Moving it to continuous, near real-time is a meaningful shift.
At its annual Innovate conference in Dallas, SAS presented a suite of purpose-built AI agents and model pipelines aimed at some of the most consequential and least forgiving domains in modern enterprise life — supply chains, workplace safety, government benefits, and financial fraud. The announcement arrives at a moment when the gap between AI ambition and AI readiness has become impossible to ignore: organizations are deploying fast, but governance is fraying. SAS is betting that pre-built, industry-specific solutions can close that gap before the consequences of unguided AI become irreversible.
- A joint industry study found that 75% of anti-fraud professionals are watching financial fraud surge, yet only 7% feel their organizations are meaningfully prepared — a readiness crisis that SAS's new fraud models are designed to address.
- Supply chain planning, long a monthly ordeal of spreadsheets and cross-departmental marathons, is being challenged by a continuous AI agent that can model demand shocks and explain its reasoning in plain language through a chat interface.
- Digital twins built inside a game engine are helping a medical device sterilization company unblock surgical instrument pipelines — and training computer vision models on simulated workplace hazards that real footage cannot safely capture.
- State governments face direct federal fines when benefit payment errors exceed thresholds, and SAS is deploying machine learning in Nevada and elsewhere to surface those errors without requiring states to rebuild their existing infrastructure.
- SAS is framing its entire accelerator portfolio as a corrective to the most common AI failure mode it observes: organizations assembling ad-hoc experiments that never cohere into competitive advantage, especially as governance pressure intensifies in regulated industries.
At SAS Innovate 2026 in Dallas, the analytics company used its 50th-anniversary conference to name what it sees as enterprise AI's central failure: speed without governance. Its response is a portfolio of industry accelerators — pre-built AI agents and model pipelines tuned for high-stakes, regulated environments — anchored by the SAS Supply Chain Agent, now in private preview.
Supply and operations planning has long been one of corporate life's more punishing rituals: a monthly, multi-day process pulling teams into spreadsheet-driven forecasting sessions. The new agent runs continuously, balancing demand and supply in near real time, and lets users query it conversationally — asking, for instance, how a sudden 15 percent demand drop should be absorbed — and receive not just projections but explanations. IDC's Kathy Lange observed that most pre-packaged agents handle basic processes, and that compressing S&OP's complexity into an agentic framework is a meaningful step forward.
Elsewhere, SAS showcased digital twins built inside Epic's Unreal Engine. A medical device sterilization company, whose bottlenecks can delay surgical procedures, used simulations to discover that a buffer lift was acting as a distribution hub rather than a simple conveyor — a misdiagnosis that targeted adjustments corrected. The same simulation logic powers SAS Worker Safety, which generates synthetic training footage of rare hazards like forklift collisions, training computer vision models that then monitor facilities in real time without exposing any real employee data.
On the government side, SAS is helping states including Nevada manage SNAP payment integrity. Federal rules now fine state budgets directly when benefit error rates exceed set thresholds, and those compounding miscalculations cost states millions while leaving eligible families underserved. SAS's tool connects to existing eligibility and case management records without requiring infrastructure overhauls, using machine learning to surface error patterns for already-stretched caseworkers.
The financial fraud picture is, by the industry's own measure, dire. A SAS and ACFE study found 75 percent of anti-fraud professionals reporting surging fraud, with 55 percent expecting deepfake and AI-generated document forgery to worsen — yet only 7 percent feel adequately prepared. SAS Fraud Decisioning for Payments draws on a consortium dataset from major global institutions, covering everything from credit card fraud to money mule detection, so adopters begin with the accumulated intelligence of millions of catalogued fraud events.
All of it falls under SAS's stated $1 billion investment in industry solutions. Global Market Strategy Lead Manisha Khanna framed the accelerator approach as a direct answer to the failure mode she sees most: organizations stitching together AI experiments that never cohere into real advantage. Whether enterprise buyers will embrace purpose-built solutions — or keep building their own — is the question the market will now answer.
In a convention hall in Dallas last week, SAS used its annual Innovate conference to lay out what it believes is the central problem with enterprise AI adoption: companies are moving fast, breaking things, and quietly abandoning the governance guardrails that make AI trustworthy in the first place. The company's answer is a portfolio of what it calls industry accelerators — pre-built AI agents and model pipelines designed for specific, high-stakes problems in regulated industries. The centerpiece announcement was SAS Supply Chain Agent, now in private preview and heading toward a global enterprise rollout.
Supply and operations planning — the process manufacturers and retailers use to manage inventory across six to twelve months of shifting demand — has long been one of the more punishing exercises in corporate life. It typically takes multiple days, pulls professionals from several departments into spreadsheet-driven forecasting sessions, and happens, at best, once a month. The scale of managing thousands of supply chains through that kind of manual process is, as SAS describes it, a longstanding wicked problem. The new agent runs continuously, balancing demand, supply, and operations in near real time, and lets business users interact with it through a chat interface. A user can ask it to model a scenario — say, a sudden 15 percent drop in demand — and receive not just a projection but an explanation of how the agent reached its conclusions. Kathy Lange, Research Director at IDC's AI, Data, and Automation Software practice, noted that most pre-packaged agents on the market today handle basic processes; compressing something as complex as S&OP into an agentic framework, she said, positions SAS to bring its deep supply chain knowledge into a new generation of AI solutions.
The conference also showcased work SAS first previewed at its 2025 event: digital twins built inside Epic Games' Unreal Engine. One of the more striking examples involves a major provider of medical device sterilization. Surgical teams depend on fully sterilized instruments — scalpels, clamps, and the rest — and any bottleneck in the sterilization process can delay lifesaving procedures. The company believed trays of medical tools were getting stuck in a buffer lift that lined them up for cleaning. By building a digital twin of their facility and running simulations, they discovered the real issue: the buffer lift was functioning as a central distribution point, not a simple conveyor. Targeted adjustments broke the bottleneck and accelerated production.
Worker safety is another domain where SAS is deploying synthetic data and computer vision. The Bureau of Labor Statistics puts the annual count of fatal workplace injuries in the United States at more than 5,000, with falls, machinery accidents, and improperly worn protective equipment accounting for a significant share. SAS Worker Safety uses digital twins to generate realistic training footage for computer vision models — footage that can capture rare but plausible events, like a forklift collision, for which real video may not exist. Once trained, these models run on cameras throughout a facility, issuing real-time alerts when a helmet is mispositioned or a mask slips in a medical setting. Because the training uses fully simulated worker personas, no real employees are involved and no personal data is exposed.
On the government side, SAS is working with multiple states, including Nevada, to address errors and fraud in the Supplemental Nutrition Assistance Program. Federal rules now allow regulators to fine state budgets directly when payment error rates — benefits over- or under-awarded due to eligibility miscalculations or undetected fraud — exceed a set threshold. Those compounding errors can cost states millions in federal funding, and the families who need assistance most may be receiving less than they qualify for. SAS Payment Integrity for Food Assistance connects to a state's existing eligibility records, case management files, and transaction histories without requiring a full infrastructure overhaul, then uses machine learning to surface error patterns and prioritize leads for caseworkers who are already stretched thin.
The financial fraud picture is, by the industry's own accounting, grim. A joint study by SAS and the Association of Certified Fraud Professionals found that 75 percent of anti-fraud professionals are seeing a surge in financial fraud and scams, and 55 percent expect deepfake social engineering and AI-generated document forgery to increase significantly over the next two years. Only 7 percent of those surveyed felt their organizations were more than moderately prepared to detect or prevent AI-fueled fraud. SAS Fraud Decisioning for Payments draws on models trained across a consortium dataset contributed by major global financial institutions, covering credit card, debit card, ATM, digital wallet, and application fraud, as well as money mule detection. Institutions deploying these models are not starting from scratch — they are learning from millions of fraud events already catalogued across the industry.
All of this sits under the umbrella of SAS's stated $1 billion investment in industry solutions, announced as the company marks its 50th year. Manisha Khanna, Global Market Strategy Lead for Applied AI at SAS, framed the accelerator approach as a direct response to the failure mode she sees most often: organizations stitching together ad-hoc AI experiments and never achieving the competitive edge they were chasing. The question now is whether enterprise buyers, facing intensifying pressure around AI governance in regulated industries, will move toward purpose-built solutions — or keep building their own.
Citações Notáveis
Current pre-packaged agents tend to tackle basic processes; with Supply Chain Agent, SAS is compressing a very complex process, which could deliver significant value.— Kathy Lange, Research Director, IDC AI Data and Automation Software practice
When organizations are left stitching together ad-hoc AI frameworks and experiments, they often fail to achieve the competitive edge they're looking for.— Manisha Khanna, Global Market Strategy Lead, Applied AI at SAS
A Conversa do Hearth Outra perspectiva sobre a história
What's the actual problem SAS is trying to solve here — isn't AI everywhere already?
Everywhere in theory, yes. But most organizations are still stitching experiments together. The gap between a proof of concept and something that runs reliably in a regulated environment is enormous.
Why does the supply chain agent feel like the headline announcement?
Because S&OP is genuinely one of the most painful processes in enterprise operations. Running it once a month, manually, across thousands of supply chains — that's not a small inefficiency. Moving it to continuous, near real-time is a meaningful shift.
The digital twin work is interesting. What does building a twin inside a game engine actually get you?
Photorealistic simulation without the cost or risk of physical testing. For the sterilization company, it meant discovering that their bottleneck diagnosis was wrong — before they spent money fixing the wrong thing.
The worker safety application surprised me. Why use synthetic data instead of real footage?
Two reasons. Real footage of serious accidents is rare and ethically fraught. And synthetic data lets you vary conditions endlessly — lighting, equipment color, body position — in ways real footage never could.
The SNAP fraud angle feels different from the others. What's the stakes there?
Federal fines tied directly to state budgets, and families who may be getting less assistance than they qualify for. It's a compliance problem and a human welfare problem at the same time.
The fraud numbers are striking — only 7 percent feel prepared. Is that a SAS talking point or a real finding?
It came from a joint study with the Association of Certified Fraud Professionals, so it reflects the industry's own self-assessment. That's a candid number for a field that tends toward confidence.
What's the thread connecting all of these — supply chains, safety, food assistance, fraud?
Regulated environments where the cost of getting it wrong is high and the talent to build AI from scratch is scarce. SAS is betting that pre-built, domain-specific tools beat general-purpose ones in those conditions.