Amazon Web Services’ re:Invent conference was quite the experience and included hundreds of sessions covering a wide range of topics. But the clear frontrunners were analytics and machine learning (ML).
Andy Jessy, CEO of AWS, emphasized AWS’ ability to provide AI/ML tooling, process and infrastructure. A good portion of his announcements through the first two hours were on new offerings that facilitate machine learning. Along with these new offerings, Amazon’s view of ML tooling has evolved to consist of a three-tiered approach.
- The top tier is AI Services – fully functional services that are powered by AI. This consists of services like Amazon Forecast and Amazon Personalize.
- The second tier is ML Services. This is where the AWS SageMaker product shines by bringing more power to the customization of your own ML tooling, along with add-ons like the ML marketplace and other ML specific tooling.
- The bottom tier is for expert ML practitioners. It consists of tight integrations with the hardware AWS runs to the major ML frameworks (Tensorflow, PyTorch, MXNet, etc.)
For Atrium, this allows us to bring quick wins to companies that want to see the benefits of machine learning and really helps us move the needle when a more custom approach is necessary.
Here are some of the AWS tools and services that we learned more about at re:Invent.
This is the service I continue to be most excited about. Announced during Andy Jessy’s keynote on Tuesday, SageMaker Studio is meant to be an Integrated Development Environment for the entire model life cycle – build, deploy, run, monitor – and comes with the following components:
- Experiments – run a number of different models to see which fits best
- Debugger – debugging for model training
- ModelMonitor – checking in on model viability over time
- AutoPilot – AutoML-like capability for quick model fitting
Amazon has spent 20 years working on product recommendations and they are now working to sell this capability to their AWS clients with Amazon Personalize. This offering breaks down Related Products into two categories:
- People who bought this item also bought – based on historical customer behavior
- Similar Items – items that have similar descriptions or characteristics
Some consumer product companies are afraid to use Amazon services, for fear of giving up information about their business or products. This tool might be a cheap and easy way for them to get product recommendation functionality on their site without having to build it themselves. One has to wonder whether the perceived ease and power of this product might make some of those CPG companies reconsider their position.
Amazon Connect Integration with SFDC Service Cloud
As part of its Service Cloud call center solution, Salesforce will be offering AWS telephony and call transcription services through Amazon Connect. I am excited to see what the possibilities might be for expanded transcription of service calls and chatbot sessions – such as transcripts being examined using sentiment analysis and combined with structured data on account performance to give some great models on customer churn and success.
Overall, re:Invent was a week well spent. At this point, I’ve recovered from the long week in Vegas and am looking forward to working with SageMaker Studio as a new technology platform for helping our customers solve challenging business problems. We expect to see immediate value using AutoPilot and Experiments on existing datasets to find better model fits with complex ML algorithms.
One of the biggest challenges with data science is getting a predictive model from ideation to production with high levels of adoption. We believe that SageMaker will be a key tool to enable us to shorten development cycles and significantly exceed industry standard success rates. There continues to be a lot of optimism that the market for ML is maturing. The types of tools and services AWS is bringing to the table will help make the benefits of ML a reality for more companies.