion 360, now include built-in tools for Design for Manufacturing (DFM) analysis, which helps designers create parts that are easier and more cost-effective to manufacture. This requires not only the assessment of manufacturability but also the ability to generate manufacturing instructions or toolpaths for CNC machining or 3D printing.? These technologies are helping product teams advance towards a more seamless workflow where AI-generated designs can be translated into physical prototypes for testing and validation.

To increase the likelihood of the manufacturability of designs, AI Engineers often create vector embeddings as a way to represent complex data, like design geometries or material properties, in a mathematical format that AI models can understand and process.? Machine Learning Engineers then train and test the AI models to assess the feasibility of the build and compliance with industry standards.? This can be done by incorporating data from various sources, such as material databases, manufacturing process specifications, and industry-specific standards databases. This teaches a vital lesson for product managers - with AI the outcomes are a product of the data sourced driven by prompt sequences. Product managers and designers must be vigilant in seeking to understand, and validate, the underlying data that produced the concept. In my own experience, I have been surprised to learn of the sources, or lack thereof, of AI-generated results.

Other CAD/CAM/CAE suites are also rapidly evolving to reduce the iterations between design and prototype. Siemens NX Software leverages AI to generate optimized geometries for complex components in industries like aerospace and automotive. Here, human designers work in tandem with the AI, refining the designs and ensuring they meet real-world manufacturing constraints. nTopology is another platform that empowers engineers to create intricate designs for 3D printing. By inputting functional requirements, the AI generates optimized structures, while human experts ensure that the final designs are both innovative and practical.

In terms of ramping up the power and speed needed to make this truly an interactive and iterative process, Luminary Cloud is advancing the capabilities of rapid engineering to keep up with the proliferation of AI-inspired designs. Their cloud-based CAE simulation software, powered by NVIDIA’s GPUS, can ingest CAD files, generate meshes, run solutions, and extract for visualization in a fraction of the time usually required to create realistic simulations.(10) One can see how the power of the latest GPUs coupled with new GenAI design capabilities are knitting together a design workflow that will eventually enable rapid iterations in the research to design to engineering workflows. Yet, even with these advances it seems a high degree of expertise will be needed to modify meshes, validate the calculations, and interpret the data.

From Reality to Marketability: Compliance with Industry Standards

Many industries are heavily regulated and have stringent standards and specifications set by organizations like the Society of Automotive Engineers (SAE), International Organization for Standardization (ISO), Underwriters Laboratories (UL), ASTM International, and the American National Standards Institute (ANSI). Incorporating these standards into generative AI-powered design processes is crucial to ensure safety, performance, regulatory compliance, interoperability, and environmental sustainability.

Product companies, CAD and CAE software firms, and standards organizations are responding by developing integration capabilities into the design and engineering workflows. Those capabilities include API integrations that connect generative AI design tools directly to standards databases. The APIs allow the AI models to access and reference the latest standards and specifications in real-time during the design process. This ensures that the generated designs adhere to the necessary guidelines and requirements, reducing the risk of non-compliance and costly rework later in the development cycle. REST (Representational State Transfer) APIs are simple, flexible, and easy to integrate with different platforms and languages. GraphQL is another API option that allows CAE software to request only the specific data needed, minimizing data transfer and improving performance. API integrations can occur by partnering with specialist boutique firms such as IHS Markit API (part of S&P Global) or Techstreet who offer solutions for seamless integration of standards into software platforms, or well-known IT consultancies such as Tata Consultancy Services and Accenture.

Other possibilities for addressing standards compliance in an AI-driven design and engineering process include rule-based validation in which companies automatically check designs against the relevant standards using pre-defined constraints based on standards to flag deviations. Machine learning solutions can also help companies develop models to learn and predict compliance. These models can be trained on both compliant and non-compliant designs allowing them to identify patterns and features associated with compliance and can be continually refined. Finally, companies can use simulation and testing software, such as Luminary Cloud, to validate the performance and safety of AI-generated designs under various conditions. These simulations can incorporate real-world scenarios and test the designs by integrating the standards databases to ensure they meet the required performance criteria.

In this new landscape, engineering expertise remains essential, orchestrating the use of AI and interacting with software and published standards to identify and rectify problems before they are implemented in prototypes. This human-AI collaboration ensures that products are not only innovative but also safe, reliable, compliant, sustainable, and interoperable with other components and subassemblies.

Summary

-? Rex Kwan Do from Napoleon Dynamite

Generative AI is emerging as a game-changing weapon for product managers, promising to revolutionize the creativity and speed of how products are conceived and designed and arming them with the potential to achieve greater productivity and market traction. However, it's not a silver bullet. The inherent risks of using Gen AI in the design of hardware products include ‘hallucinations’ - the implausibility of building suggested concepts, the “black box” challenge of researching and grasping the logic behind specific design suggestions, and the danger of data bias (Gen AI output may be based on an insufficient sample size, an overworked dataset, or a misunderstanding of the human prompts).

While AI may have “the strength of a grizzly and the reflexes of a puma…”, it does not have human wisdom. Perhaps the most fascinating aspect of the Gen AI journey in product management is grasping that the flipside of the inherent limitations we humans have from perspectives borne from our experiences, is that it becomes a strength when working with AI.? We are gifted with a nearly boundless mix of context, affinities, emotions, consciousness, intuition, and experiences. The wisdom gained from all of this can provide a chord of resonance or serve as a warning signal and guide us in filtering AI recommendations. Sometimes, it can be a gut feel that a certain design won’t be well-received, but other times it can be insider knowledge on risks related to manufacturability, costs, non-compliance, safety, competitor strategies, or user experience.

There is hard work to be done by product managers and designers in incorporating Gen AI to produce winning products. That work includes building skills in interacting with AI platforms, managing what and how data is fed to the model, detecting bias and errors, and experimenting with prompts to produce more high value design concepts. Further, we are still early in the Gen AI revolution of product management, and the fungibility of the various tool sets across the product management cycle is onerous and requires a knowledge of what platform to use and when (an example appears in the postscript).?

Product managers should embrace the age of AI that has dawned upon us - there is no turning back - and recognize our unique capacities to transform its power into products that result in greater benefit for our customers and greater business value for our organizations.

===========================================================

Postscript: Example of Optimizing Various Gen AI Platforms in Design

As most product managers experimenting with Gen AI can testify, it can be daunting knowing which platform to use and when. It seems that weekly there are upgrades, new capabilities, and new entrants adding to the complexity of comparison and selection. In my view, there has not yet emerged a dominant paradigm for product managers that offer a fully integrated and versatile experience in the product management cycle. I present here a hypothetical work flow where the product manager or designer selects different platforms to complete various tasks.

Scenario

A company seeks to redesign its flagship smartwatch, prioritizing user comfort, functionality, and aesthetic appeal. The product managers are seeking a design that is disruptive, yet retains some trademark characteristics of the brand. Their Gen AI-infused design process might look something like this:

Stage 1 - Data Preparation and Analysis

The product manager has gathered extensive customer research data through surveys, social media, customer support chats, and transcriptions of sales calls. To glean actionable insights from this treasure trove of data, our product team turns to MonkeyLearn. This AI-powered platform excels at analyzing text-based data, effortlessly identifying recurring themes, sentiments, and specific feature requests. It categorizes feedback, spotlights pain points, and unveils emerging trends, providing a clear understanding of customer desires and frustrations.

Simultaneously, the team leverages tools like Amplitude or Pendo to delve into product usage data. By understanding how customers interact with existing features, identifying favorites and pain points, the team gains invaluable insights into areas ripe for improvement.

Stage 2 - Concept Generation and Refinement

Armed with data-driven insights, the design team now enters the exciting phase of concept generation. They enlist the help of ChatGPT (latest version), a powerful language model that thrives on creative prompts. By feeding ChatGPT the summarized insights and design constraints, the team receives a diverse range of textual descriptions of potential smartwatch concepts.

These textual descriptions are then brought to life through the visual prowess of Midjourney or DALL-E 3. These image generation platforms transform words into stunning visuals, providing a springboard for design exploration. The team iterates on prompts, generating multiple concept images to explore various styles and aesthetics.

To further refine these concepts, Adobe Firefly comes into play. This versatile tool allows the team to tweak colors, materials, and textures, creating photorealistic renders that accurately represent the envisioned final product.

Stage 3 - Manufacturability and Compliance Assessment

With promising concept images in hand, the team transitions to ensuring the designs are not just visually appealing but also practical. They utilize CAD/CAE software (like Fusion 360, nTopology, or Siemens NX) to build 3D models from the AI-generated images. These tools analyze the design for manufacturability, suggesting modifications for compatibility with chosen materials and processes. They can also check for compliance with industry standards by incorporating APIs that integrate CAE software with databases of organizations having standards for wearable technologies, such as ISO or UL.

Stage 4 - User Feedback and Iteration

Before finalizing the design, the team turns to Segment, a platform that helps divide the target audience into distinct segments based on demographics, preferences, and behaviors. Concept images are shared with these segments, and feedback is gathered.

This feedback loop is further enhanced by Gemini Advanced or ChatGPT. These AI tools analyze user responses, providing valuable insights and suggestions for design refinements, ensuring the final product aligns with customer desires.

Conclusion

This hypothetical workflow demonstrates that by leveraging the strengths of different AI platforms, product managers can revolutionize the product design process. This synthesized approach might not only accelerate innovation but also mitigate certain biases from a monolithic model and build confidence that the final product is well-informed by user research, manufacturable, and compliant with standards. The weaknesses in this method are the time and complexity of managing each platform within the process and, of course, the cost of having multiple platforms. Throughout the entire process, the product manager and designer play a crucial role in interpreting AI-generated insights, evaluating design options, making strategic decisions, and ensuring that the final product meets both customer needs and business goals.

References

1. ?“IBM AI Global Adoption Index 2022.” IBM Corporation, May 2022. https://www.ibm.com/downloads/cas/GVAGA3JP
2. Rahul Saluja. “Leveraging Generative AI For Breakthroughs In Medical Device Design.” Forbes, 5.2.24. https://www.forbes.com/sites/forbesbusinessdevelopmentcouncil/2024/05/02/leveraging-generative-ai-for-breakthroughs-in-medical-device-design/?sh=4c19825373ee
3. Priyanka Rawal. “The Promise and Potential Pitfalls of AI in Medical Device Design.” MedTech Intelligence, 8.1.23. https://medtechintelligence.com/column/the-promise-and-potential-pitfalls-of-ai-in-medical-device-design/
4. Sebastian Blanco. “AI, digital twins reshaping advertising and car creation at JLR .” SAE Mobilus, 3.29.24. https://www.sae.org/news/2024/03/jlr-generative-ai-ads
5. Roland Mayr, Raju Ivaturi, Harminder Anand. “Generative AI in the Driver’s Seat.” Accenture, 2024. https://www.accenture.com/us-en/case-studies/automotive/bmw-puts-generative-ai-in-the-drivers-seat
6. “Carbon Lattice Innovation: The Adidas Story.” Carbon. https://www.carbon3d.com/resources/whitepaper/the-adidas-story
7. “Reimagining the Future of Air Travel.” Autodesk. https://www.autodesk.com/customer-stories/airbus
8. Bryce Booth, Jack Donohew, Chris Wleizein, Winnie Wu. “Generative AI fuels creative physical product design but is no magic wand.” McKinsey, 3.5.24. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/generative-ai-fuels-creative-physical-product-design-but-is-no-magic-wand
9. Ibid.
10. Sebastian Blanco. “Luminary Cloud takes CAE to the cloud.” SAE Mobilus, 4.19.24. https://www.sae.org/news/2024/04/luminary-cloud