SAS – who dares wins in database analytics

SAS was founded in 1976 by Dr Jim Goodnight, initially producing software to analyse agricultural data. It has grown from these humble beginnings to over 12,000 employees around the world, operating in 140 countries with 70,000 customers, and having 1,700 partners. It has a famously low employee turnover rate of around 4%, much lower than the technology industry norm of 15% (which rose to 21% in 2023). The company is still privately held, with Jim Goodnight owning two-thirds of the shares and his co-founder John Sall owning the rest. The company has over 700 patents and half of its engineers have advanced degrees (either masters or doctorates).

SAS software ranges from analytics, statistical analysis and visualisation to systems management, with a wide range of vertical solutions such as bank fraud detection, healthcare and insurance solutions, oil and gas exploration analytics and many more. SAS Viya is their cloud-enabled in-memory analytic engine, complementing their previous on-premise software implementation of the various SAS products.

The annual analyst conference was held late in February 2024 at the sprawling SAS campus, set in nine hundred acres of woodland near the town of Cary in North Carolina.

The conference began with some brief thoughts from Jim Goodnight, the founder and CEO of SAS, who is 81 years young. He observed that in the 1970s the industry had time-sharing of mainframe computers (which were scarce and expensive) and now the world has reverted to a rather similar model with cloud computing where you rent processing power from computer clouds like AWS, Azure and Google Cloud Platform.

SAS achieved 8% revenue growth in 2023 on revenues of around $3.5 billion. The company has been profitable throughout every year of its 47-year life and has no debt. Cloud revenue grew 30% last year, with over $500 million of cloud revenue in all.

One Canadian banking customer who spoke at the conference has 52TB of datasets and 300 applications written in SAS including ones that do critical things like employee bonus calculation, regulatory statements and more. All are being steadily migrated to the SAS Viya cloud. As an aside, the customer observed that a recent switch that the bank made from AWS to Azure took over three years and millions of dollars.

A wholesale risk rating application at the bank assesses loan risk, and was the pilot project for migrating from older SAS to Viya – this took five weeks to migrate. They were pleased to find that their code executed around ten times faster than previously. They found they could deploy new models at around half the level of effort needed with the older SAS technology.

SAS has a lot of public sector customers and in 2024 they have an initiative to gain cloud security certification in various countries including FedRAMP in the USA. SAS has invested a billion dollars in AI, from speech and audio processing to machine learning and natural language processing. SAS will in 2024 deliver a rapid development environment for building AI applications. They will also introduce industry-specific AI models as separate products, of which more anon.

The SAS software pre-processes AI prompts that align similar but subtly different phrasing of natural language questions and prompts from users to ensure greater consistency in AI answers. Viya also caches common AI results, which reduces the processing costs of AI by avoiding repeating the same query to (potentially expensive) AI resources.

SAS customer Georgia Pacific (a pulp paper company with $17 billion annual revenue and 30,000 employees) processes two trillion records of data produced by 500,000 sensors in its factories, with 20,000 models running concurrently to do things like locate misaligned boards on its production line via computer image analysis of its machines. Reducing production line interruptions in this way saves them several million dollars a year. They also use AI (using Claude AI from Anthropic) to try to predict paper machine failure based on the sensor data. 20,000 data points per second are interpreted to spot potential problems with the machines and recommend preventative corrective action. The idea is to make better use of older production line machine assets to extend their useful life and defer replacement with expensive new equipment.

A speaker from Lockheed Martin talked about SAS Viya. They are planning to shift the analytics to physically be present on various military aircraft models (rather than running in a cloud) for anomaly detection. He mentioned that getting clean data is quite a big issue when you have the realities of human engineers on the ground actually inputting that data. They reduced downtime by 280 hours per plane after their SAS project.

SAS Viya workbench is a cloud-native development environment, for either SAS language code or Python or (soon) in the R language. A demo of this was given by Jim Goodnight himself. In one example, 60,000 images of apparel were used to train an AI model and then generate images. In a further example, they compared some Python libraries with SAS Viya processes, which are highly machine-efficient and so can speed up many processor-intensive calculations.

They also showed an AI-powered “co-pilot” that helps developers write SAS code.  The tool tries to anticipate the intentions of the developer using natural language prompts; draft code is generated based on the inputs. The tool can also be used to properly generate comments and explanations for the code generated or indeed existing code.

Another feature is the generation of synthetic data, used where sufficient quantities of real training data are either unavailable or cannot be used for privacy reasons. Being able to trace the origins of synthetic data is important for transparency and audit purposes.

In an additional customer example, Rugby France uses SAS to analyse detailed player performance across 150 rugby matches with 540 million data points. Independent firm Futurium Group were commissioned to run 1,500 tests of mathematically intensive calculations on Viya versus several alternative platforms. They found SAS was, on average, 30 times faster than open-source offerings and 49 times faster than an (unnamed) commercial competitor. In a real engineering customer situation that ran Azure and open-source tools on a complex workload, a set of calculations that took 15 hours using this approach was reduced to 15 minutes in SAS Viya.

A development appearing in June 2024 in SAS is the ability to generate AI “model cards” which provide short documents (an idea pioneered by Google) that audit existing AI models. The cards highlight things like the training data used in the model and various performance metrics about the model such as how the accuracy of the model changes over time.

SAS will sell pre-built industry-specific models that will be available either with SAS itself or could be deployed in other non-SAS situations. The idea is to productise these models and release new versions of them based on customer experience. These will include payment fraud detection, document conversion for medical records and AI assistants for inventory planning and law enforcement. As an example, SAS worked with Nestle on an inventory planning issue. Nestle wanted to store fast-moving products as close as possible to the customers that need them delivered. An AI model was developed to help with this optimisation problem. This approach is arguably similar to pre-built AI models sold by Amazon in the AWS marketplace, amongst other examples.

SAS has numerous vertical solutions. In banking, one example is integrated balance sheet management, another is “Customer Intelligence 360” (used for example by Commerzbank), and another is “Anti Money Laundering”. Fifth Third Bank (a US bank with 20,000 employees) have managed to reduce their money laundering alerts by half after implementing this product.

In healthcare and life sciences SAS also has a range of applications as well as in retail supply chain. Calimax (a Mexican retailer) needed to improve demand planning for perishable foods and used SAS for this, reducing error rates in this area from 56% to 12% within four months, resulting in better sales and profit margins. Turkish supermarket retailer Migros experienced 40% incremental sales growth after implementing the SAS supply chain solution to help with putting in a loyalty program that signed up 90,000 members.

A Deloitte practice leader gave a talk about their partnership with SAS, in their case, particularly in the financial services industry. They developed a joint approach to offering industry solutions like commercial claims counter-fraud. SAS also partners with technology companies like Intel, who provide chips to SAS optimised for specific types of processor workload. The fast database Singlestore is another partner, who SAS deploy in appropriate situations, for example migrating customers off the old SAS internal database application.

Moving forward, SAS will continue to emphasise Viya as a data and AI product. There will be a greater marketing emphasis on vertical industry solutions. They have developed campaigns aimed at specific categories of customers such as data scientists, and not just senior IT executives. SAS acknowledged that it currently spends less on marketing than the industry norms (though at present it does not share the exact percentage of revenue that it spends on marketing) and it will be interesting to see if this changes at all in the future.  SAS is doing preparatory work for a potential IPO, possibly in 2025.

SAS is a company with an unusual culture in the software industry. It was summed up quite well by a SAS customer at the conference who had worked with them for several years. He told me in a conversation we had that SAS was not particularly aggressive at sales and marketing but their staff were refreshingly honest about what their product could and could not do, which he had found was a real rarity amongst software vendors. The SAS software is functionally mature and he had found that the SAS staff he had dealt with were very committed to fixing any issues that did crop up and making sure that he succeeded as a customer. That seems quite reassuring.