Examples of AI and Operations Excellence from two Podcasts with Motorola Solutions
I just listened to two podcasts from leaders at Motorola Solutions.
The first, from an investor perspective, is from January 2023 on Tech Disruptors (Apple, Spotify).
The second, from an AI perspective, is from Oct 2022 (before the release of ChatGPT) on Gradient Dissent (Apple, Spotify).
Both noted that the general public thinks Motorola makes cell phones. However, they sold the cell phone business to Google in 2012.
Instead of cell phones, Motorola has a great and fascinating ~$10B growing business focused on public safety and enterprise security. Check them out to learn how they’ve combined hardware (like Apple), software (with ERP, Cloud, and Salesforce-like capabilities), and AI.
The two podcasts had plenty of good examples I’ll use in my class:
One, removing time is important for operations excellence.
In the case of public safety, they cite that removing one minute from emergency response time would save 10,000 lives. They realize that coming up with many small improvements will lead to a significant reduction in total time.
Many case studies reinforce this idea. Operational excellence is often about focusing on all the little improvements that add up.
Two, they provide an example of the AI Factory concept from the book Competing in the Age of AI.
In the podcast, they discuss many different uses of AI. This is similar to the idea of the AI Factory: companies should build a platform to support many different applications.
I think others could get ideas from how they broke down their use of AI:
Automating the mundane. This includes using AI to watch security videos and have the people respond to alerts, having AI help fill in reports, and using AI for an assisted narrative.
Accelerate Analysis. In an emergency, understanding the details is challenging. AI can help sort through the many different unstructured data sources: multiple video feeds, interviews, calls, dispatcher requests, and observations from the responders.
Enhance Privacy. I always hear about privacy as a negative in terms of AI. In this case, they are using AI to protect privacy. Different parties want access to the information from an event. A time-consuming part of this is redacting text, audio, or video to protect the privacy of individuals who should not be shown. AI can help get information faster to those who need it while safeguarding privacy.
Three, product management skills are important for AI leaders (and vice versa).
The class I’m teaching now with Yuri Balasanov is to prepare engineering and product managers about AI. We think that it is important for them to understand this space.
The second podcast discusses how AI managers also need to be good product designers. A statement from the podcast that summed up the need quite well: If it takes 30 minutes to explain what your AI system is doing, you’ve probably already lost them.
Four, don’t forget change management.
The first podcast points out that Motorola systems are used by experts—911 call takers, dispatchers, and first responders. These experts all have tools they use to do their jobs. If you build new tools for experts, you need to work with them throughout the entire process.
Five, deep learning systems are challenging.
Several years ago, when it was clear that more firms would use deep learning for things like video, a16z came out with an article to educate investors and IT executives on why the unit economics of deep learning was different.
The article argues that traditional software has excellent unit economics—you write it once, maintain it, and it can keep scaling. However, deep learning has significant costs in training and continually feeding these models.
The second podcast doesn’t discuss unit economics. Still, as you listen to it, you get a sense of the training, the expense of data (cleaning, labeling, and using synthetic data), and the need for ongoing training and adding features.
Six, don’t be afraid of old-school rules and cleaning cobwebs
Real applications require a lot of algorithms. The second podcast talked a lot about analyzing video feeds. But they also used rules to help send alerts. He mentioned that they are working to replace those rules with more sophisticated algorithms.
However, this was a good lesson. To get systems into production, you may have very sophisticated algorithms working with simple algorithms.
I laughed when they talked about troubleshooting AI systems. Sometimes, sophisticated tools and retraining fix problems. Other times, you just need to remove a big cobweb from one of your video cameras!
Seven, pre-ChatGPT talk sounded like people were ready for it.
The second, the technical podcast, has a segment about the need for natural language interfaces. Then, I realized it was recorded before ChatGPT's release. It made me realize that sophisticated teams were already thinking about and getting close to the capability that ChatGPT introduced to the world.