Top AI/ML News for 2020
In June, SAS announced that its revenue declined last year. Oh, wait, SAS didn’t announce that, they quietly slipped it into the back of their annual report. SAS revenue declined from $3.3 billion in 2018 to $3.1 billion in 2019. SAS blamed the decline on three things: (1) new pricing strategy; (2) adoption of the ASC 606 accounting standard; and currency movements. Blaming the decline on an accounting standard is preposterous; ASC 606 went into effect at the beginning of 2018, so it should have no effect on 2019 revenue. In any case, ASC 606 generally favors companies like SAS that sell subscriptions.
SAS Chief Revenue Officer David Macdonald left for greener pastures in October. Pro tip: CROs rarely leave when revenue looks good. Dave Mac is a great guy and a sales pro — if he can’t change the revenue story at SAS, nobody can.
This year IBM rebranded and repackaged its stuff again; IBM is very good at rebranding and repackaging. The product previously known as IBM Data Science Experience morphed into IBM Watson Studio a couple of years ago when IBM bundled it with IBM Data Refinery. Watson Studio used to have machine learning capabilities and a rudimentary AutoML tool, but IBM repackaged those into IBM Watson Machine Learning, bundled with — wait for it — SPSS Modeler.
Wait! There’s more. IBM also rebranded Red Hat OpenShift as IBM Cloud Pak and added some software, including the products listed above. They market this as IBM Cloud Pak for Data, which is available on-premises or in private clouds, or in what used to be called IBM Cloud but is now called IBM Cloud Pak As A Service.
IBM also continues to bundle its AI/ML software into enterprise agreements and call it “free.” IBM client executives are very good at selling AI/ML software as long as they don’t have to ask customers for money.
Mathworks kept on doing what IBM can’t do: make and sell software that users want to buy.
At its annual re:Invent show, Amazon Web Services introduced a half-dozen new SageMaker services. The most notable of these is a no-code data prep tool, so SageMaker users can prepare data without writing code. The rest of SageMaker still appeals primarily to developers.
Microsoft introduced the Designer drag-and-drop user interface for Azure Machine Learning. It’s a nice front end, but the features available to Designer users are not the same as the features available to SDK users. This is a source of endless confusion for users.
Google Cloud Platform used to have two siloed products for machine learning, Google AI Platform and Google Cloud AutoML. Google now markets these as one product, but they still don’t work together very well. Google Cloud AutoML Tables remains in beta after eighteen months.
After Alteryx acquired the Feature Labs team, some people had high expectations for Alteryx Assisted Modeling. More sad faces. After a year in beta, Alteryx released the product to GA, with a price tag — $2,300 per seat, over and above the $5,195 one pays for Designer. The UI seems nice, but features are scarce, and the models you build can’t go anywhere.
In other Alteryx news, founder Dean Stoeker departed into the sunset. He didn’t take the camper van, but I guess he can afford to buy his own.
Databricks chugged along, minding its business. The company donated MLFlow to the Linux Foundation, but the top contributors are still mostly Databricks people.
In August, Dataiku raised $100 million in a Series D round. It’s a nice raise, but possibly some sad faces in Paris — sources say it was a down round. The company recently soft-launched a managed service for SMB, and an MLOps product is in the works. Keep that in mind when your Dataiku rep tries to tell you that Dataiku already does MLOps.
Last spring, competitors laughed and whispered that “DataRobot is in trouble.” They’re not laughing now. In November, DataRobot announced a $270 million Series F led by Altimeter, on a $2.7 billion valuation. For icing on the cake, in December the company announced additional investments from Salesforce, Snowflake, and HPE.
H2O.ai recently shoved its long-touted MLOps product out the door. Sadly, they forgot to finish the documentation. It’s not a big deal, but it does make you wonder what else they forgot.
Domino Data Lab snagged $43 million in a Series E round. The company also launched the Domino Model Monitor, which does what it sounds like. Domino also hyped its partnership with SAS, featuring SAS for Containers, a product that nobody uses.