SAS’ Florida Vacation — 2019 Edition
Every year, in February, SAS hosts a beauty show. It’s a ritual. Industry analysts fly to a posh resort. There, they see shiny new things most users will never see. They listen to executives talk about how the company finally has its shit together.
This year is no exception. SAS held its analyst conference February 11-12 at Vanderbilt Beach, Florida. On the agenda this year: SAS is saving wildlife, curing cancer, and solving the opioid crisis.
Corporate virtue-signaling is tiresome. SAS was perfectly happy to sell software to tobacco companies and to opioid marketers. Some might say that if SAS gets credit for solving problems, it should also get the blame for creating them.
But that would be mean-spirited. I would never say such a thing.
Forrester’s Mike Gualtieri attended. He tweeted. Once.
Must have been an exciting show.
Jim Goodnight kicked off, noting SAS’ 43 years of growth. He’d be more accurate if he touted 37 years of growth, 5 years of no growth, and one year of undisclosed growth. SAS has not published its 2018 revenue, something it does every January except this year.
SAS is a privately held company, and it’s not required to disclose revenue at all. But when an organization reports revenue for 42 consecutive years and then suddenly goes quiet, you figure the latest numbers aren’t pretty. It’s like when your kid slinks in the back door and buries her report card in yesterday’s mail, you know it’s a stinker.
Analysts Lisa Dodson, Alex Kwiatkowski, and Jared Peterson tweeted glowing reports:
Oh, wait. They aren’t analysts, they work for SAS.
Gartner’s Cindi Howson liked the cookies:
Those are enormous cookies. You could build a house with those cookies.
Cindi also approves SAS talk about more flexible pricing. But SAS customers do not want more flexible pricing. They want lower TCO.
Every year, after the conference, analysts return home and write about the shiny new things SAS showed them. Tony Baer does not disappoint. Here are a few snippets:
When BI emerged back in the 1990s, SAS was already a mature player that catered to a narrow elite of data miners and quants. You may have used Business Objects or MicroStrategy to generate sales by region dashboards, but when it came to crunching deep analytic models, SAS had few rivals.
Stop right there. I used SAS in the 1990s at AT&T, Citibank, and Chase. Analysts used SAS back then because it ran on a mainframe, where a lot of companies kept their data. Few people liked using SAS. There were plenty of other options for “deep” analysis. We’d write SAS jobs to yank data from DB2/zOS and export it to the analysis software we wanted to use.
The most popular SAS procedures? FREQ, MEANS, PRINT, RANK, SORT, SUMMARY, and TABULATE. Now and then, we might run a linear model or logistic regression. One junior analyst at AT&T was completely in love with PROC STEPWISE. He kept banging away with it until his manager told him to stop because it produced idiotic models. Interaction effects but no main effects, stupid shit like that.
Deep analytic models my ass. 80% of SAS workload is ad hoc analysis. You can’t use Business Objects or MicroStrategy for that because they aren’t designed for self-service and it takes IT three months to do anything.
Executives like their ad hoc analysis while they’re young.
For SAS, the overriding theme in recent years has been broadening its footprint in a landscape where AI (mostly machine learning) is reshaping BI and driving the need to treat modeling, not as a one-off but as a lifecycle that integrates with operational systems both on-premises and in the public cloud.
I’ve seen freight trains in Nebraska shorter than that sentence.
Both developments are making the landscape where SAS competes increasingly crowded, from the self-service visualization tools of business analysts, and the data science collaboration tools that are coming from an expanding array of venture-backed startups.
Plus, most companies don’t keep their data on a mainframe anymore.
Then there’s open source, the elephant in the room, which SAS no longer treats as a mortal threat but a heterogenous system; SAS’s contribution is applying process and governance that picks up where open source leaves off. When it introduced SAS Viya three years ago, SAS opened a path for data scientists to write models using Python, and later, Java, Python, Lua, and R, then invoke SAS routines in Viya on the back end.
Open source is a mortal threat to SAS and other commercial software vendors who do not adapt.
Yes, you can invoke Viya from Python, Java, etc. Who cares? If you use Python to build your data pipeline, and you’re going to embed your machine learning model in a Python application, which of the following options is your best choice for machine learning?
- SAS Viya
- One of the more than a thousand machine learning packages with Python bindings
If you chose option (1), you work for SAS.
The SAS hybrid open source model is backward. Customers want an open source hub, with commercial tools that provide unique capabilities on the edge. SAS wants you to build a SAS hub, with open source tools on the edge. That’s not going to fly.
Oh, by the way, the open source code runs outside of SAS. SAS doesn’t apply “process and governance” to it. And SAS built the Lua interface to Viya first. Because, you know, there are so many data scientists who use Lua.
But until recently, SAS had a perception issue with Viya, as many prospective customers had the impression that it was merely a Tableau-style pretty face instead of a new analytics platform tackling the full lifecycle.
No. Customers had the impression that it was a complex mess. This is SAS’ simplified Viya architecture:
That’s not a perception issue.
But SAS is finally starting to get the Viya story out. Given that SAS’s growth has hovered around the 1 – 2 % range for the past five years, perceptions are critical for the company to spark renewed growth.
2018 was where Viya finally took off, with sales roughly doubling. Meanwhile, SAS’s cloud business grew just over 30%, primarily from the Results as a Service offering, where you feed your data to SAS-managed models.
Yes, 2018 was a bang-up year for SAS sales. So good that the Chief Sales Officer departed to spend more time with his family. And SAS is playing “hide the pickle” with the revenue numbers.
To those…who want to get updated on what we saw at this year’s SAS analyst event, the focus was not on the product roadmap per se, but the end-to-end journey.
The message is that, beyond the large portfolio of SAS analytics functions are tools that support each step of the analytic lifecycle where the appeal is being broadened beyond SAS’s usual constituency and into enterprise operations when analytics becomes key.
The trip spanned from data management to model management, data discovery, and deployment featuring deep dives into selected use cases. Examples included showing how SAS ESP ingests and transforms real-time streaming data in a data pipeline that also enriches with data at rest to support cancer research. From that pipeline, you can use SAS or third-party analytic tools. There was the phase of data discovery, using an example with credit fraud prevention that provided GUI-driven environments for the business user and model building, competition, and training for the data scientist. Then there was the focus on deployment, providing the metrics on deployment and lineage (of data and the model), and last-mile applications such as SAS Intelligent Decisioning being applied to customer next-best offer scenarios.
Clearly, SAS has plenty of company in different parts of the analytics lifecycle, from modeling and model lifecycle management to analytics, reporting, and self-service. SAS’s big draw has always been the breadth and depth of its analytics functions. As one customer put it, open source frameworks or libraries will typically take the 80/20 rule, such in classification or regression analysis, while SAS will venture beyond with the more esoteric stuff or analytic functions that are shaped for specific industry sectors.
This is completely wrong. Developers publish esoteric and innovative stuff as open source software first because there are no barriers to entry. It’s the long tail effect. Commercial software publishers like SAS are more conservative. They typically wait until they see clear market demand before investing in a particular capability. As a result, open source functionality is a superset of commercial software functionality, not a subset. All of the most innovative developments in data science were available in open source long before SAS even thought about introducing them.
Anyone who doesn’t understand this has no business writing about machine learning and AI.
But the functional gulf between SAS and the open source world is a battle for time that has a generational spin to it.
A colleague who attended the last SAS Global Forum says it seemed like an AARP convention.
In 40 years, SAS has had the opportunity to develop a deep and wide portfolio of analytic functions.
PROC FREQ, PROC MEANS, PROC RANK, PROC REPORT, PROC SORT, PROC SUMMARY, PROC TABULATE.
With the enthusiasm for open source, especially among incoming data scientists and data engineers, there is a strong push in the community to expand the portfolio in the open source wild. We’ve also seen how sectors that traditionally kept development to themselves are increasingly embracing open source, either through open source-first strategies (use open source technology where it is viable) or building their own open source projects. A decade ago, tech meetups were practically non-existent in New York. Today, a week rarely goes by without somebody from Wall Street or the media world demonstrating their new open source contributions.
Well yes, that’s the rub. Working data scientists prefer open source software. Dogs don’t like the dog food. And people wonder why SAS revenue is flat.
As noted above, SAS has taken the first steps by meeting the open source community halfway by making SAS analytics on Viya accessible to non-SAS languages. We’ve gone on record previously that SAS should take the next step and start contributing to open source as well. At the end of the day, SAS’s core value-add will not just be its rich analytics functionality, but also how it helps you manage and implement analytics, and how SAS delivers it.
SAS has nothing to offer the open source community.
Consider what SAS has already delivered with its Viya portfolio for repackaging, and its Results as a Service for delivery, Now imagine once SAS containerizes its portfolio in cloud-native format; that could help SAS accelerate the refactoring the analytics portfolio as cloud services.
Did somebody say ‘containers?’ SAS does not run in containers unless you license SAS for Containers, which hardly anyone does.
It’s really quite simple. Suppose a customer licenses SAS to run on 10 4-core servers. That’s 40 cores, so SAS revenue is 40X.
But those servers are only about 15% utilized. The customer figures that if they could virtualize the infrastructure, they could support the same workloads in Docker running on an 8-core server. If SAS lets the customer do that, their revenue declines from 40X to 8X.
Of course, SAS isn’t going to let the customer do that. They will price the software to ensure that virtualization is revenue-neutral.
That is SAS’ problem in a nutshell. SAS can tinker on the margins with “flexible pricing.” But SAS customers want to reduce SAS TCO. There is no formula that delivers cost relief without markedly reducing SAS revenue. It’s just math.