SAS's Unconventional AI Playbook for Productionizing Prototypes

Data science and AI teams are all familiar with the paradox that enterprise AI currently faces: impressive prototypes and experiments but little tangible return on their substantial investments. This has led many CEOs to lament failed implementations, regulatory compliance nightmares, and a widening gap in turning proof-of-concepts into production reality behind closed doors. They realize that GenAI alone cannot solve complex enterprise data challenges.

Established players like SAS see this POC-to-production crisis as an unexpected opportunity. The 49-year analytics veteran is leveraging strategic acquisitions, investments in emerging technologies like quantum computing and digital twins, and a renewed focus on multi-cloud flexibility.  The question is whether it's repositioning as the anti-hype alternative will resonate in a market tired of hyperbole.

The right technology for the right problem

Research shows the global market for agentic AI will reach over USD93 billion by 2032, with many organizations already integrating AI agents into their workflows. Yet, beneath these promising statistics lies a troubling disconnect between AI investment and measurable outcomes.

Speaking at the sidelines of SAS Innovate on tour 2025 in Singapore, Gavin Day, SAS’s executive vice president and chief operating officer, acknowledges that many customers have spent “significant amounts of money prototyping and doing cool stuff” with GenAI, but they struggle to “make the leap from what you’re doing in the back of the room… into something that generates value”.

SAS is zeroing in on this production gap by emphasizing a core principle: using the right AI technology for the right problem. This means recognizing that there is a time and place for traditional machine learning and statistical methods, a time for GenAI, and a time for new approaches like agentic AI. The goal is to move beyond the buzzwords and deliver solutions that are reliable, compliant, and scalable for enterprise deployment.

Agentic AI: SAS’s answer to GenAI limitations

SAS is positioning agentic AI — AI systems that can operate with varying degrees of human oversight — as the next evolution beyond simple GenAI models. As senior vice president of global marketing, Patrick Xhonneux explained in his opening keynote, SAS defines this as a spectrum between “human out of the loop” for autonomous decisions and “human in the loop” for flagged interventions requiring human review.

In a mortgage lending scenario, an intelligent agent analyzes irregular transactions, calculates risk, and compiles a summary for human review. The key differentiator, according to Xhonneux, is not just the agent but the transparency and explainability it provides. The platform offers “model cards” that function like a “nutrition label” for the model, providing details on usage, data summary, performance, and fairness. For a data scientist, the platform also provides a “decision logic flow” that gives a transparent look into how each applicant moves through multiple decision points in the workflow. This approach uses a deterministic model to make the decision while using probabilistic LLMs to provide the explanation.

This focus on explainability directly addresses a critical pain point: systemic bias. A recent study found that large language models recommended “more denials and higher interest rates for black applicants” compared to white applicants with identical profiles. The discrepancy was found “across all language models, including OpenAI, Anthropic and Meta”. SAS’s model governance technology, which has been developed over decades, is now adapted to monitor for drift and bias in newer technologies, running “model tournaments” to ensure models are performing as expected.

Why synthetic data matters to enterprise AI analytics

SAS’s acquisition strategy is less about buying revenue or customers, but more about acquiring “pieces of technology that complement either gaps that we have or to accelerate time to market,” said Day. The acquisition of Hazy, a synthetic data specialist, exemplifies this. The purchase directly addresses training data limitations in regulated industries where “the data training needs exceeded what they could generate and exceeded what they have”.

The synthetic data play serves dual purposes. It enables innovation in data-starved environments and provides a cost and productivity benefit. As Riad Gydien, executive vice president and chief sales officer, noted, in a regulated industry, you cannot make sensitive data “generally available to analysts everywhere to go and innovate”. Synthetic data reduces the cost and complexity of innovation, allowing analysts to work with “almost freely available data” before a model moves into production, where real data is used for inferencing.

Reimagining analytics gaming engines and quantum AI

Perhaps SAS’s most ambitious AI integration involves partnering with Epic Games to bring Fortnite’s Unreal Engine into industrial analytics. The collaboration with Georgia-Pacific demonstrates this approach, where SAS optimized manufacturing plant operations using photorealistic digital twins.

The results were counterintuitive: “Adding extra AGVs (automated guided vehicles) actually is going to slow down your operation” due to traffic jams, requiring “to decrease the fleet side by three vehicles” to improve average mission time by 8%. This type of complex optimization problem, involving 400-meter-long facilities “larger than four football fields,” showcases how game engines can surface insights impossible to visualize through traditional dashboards, said Xhonneux.

The digital twin strategy extends beyond manufacturing into employee safety, where “we can add digital humans...test different scenarios to identify potential dangerous situations without causing risk to employees in the real world.”

SAS is also exploring quantum computing as an optimization accelerator, positioning QPUs alongside CPUs and GPUs in hybrid architectures. Early results are promising: a recent implementation for “a very large CPG organization” achieved “300% improvement in results and a 97% reduction in time of computation,” said Xhonneux.

The company’s quantum strategy emphasizes seamless integration: “SAS software will automatically select which portion of your problem should use the CPU, the GPU or the QPU,” making quantum acceleration transparent to end users.

Breaking down the “walled garden”

SAS is making a calculated bet against the "walled garden" approach adopted by some competitors. The company’s Viya platform supports multiple programming languages (Python, R, SAS) and deployment models (on-premises, hybrid, multi-cloud). “SAS is not a walled garden,” Day stated definitively. “The days of monolithic software vendors are over”. This architectural flexibility addresses a critical enterprise reality where customers need to use a "best of breed" approach and integrate with various vendors.

The strategy also addresses the challenges of data movement and sovereignty. SAS has engineered its technology to “bring the analytics to the data” rather than moving the data to the analytic engine, said Day. By placing the SAS container next to a Snowflake or other containers, analytics can run right next to the data, reducing data movement costs and improving performance.

The challenges of a 49-year legacy

SAS’s long track record is both an asset and a challenge. Day acknowledged that the company grew organically, which led to multiple financial systems and self-contained business units in various countries that “don’t scale”. The company has undergone a significant infrastructure modernization journey, centralizing finance and HR and consolidating financial systems into Oracle Fusion in preparation for potential public markets.

Gydien also noted that while SAS has a “global install base of hundreds of thousands of sites,” many of these customers “do more than one thing with us out of the entire portfolio we have”. This suggests that while new customer acquisition is essential, a significant opportunity lies in expanding their footprint within existing accounts. This is a “much more nuanced go-to-market approach” than simply focusing on new logo acquisition, Gydien added.

Shaping the future of enterprise AI analytics

As the analytics landscape evolves with AI, SAS is betting that enterprises will ultimately prioritize governance, explainability, and production reliability over experimental capabilities. This approach is informed by the company’s multi-decade perspective on technology hype cycles, which helps them “look and prioritize our investments”.

For data scientists and machine learning engineers, a critical shift is underway: the competitive edge in enterprise AI is no longer about experimental capabilities but about production-ready, governed systems that deliver measurable ROI. The conversations at SAS Innovate on tour 2025 in Singapore confirm that this change couldn’t have come a moment too soon, signaling that SAS is listening closely to the market pulse.

It also means that the real disruption isn’t just happening in Silicon Valley startups. Instead, it’s occurring in the unglamorous but essential work of making AI systems reliable, compliant, and scalable for enterprise deployment. And based on SAS’s moves, that is where the real battle for mindshare is occurring. 

Image credit: iStockphoto/IvelinRadkov