Masterspeak 2017

"I like to hire really strong people and let them do their thing"

Analytics mogul, Jim Goodnight talks about SAS’ culture, machine learning, and much more

Published 7 years ago on Sep 01, 2017 15 minutes Read

Cary, North Carolina/ July 19, 2017

Spread over 900 acres, the SAS headquarters at Cary, North Carolina is probably the biggest single-site office location in the world. And if not a data scientist, its CEO Jim Goodnight could have easily been a geologist, given his fascination for collecting rocks. The conference room where we are interviewing him has lots of it neatly named and arranged, his favorite being the deceptively heavy Gibeon Meteorite from Namibia. The collection built over the years now totals 400 and gets added via the annual shopping trip that Goodnight makes to Tucson every February. With revenue of $3.2 billion, SAS which he co-founded in 1976 is one of the world’s biggest privately-owned companies but its private ownership has not prevented it from becoming a highly-rated employer of choice. There is possibly no enterprise of consequence that does not use SAS’ analytical software. Not only are 90% of the Fortune Global 500 companies its clients, the IRS also uses its software to identify fraudulent tax filings. To ensure that its offerings stay cutting edge SAS spends 25% of its revenue on R&D each year and now has its sights on machine learning.

How did you get started on analytics? Can you take us through the work that you used to do, and how is it different from what SAS does now?

We started SAS back in 1966 at North Carolina (NC) State University. I was actually working part time, for the Department of Statistics, as I put myself through college. We were in charge of analysing all the agricultural-experiments conducted on campus. NC State is a land-grant college that was established to study agriculture and engineering. There are a lot of agriculture related departments over there. I was doing experiments all the time, to try to improve the crops, try to make cows give more milk or grow faster.

The Department of Statistics would help design the experiments. I was part of the group that analysed the data. We constantly had to rewrite the programs to analyse different kinds of data. What we needed was a method whereby we didn’t have to rewrite everything, every time; just change a few parameters. That’s how SAS got started.

Back in those days, we were predominantly focused on planned experiments. They were laid out in a way that made it possible to analyse all the experimental designs. It still goes on today, it is big in the universities, in the pharmaceutical companies where they are testing new drugs. And we had to test for the efficacy of the drug, if it’s better than the existing ones or if it had any side-effects.

But those were planned experiments. Sometime around 1990, banks started developing models from the data that they had collected. Today, we do things like credit card fraud modeling. We look for certain patterns that are unusual for the owner of that card, and use some very sophisticated neural network models to determine whether the card is being used fraudulently. Nowadays, we see more and more building of models to try to predict customer behaviour, try to score things like: what’s this person’s credit score, should we loan this person money; should we send this person a catalogue, what ad or website should we pop in front of him. We see much more use of analytics for that purpose than we do for planned experiments.

Can you elaborate on the kind of services you provide to your clients across different sectors?

Credit card fraud detection is a very specific solution we provide. But we also have a number of other banking solutions, one of those is IFRS9, which is our reserve requirement based on the probability of default on some of the outstanding loans. It is a global requirement. Here in America it is called CECL (Current Expected Credit Loss), which will be fully implemented by 2020, so we’re working on it with banks right now. 

Retail is another interesting area where we are able to pack optimisation in various sizes for different stores and different locales. The demographics of the people, who shop in each locality or store may be different. Some stores may have more “large” people visiting, some may have a number of “small” people frequenting. By dividing the stores into different categories we are able to stock them with different sizes that were most appropriate for their store. We also do merchandise planning, where we help the merchants plan what to buy for the next year and how much of each thing to buy. We also do pricing optimisation where we help them establish optimal prices. Retail is a very big user right now, especially for advertising. It can really attract people, so they try to find the most effective way to advertise. Of course right now most of our advertising, and our main budget here at SAS, is digital. We’re putting up promotions or ads on websites based on click-through — we get paid a little more money when someone clicks. So if someone is pestering you with an ad all the time, just click on it! 

For manufacturing we have solutions that allow you to predict when one of your assets is going to fail. In the North Sea, we have a lot of oil wells. Each platform has about 200 different pumps on it and we predict when those pumps are going to fail. If our prediction is that they are going to fail in a couple of months, they will go and change them when they have their next maintenance. That type of predictive maintenance is something we are also doing for one of the large truck fleets. All the car manufacturers are interested in that too, as they put more and more sensors on their cars to collect data. They want us to be able to look at if the car is going to fail or their parts are going to fail for some reason. So we create models to predict all of these.

Fortunately, SAS operates in a segment that has only seen a rising curve — right from the time when you started becoming more and more popular. Has there been any point in history, through these 40 years, where you’ve had to rethink your approach, strategy or model in any way?

Back in the early 80s, right after the PC was announced and our customers started asking us to help with it — up until that time we had started running on three mini computers: Digital, Data General, Prime. Those are names you don’t hear anymore. Those were million instructions per second machines. They were used in departments and were much cheaper than mainframes. But for each one, we would have to do a lot of rewriting. We realised that with the proliferation of machines, especially PCs and new operating systems, we needed a better way to be able to make SAS more portable from one machine to another. So we decided to rewrite SAS entirely in C because that was the most prevalent language amongst mini computers. 

You didn’t have to reinvent the company in the sense that it didn’t require a mass reskilling of people or anything like that?

No, we did. We had to retrain everybody to use C. That allowed us then to grow from machine-to-machine and make SAS completely portable. Our latest transition has been over the last seven or eight years, where we realised that as data and problems get bigger, we needed to run massively parallel. A single processor is not going to work anymore for some of the really big jobs. They are still very fast and we came up to 2.5 to 3 billion instructions every second now with these Intel chips. But still, there are things — some of the neural networks, some of the really big data where we want to take the data and spread it out among many different servers. 

Now, if it is a two-slot server, it can have up to 44 cores in it. Twenty-two cores is what they append right now, that you can buy and have on a chip. So you get 44 cores in all. Each of those can be hyper-threaded to have 88 processors, or 88 processors running on one server. And we’d put a terabyte of memory on it. That’s a lot of computing power. You take 10 of those beside each other and we can spread data to all 10 of them. In that way, we can take jobs that used to take 18 hours and run them in 12 minutes. So, that is where we are going right now — massively parallel computing with big data. 

When you identify a new growth opportunity or see a transition in terms of the skill set you may require for the future, how do you go about your people strategy? 

For cyber security, which we have been working on for two-and-a-half years, we hired an ex-NSA employee who really understood it. He in turn hired some others that were very good. We set him off using our administering processing engine to capture all the transactions that occur inside the data centre, as that can be billions a day. We want to know if a desktop machine is talking to a server that it has never talked to before, or is making lots of inquiries which are completely out of the norm for that particular server or for that particular group of servers. Then, we will flag that as a probable security problem and have security look into it, and decide whether we want to shut that machine down. For risk-management and banking, we brought on board a risk manager from JP Morgan Chase. He hired a number of other risk experts from other banks. This allowed us to build that skill set. 

In the tech sector, stock options are a favoured way of rewarding people. So how have you been able to retain talent with such low attrition over the last 40 years?

We pay real money as bonuses and profit sharing instead of a piece of paper. So many people who got their paper back in 1999-2000, turned out they were absolutely worthless. All of that time they spent working – they got paid nothing, basically. I figured that having a great place to work, sharing the profits, with good bonuses in good years is a good way to reward people.

But what was your thought process about keeping the company private at a time when it was just too fashionable to go public and get a fancy valuation? You tried a flotation in 2000, but then kind of let go.

We did talk to several investment bankers and they said, “Your company is kind of ho-hum and you are growing at just 15% a year, which is not saleable in the market today. It’s the ‘new economy’. You have to have a massive number of users and websites.” And I looked at them and wondered, “Are you people crazy?”

We are starting to see a number of tech stocks that got really hot. And now they’re realising that they have never really seen a return on this investment. The stock prices are also going down. So, I think I have a good way of just rewarding people that work here. 

So, does staying private has its own advantage?

Yes, I don’t have some 28-year-old kid on Wall Street telling me what my earnings should be. We don’t have to worry about what the earnings are – we worry about them, but we don’t have to worry about talking about the earnings.

By and large, we are focused on the long term, not the short term. So, up or down a quarter does not really make a whole lot of difference, as long as we are up for years. But all the filings you have to do! So, it’s tough. Michael Dell used to tell me that he was really envious and proud of me. So when he finally went private again, I called him and said, “Congratulations Michael, welcome to the club”. He was very happy because being public is not much fun.

Can you share the leadership lessons you’ve learned over the years? How has your management style changed, if at all?

I don’t think it’s changed much. I’ve never been a micro-manager. I like to hire really strong people and then let them do their thing. Only occasionally I’ll come and say, “You need to change what you are doing” or “I don’t like your approach”, but not very often. We let people make mistakes. I make mistakes. You have to try new things, and if they are not working it’s okay to drop them. Taking risk is okay. We know sometimes you’re going to fail, but you are not going to fail as much as I have.

I have wasted a whole lot of money doing acquisitions that were not the right thing to do, starting companies that ended up going further into the hole. But I do have an expression which I use in all my graduation talks: “When the hole is deep enough, get out of it. Quit digging it.” Some people, because of their pride, will just keep digging it until the hole is so big that the company might collapse around it. Starting a new hole is cheaper! 

Can you give us a couple of examples that taught you the most – that epitomise your expression ‘when the hole is deep enough, get out of it’?

I bought an airline one time and that’s not a good business to be in. After what happened on 9/11 they went belly-up and I wouldn’t put more money into it. I just walked away. When somebody goes bankrupt, everybody who owns stock loses, the debtors end up owning the company. We had a game business one time and we were the sixth largest game distributor in the United States. Last year, before we sold it, the revenue was about $55 million. Then we looked a little further and the expenses were $85 million or something. I couldn’t afford to let this thing get any bigger. These are little things and when you realise that it is just not going to work, you should quit spending money on it and get out of it.

Anything with respect to the core business when you decided to get out as it didn’t quite grow as fast or as big as you thought it would?

Once we develop a new product, we give it about five years and expect to see about a million dollars per developer of revenue coming in. If it’s way below that, then we know that we need to cut some developers on this project and reassign them to something else. So that’s one of the internal measurements we use. We spend about 25% of our revenue on R&D. So the idea is to make sure that they are producing the right products. And if the sales team can’t sell them, then we drop them. 

What interests you most today? Where is the R&D happening? 

We are expanding the different types of neural networks right now. We are opening up the availability to run our Viya massively parallel from any other platform. If you submit a job that is a Viya job, we will automatically see that it is moved up to the cloud, run there and the results go back to you.  

What are the big trends in analytics? You talked about machine learning being one. Have any other sectors surprised you based on the volume of business coming from there?

We are doing an extremely large amount of revenue in administering processing. We have probably the best administering process engine in the world that can handle a billion inputs a second. They get moved through different windows until the transaction is completed. We will write a lot of those up in the Viya and display it. That’s a machine everybody is working on, trying to find the internet of things. This is vague, but it basically means that you’ve got a lot of sensors out there and you are collecting all that data, trying to make sense out of it to help customers.

Our town of Cary actually has smart meters for all the water we use. If they get an unusual high reading, they’ll notify the customer about a leaky faucet.

As part of our smart cities project, we have what we call, a smart campus. Here we monitor all the heating and cooling units, and electrical usage. We are able to forecast when we are going to have a failure in our air conditioning units, so we can get service teams to work on it, right away. There’s a lot of predictive maintenance that goes on to be able to make use of all data that’s flowing in.

The fall in hardware computing costs has really helped SAS and then there is the emergence of big data. Besides these, have there been any other definitive turning points?

We have had steady growth year after year. We specialise not only in statistics, but also in econometric forecasting and operations research, to be able to optimise problems. Now that we have got massive parallel computing, the optimisation can handle much bigger problems, and make it run a little faster. So, you know how to optimise deliveries of your goods and services.

More of our business is moving to the customer who wants us to take their data, and do more than just maintain it — analyse it and send the results back. So we are seeing more of that. That also means that we are bringing their data off-site, so our data centre is growing, our use of virtualisation is growing as well to be able to satisfy all these customers. So that’s been a trend. Most software companies are moving towards cloud strategies including Microsoft.

What do you think are the new frontiers for analytics?

It’s going to be all about machine learning over the next few years. I think there’s no question about that. There is also experimentation to find out what neural networks work for machine vision. They must have tried hundreds of different combinations to try to find out the best solution for this problem. That’s sort of the same thing we did over the years for predicting and probability. But we just know, that’s the way we do it. A lot of model developed is very heuristic. It’s based on trial-and-error. They come up with something that seems to work really well, and then they let everybody know about that. 

With machine learning right now, each instance is a very specific thing. Like, vision: being able to recognise objects. That’s the same machine. It can’t do anything else.

We play a little bit with self-recognising handwritten numbers, and it’s about almost 99% accurate. It recognises 0 to 9 but if you write a ‘B’, it won’t know what it is. If you ask it to tell you what object is in the picture, it’ll probably say 5 or 3. The only result it can give is 0 to 9. It cannot answer any other question. So, we are a long way from machines being able to take over the world. It will probably be another 20 years at least. Because they are all very limited, and there’s always something specific they’ve been trained on. 

How does SAS in 2030 look different from what it is right now?

I don’t know. My time frame is about two years. That’s all I worry about because lots is going to happen in the next two years that we are going to have to react to. I don’t want to be in the middle of a five-year rewrite where we can’t change or take advantage of something new that’s coming out. Two years is about the maximum time it takes us to start a product, and then get it to market. When the iPad came out for example, we said we have to run on the iPad. So, we moved rapidly to ensure that. I think being flexible is very much a requirement.