Machine LearningAll three of the new iPhones have A11 bionic chip. The word "bionic" is a hint to some of the capabilities of the processor. Of course, it's faster and has more processing cores than previous iPhones, but the real magic in there is the fact that it has hardware support for machine learning. So it has special circuitry to optimize the performance of machine learning algorithms. Without custom machine learning circuitry, apps had to either call services or use the built-in GPU. Both of these approaches have compromises in either latency or battery life.
Right off the bat, auto-correct suggestions will utilize that, but it will also be utilized for some of the new photography functions like custom shading and FaceID. There is a lot that you can do as an enterprise to leverage machine learning on the phone in real-time. Now it won't take network calls and a half-second latency to build machine learning into your app. Using CoreML and the new machine learning hardware, apps can use trained machines in real-time to impact the behavior of the app, all without querying server-side resources. Image recognition is a big winner with this capability.
Sentiment analysis is another potential application, say someone's typing in a message you will be able to tell if they are they happy or angry. And the best time to look for these insights is as they are typing it in real-time right on the phone. This also goes for natural language processing, right now they have speech to text but text to intent can be done with machine learning. It's not limited to deciding if someone is angry, it can be as simple as someone saying, "What is my account balance?" Now there will be no need to do a server call, you can pull up the balance before they're hardly finished with their thought.
My biggest piece of advice in the short term for companies is to collect data. Data is the real gold in this machine learning revolution. The algorithms are common. They are known. They are shared, most are even open source. The secret sauce is the data you collect. The algorithms in isolation aren't that useful. So you really need to make sure you're collecting as much data as you can.