Register for "A Guide to Data Products: Everything You Need to Understand, Plan, and Implement" - Friday, May 31,1:00 p.m. ET

SAS Addresses Five Key Analytics Challenges

We Need Innovation in Our Analytics Platforms

Less than one year ago I researched and wrote a report on data science platforms and how they were supporting the complete analytics production line. The platforms have already made a lot of progress since that research came out, including a growing understanding of the importance of data preparation. This has led to the crystallization of the concept of DataOps and its application to machine learning.  

There is still, however, work to be done. Specifically, these platforms need to support more user stories from a greater variety of user types, more analytics modules (especially to support AI as they became available) and cover the complete analytics production line (from raw data to measurable and optimizable decisions).

Current Challenges for Analytics Platforms

Here are five key challenges being addressed by analytics platforms today:

  1. Automation. The analytics production line still takes too long to complete and there are still parts of it that can be automated that are not yet automated. Example: determining the optimal time window for a time series or picking the best machine learning algorithm.

  2. Explainability. Predictions need to make sense and be explainable if they are to be used for business. Example: With the new privacy laws in the European Union, decisions must be explained to consumers, must include a human judgment, and cannot be solely based on an algorithm.

  3. Computer Vision. Computer vision has become one of the ‘killer apps’ of AI and pre-built and reusable modules are needed. Example: human facial recognition is used in many AI applications and a prebuilt module could often be used rather than having to create one from scratch each time it is needed.

  4. Decisions at Scale. Models and predictions must be balanced against known common sense business constraints in order for them to be useful. Example: a model might recommend shipping 100,000 units of a product to optimize profits but only 50,000 units exist in inventory.

  5. Customer Experience. Companies need to shift appropriate parts of the analytics burden from expensive data scientists to business users. Example: “Next best offer” models could be used and updated by business users once data scientists have done some of initial heavy lifting and data preparation.

SAS Expands its Analytics Platform

SAS has done a nice job of addressing these challenge areas in their recent release of SAS Visual Data Mining and Machine Learning 8.4:

  1. Automation – SAS now has automated rule generation, automated sub-segmenting for time series data and automated selection of the best AI or machine learning algorithm.

  2. Explainability – SAS has enhanced its natural language and storytelling capabilities to give human-understandable text-based explanations of reports and predictions. It also utilizes the algorithms for computing the Shapley value, which can report on which factors have the most impact in a decision in more complex decision spaces.

  3. Computer Vision – SAS now has support for the DICOM and DICOM RT standards for medical images. It also has improved accuracy for object detection with YOLOv2, TinyYOLO, and R-CNN systems. They also include support for MobileNet, and ShuffleNet which are able to run effectively on less powerful mobile devices. This is important for performing analytics at the edge.

  4. Decisions at Scale One of SAS’s unique strengths is their inclusion of decision support tools as part of their platform. They have now extended these to include additional real-time capabilities that would, for example, allow a physician to access predictive models while meeting with a patient and deliver an immediate diagnosis. They also provide ‘container recipes’ published on GitHub for AWS and Azure to make it easy to scale on these cloud platforms.

  5. Customer Experience – SAS has simplified its pricing and now prices its platforms in three main ways: 1). by number of users, 2). by number of CPUs, and 3). unlimited for the enterprise. They are also aggressively reaching out to create new SAS users by providing SAS for educators at no cost and SAS for students at 95% discounts off all public courses.

What Does this All Mean? 

SAS is the gorilla in the room when it comes to analytics but they are nonetheless proving to be nimble in responding to existential challenges such as the cloud (AWS and Azure) and open source. They are also, like all analytics companies, trying to keep up with the very good news that AI is spreading from top to bottom in every organization. Because of this, AI needs to be operationalized, made more robust and automated. 

In the future expect SAS to make more of a marketing splash talking about and providing support for AI. As any good data scientist understands, the goals and techniques of analytics are very similar whether you call them AI, machine learning, data science, statistics, or predictive analytics. The distinctions have more to do with the applications of analytics than differences in the fundamental science driving the algorithms. But making it easy and scalable to apply analytics remains the challenge for these platforms.

A Step in the Right Direction

It is good to see large analytics platform providers like SAS, continually innovate to meet the evolving needs of data scientists, AI researchers and now business users. The result is a more fully functioned platform that satisfies more users, with more functionality across the entire analytics production line. 

Related articles: 

Stephen J. Smith

Stephen Smith is a well-respected expert in the fields of data science, predictive analytics and their application in the education, pharmaceutical, healthcare, telecom and finance...

More About Stephen J. Smith

Books by Our Experts