Using AI for managing internal operations

- Three major elements of any company's internal operations are, designing and building the factory or service center, day-to-day operations, and quality control. Using concrete examples from BMW, DoorDash, and landing.ai, let's look at how these companies are deploying AI across the three domains. BMW is partnering with NVIDIA to use AI for designing a new drivetrain factory. The partners have deployed NVIDIA's Omniverse software to create a digital twin of the entire factory, enabling managers to see in photorealistic detail how the factory will look and operate. The virtual factory shows all machine tools and workers exactly where they will be stationed. As virtual parts move through the virtual factory, the software even takes into account the effects of gravity, material type, and workers' tools. While factory builders have used simulations for many years, today's digital twins, like BMW's, are radically different in terms of the number of factors they incorporate. This richness of inputs enables engineers to simulate the actual working of the factory at a level much closer to reality. And many of today's AI algorithms simply did not exist even 15 years back. As these algorithms run through millions of cycles to gather training data, they help the engineers understand what's suboptimal and how they might tweak the overall factory design. DoorDash showcases the use of AI in day-to-day operations. DoorDash is a logistics company that enables customers to order and get goods delivered from local merchants. To use DoorDash to order from, say, a restaurant, you log in and choose the restaurant. Once you pick what you want, DoorDash shows you the bill and the estimated delivery time. You check out and wait. How does DoorDash predict the delivery time? The company uses a supervised learning algorithm to train its ML model. Historical training data include the actual delivery time, as well as input details pertaining to each order, the merchant, a timestamp, as well as details pertaining to the neighborhood, such as number of outstanding orders, number of available couriers, day of the week, holiday or not, et cetera. As new data emerges, the model gets retrained regularly. The goal is to be accurate to the last minute. Overshoot, and customers may not order. Undershoot, and they may get angry. Behind the scenes, AI also plays a role in DoorDash's logistics engine, balancing supply and demand of couriers, planning optimal routes with changing traffic patterns, and matching couriers to merchants and customers in real time, picking the optimal solution from millions of combinations. Finally, landing.ai illustrates the use of AI for quality control. Founded by Stanford's professor, Andrew Ng, this company helps clients manage their internal operations. One such client is a steel company that must routinely test for defects in steel sheets. Defects can be problematic in terms of aesthetics, functionality, or both. However, this is not true of all defects. So the key challenge is not merely to identify a defect, but to figure out which is problematic and which is not. For AI, this is a supervised learning problem. To train the ML model, landing.ai has worked with human inspectors to label each type of defect as problematic or not, paying particular attention to accuracy and consistency across different inspectors. Once trained, the machine learning model has proved to be more accurate than the human inspectors. For a large steel mill, even 1% greater accuracy pays off very quickly for the investment in AI. Now, analyze all three aspects of your company's internal operations, designing and building a new factory or service center, day-to-day operations, and quality control. How might you apply AI to improve upon what you currently do?