AI applied: How SAP and MapR are adding AI to their platforms

Sometimes when we write about analytics, machine learning and AI, it's challenging to come up with concrete use cases. That makes it harder than it should be for readers to grasp the power of these technologies. And that's a shame, because it makes AI seem ethereal rather than useful or easily understood.

But every so often I am reminded that when one needs use cases, one need look no further than ERP (Enterprise Resource Planning) software. Sometimes ERP is disparaged as mundane. In reality, ERP is what makes businesses run, and when cool technologies are applied to ERP, their impact can be huge, and their value becomes crystal clear.

SAP and AI
SAP S/4HANA Cloud 1802 is the latest quarterly release of the canonical ERP suite, and AI figures into it prominently. SAP's Chief Product Officer, Christian Pedersen, graciously made time and broke down for me how AI has become woven into the software's very fabric.

Also read: SAP's S4/HANA master plan: The lingering questions
Also read: SAP HANA does Big Data...with ERP, CRM and BI savvy

With SAP Leonardo Machine Learning available to the software, it now features really cool capabilities. These include things like determining a deal's probability to close; predictive profit and loss, based on impact of pending orders; and a system that automates matching invoices to orders, which can observe, and learn from, how users do this manually.

Also read: SAP unveils its Data Hub

Handling the AI situation
SAP S/4HANA Cloud includes digital assistants that feature a voice command interface and -- through a partner solution -- can integrate this capability with Amazon Alexa. This release also offers automated payment processing and a new "situation handling" tool. This latter alerts users to risks in purchase order confirmations and purchase requisitions, and automates customer communications proactively.

Pedersen told me that SAP is now looking through almost every business process its software handles and determining where AI should be added. SAP's asset management functionality, for example, is getting predictive maintenance capabilities added in. In our conversation, Pedersen also made the point that AI is worthless without crucial data. SAP has that data, and combined with the HANA and Leonardo platforms, makes AI useful in an everyday manner.

MapR way to ML
If AI is worthless without critical data, and if data has gravity, then it makes sense to bring AI to data platforms. That works much better than extracting data off those platforms, moving it to some data scientist's workstation, and analyzing it there. MapR takes that point very seriously, and in a quarterly update of its own, has made good in delivering on it.

MapR's senior product manager, Ankur Desai, talked to me about the company's Extension Pack 4.1, released this week. Desai explained me that Extension Pack 4.1 adds improvements to Apache Drill and new integration between Apache Spark and the MapR-DB database, when used in JSON document store mode. Java and Python code running on Spark now have direct access to MapR's OJAI (Open JSON Application Interface), which only Scala code had previously.

Also read: MapR 6.0 converges control of data at rest and in motion on the same pane of glass

With this Extension Pack, MapR's Data Science Refinery has advanced to version 1.1 and adds support for PySpark code (i.e. Python code running on Spark) to run across the cluster. MapR achieves this through a container image that contains both the Zeppelin notebook system and the MapR client. The container image is pushed to the nodes in the cluster, allowing data science-oriented Python code to run in a distributed fashion, on Spark.

AI makes house calls
While MapR's application of AI is different from SAP's, they have some things in common. In both cases, AI functionality is being brought to the platforms that contain the critical data. Having AI meet data where it's located increases the chance of AI being applied, since data movement, especially at high volume, is cumbersome and time-consuming.

In the case of a Big Data technology like Spark, bringing AI to the data also alleviates the need to build machine learning models based on mere sampling of the data. If the AI is co-located on the data platform, then building more accurate models using all the data can become routine.

All in all, these two quarterly updates, both released this week, show how AI is burrowing into to all kinds of software, including applications as well as data analytics platforms. The more such practical AI applications -- and the less noise -- we have, the sooner AI will become useful and effective.