A Playbook for Startups in the Physical/Digital World ⚙️⚡️

The new decade will present a wave of innovation at the intersection of the physical and digital world.

In our last blog post, we broadly examined how AI can transform manufacturing. In this post, we take a deeper look at each of the stages of building a physical product and how these can be transformed using automation and AI.

We believe in a future where users can order highly customizable products — where ultra efficient supply chains produce sub-components on demand, choose the best suppliers to build them with, and where factories can autonomously assemble these components and deliver products quicker than ever. This would allow humans to focus less on the manual labor and more on the true value-creating aspects of the production process: creatively improving factory processes, addressing customer needs, and creating more meaningful product visions and designs. This is widely defined as Industry 4.0, and is hopefully a win-win for both customers and manufacturing employees.

How do we get there? We need to understand the different stages in the lifecycle of a physical product, and systematically improve each of them:

• Design
• Procurement
• Manufacturing
• Testing
• Usage/Replacement

Design:

Today, computer-aided-design (CAD) has become a norm in industries with physical designs such as automotive, semiconductors, and architecture.

However, today’s CAD software solutions are merely visualization tools — they help visual 3D objects on a screen. This means design is quite removed from the remaining steps of the manufacturing process — physical prototyping, sourcing suppliers, physical assembly and product reliability — these steps only come after the design process, creating long iterative cycles and postponing product releases.

We see a huge opportunity for design tools to be more powerful:

1. Augmented Visualization in AR/VR:

Being able to see and feel a physical product provides feedback on its effectiveness during its use case. However, making the first prototype of a new custom design product is expensive and time consuming! Visualizing a design in AR/VR facilitates the prototyping step and provides quick, inexpensive real-time feedback. Startups capitalizing on this trend include Mindesk & Spatial (real-time VR/AR collaboration), eyeCAD (real-time rendering software for architecture) and MakeVR (3D modeling software for headsets).

2. Material Selection:

Citrine Informatics is an example of a company that is building a data platform for the physical world, collecting technical data on materials, chemicals and devices. If material science insights began to pervade CAD tools further than they do today, engineers would be able to achieve a new level of granularity in the physical specifications and reliability expectations for their product.

3. Part Availability:

Given design specifications, CAD software could provide suggestions for relatively standardized parts (e.g. screws, resistors, electronics) and their real-time availability and prices across suppliers. This integrates procurement insights earlier in the design process.

4. Quick, On-demand Prototyping:

After the first prototype, physical products go through several iterations to get built correctly. How might we minimize the number of iterations? Perhaps through quick on-demand prototyping, coupled with AI-powered design-for-manufacturing insights.

• Plethora, Tempo Automation, Fictiv and Shape Prototype are some startups catering to this prototyping need, providing early DFM (Design for Manufacturability) feedback and real-time quotes.
• Xometry is providing instant part quotes and quick high quality prototypes through a network of machine shops.
• Shaper Tools is a handheld CNC tool that provides manufacturability insights without having to buy or program a full CNC.

5. Rapid Design Evaluation:

Design evaluation has been limited to complex analytical simulations like FEA (Finite Element Analysis) and CFD (Computational Fluid Dynamics), which often can not be implemented rapidly due to their computational complexity, and are only done at major milestones. Moreover, these evaluations can call out issues but aren’t useful at suggesting design improvements. With ML-based systems and evolutionary design, it becomes more and more likely for CAD tools to be prescriptive and suggest design changes (similar to the way that code-autocomplete is evolving for software engineers, an example of which is Kite).

Procurement:

Procurement involves finding suppliers that can build a part as cheaply and quickly as possible while meeting the quality and quantity requirements. Procurement is a field that can be disrupted by automation. Through automated bidding, the best suppliers can be selected based on cost, delivery time and quality. There are two types of B2B SaaS procurement solutions available today: B2B marketplaces and invoicing/spend management tools.

However, there are drawbacks to today’s procurement solutions:

• Catering to the Fortune 1000: Large software providers like SAP and Oracle might offer both marketplaces and spend management as part of their package, but they cater largely to the Fortune 1000. We believe there is a huge opportunity for startups to disrupt the end-to-end procurement SaaS space for companies smaller than the Fortune 1000.
• Direct vs. Indirect Procurement: Direct procurement deals with items on the BOM (Bill of Materials) that are assembled into the products sold by the manufacturer. Whereas indirect procurement handles items for a company’s internal consumption rather than for a customer or client (e.g. coffee, furniture, toilet paper). Indirect procurement solutions (e.g. BuyerQuest, Vroozi) are easy to scale (since the operating needs of most companies are similar). However, indirect procurement only makes up 20% of a manufacturer’s total procurement and is also of lesser strategic importance to a manufacturer. Therefore, most manufacturers tend to have in-house teams for direct procurement but are relatively comfortable outsourcing indirect procurement.

We believe there is an opportunity for startups to tackle direct procurement and, through AI, provide better deals for manufacturing clients. For a direct procurement solutions provider in a particular industry, having access to industry-wide data can help them to create a more efficient market than their customers’ in-house team might be able to. This can only become more effective with more data, using AI.

Manufacturing:

In the US, as the current generation of manufacturing experts is increasingly retiring, and manufacturing is beginning to be seen as “un-sexy” by millennials, factories are faced with a skills gap — it presents a large opportunity for startups to disrupt the shop floor.

We see three major categories for disruption:

1. Augmenting Machines:

We see an opportunity for expensive machines and equipment on the factory floor to be augmented with sensors to provide predictive maintenance and servicing. This prevents unexpected factory “line-down” situations due to malfunctioning equipment, which is something that directly impacts the bottom line of a manufacturing business.

• Oden Technologies is an example of a startup looking to digitize old factory machinery using retrofittable sensors that sense a machine’s degrading performance over time.Neewee, Maana and Augury are other companies in the equipment maintenance space.
• Bright Machines employs “micro-factories” made up of robot cells to automate electronics manufacturing and inspection.
• Element and UpTake recently partnered to collect and analyze factory floor data to prevent failures and defects.

2. Upskilling Operators:

There’s been a huge, much-needed push in the last five years to make manufacturing conditions easier for factory floor operators in terms of fairer wages and better hours. However, jobs continue to be heavily physical and manual. We see an opportunity for hardware tools to upskill operators, making their physical jobs easier by enabling them to lift/move heavy objects or even provide operator instructions.

There are two buckets of startups we see in this upskilling space:

• Visualization: Upskill.io and Light Guide Systems are using AR to provide operators with assembly instructions on the factory floor. Daqri, Atheer, Realwear and Scope AR are supporting visualization for industrial settings.
• Cobots: There is an increasing prevalence of “cobots” — collaborative robots — that augment human capability. Some examples are Esko Bionics and Sarcos that are essentially subsets of IronMan suits, allowing operators to do physical jobs which they weren’t able to earlier, and provide a new paradigm of safety on the factory floor. Other examples are SmartCap and StrongArm Technologies — a smart helmet that monitors worker fatigue.

3. Shop Floor Management:

• Today’s factory has disparate machines on disparate production lines. But what if we could one day connect all these disparate entities to isolate failures on the production floor quicker? Or to find inefficiencies and address them? Tulip.co is an example of a startup that connects manufacturing IT to the smart shop floor. MachineMetrics and SightMachine are factory floor management softwares. Foghorn and GEM are providing edge intelligence software that would provide lower latency for above solutions. And Xage Security is working on cyber security for industrial IoT applications.
• Warehouse management is another booming area — our thoughts on this topic can be found here.

Testing:

As the physical products we produce have become more complex over the ages, the need for quality control and testing has increased before products can be shipped.

1. Visual Inspection:

AI presents an opportunity to augment the highly manual task of visual inspection. For example, Landing.ai started out enabling automated optical inspection (AOI) to tag defects in printed circuit boards (PCBs). Instrumental takes pictures of objects in electronics and compares images to detect potential failures. Inspection can also be applied to machines and equipment — Re-flekt is an example of a startup using mixed reality to assess equipment issues, maintenance and tuning.

2. Insights from Test Data:

Currently, even in the high-tech industry, test data collected during the production process is only used to determine whether a product passes or fails. However, we see an opportunity to run ML models on the massive datasets collected through testing with the insights fueling better designs of the same product in the future.

3. Design/Assembly Corrections:

In-process testing also brings the opportunity for real-time correction of issues. Every physical product is made up of many sub-parts and each sub-part is imperfect to some degree, as defined in the part’s manufacturing tolerance. Even if that sub-part were somehow perfect in every dimension, it might be assembled imperfectly in the finished product. We believe AI can correlate the impact of these imperfections to the performance of the finished good. This can help engineers decouple design and assembly issues during the process of building a physical product.

Usage/Replacement:

Not all products are made equal. But most frequently, we see that the warranty of a product is the same, irrespective of which product we buy! Besides predictive maintenance of a product using sensors (e.g. the “battery health” feature on an iPhone), we see the following opportunities in product usage/replacement:

1. Holding Inventory vs Sourcing Replacement Parts:

One question that faces high-volume consumer product companies is “how much inventory of our older models should we keep?” With 3D printing becoming more prevalent, we see an opportunity for replacement parts to be produced on-demand, minimizing factory inventory, and directly impacting a company’s bottom line.

Until 3D printing becomes ubiquitous, we see an opportunity for replacement part marketplaces to flourish. Currently, finding replacement parts is a tedious process — the market is fragmented and there is no centralized database for each industry. There are certain industries where the replacement part market can be quite lucrative. For e.g., the aircraft replacement part market is an $80B market — not necessarily due to the complexity of the parts involved, but due to sheer urgency. When an aircraft is on the ground due to a defective part, it leads to millions of dollars of lost revenue for the airline. SkySelect is an example of a startup looking to build a B2B marketplace and payment solution for aircraft replacement parts.

2. New Business Models:

Traditionally, factories have purchased machines for their production lines and are responsible for their maintenance. However, a pay-by-use model for machines would be cheaper for factories, and provide machine manufacturers more performance data on their machines. This would also allow machine manufacturers to provide preventive rather than reactive machine maintenance since they can quantify machine risk better.

In conclusion, a fully customizable future of manufacturing may be far away — but in the interim, there is a tremendous opportunity for startups to ride the physical/digital wave and propel the digitization of manufacturing.

Thanks for reading! This was published based on my work last year at Baidu Ventures.

I’d love to hear from you! You can reach out to me on LinkedIn. ✌️