One of the most important investments our predictive team is making this year is rebuilding our user interface (UI) to take advantage of modern platforms such as SAP Fiori. This is really important—our technology has been generating huge amounts of customer ROI for many years, but it takes some time to learn how to use and user expectations have changed.
We have made substantial progress in this journey, and we’re getting great feedback from users. So it caused me some consternation the first time my boss Erik Marcade declared that “the best predictive UI is no UI.” What could he mean?
It turns out that it makes a lot of sense. We are focused on two main drivers: the massive predictive factory and predictive for the masses.
Massive predictive factory: To deliver value from predictive analytics, you need to build a model, put it into production, and then use it to answer business questions on an ongoing basis. In some situations a single business question may generate thousands or millions of models when replicated across product segments and regions. Each of these has to keep up to date and applied on an ongoing basis. This needs to happen in a lights-out operational setup, because there is no way a data scientist could scale to run and monitor them, and it would not be a good use of their time.
Predictive for the masses: In the end, predictive is most useful when it is used to answer operational business questions: Is this opportunity going to close? Will this machine fail? Will this contract be paid on time? Is this transaction fraudulent? Each of these processes have their own users and business applications.
The highest increase in productivity occurs when the output of predictive is embedded within their applications as native features. In many cases the output is presented as a business action, task, or score, and the user is unaware that predictive is even used.
This is the true prescriptive analytics—where the application prescribes an action directly to the user or makes a decision without user input.
Making prescriptive analytics a reality
So what do we need to build to make this a reality?
- We first need a repeatable mechanism to embed predictive in applications as core features with all of the operations they need to be accurate, including apply and retrain. We call this a predictive scenario.
- We then need to enable business users to configure their predictive scenarios to apply them to specific use cases. We call this business context.
- We then need a mechanism to allow a data scientist to update a model to make it more accurate given their understanding of the business. Predictive features need to be able to leverage the knowledge of data scientists and analysts where these skill sets are available.
- Business applications are critical components of modern business architectures and can’t be put at risk. Predictive analytics needs to fit into the overall extensibility framework of the business applications, it needs to work within established security and governance frameworks, and it needs mechanisms to deliver updated models and scores to business users in robust and secure manner. This last mile is critical to the success of predictive analytics in real-world scenarios.
The good news is that we’re building both a beautiful user experience for data scientists and analysts and an architecture and framework to make sure that it functions in situations where “the best predictive UI is no UI.”
For more on the power of predictive, see Predicitive Analytics And The Segment Of One.