A billion dollars from SAS to bring agent-based AI to regulated processes
General-purpose agent-based AI, based on horizontal assistants that rely on large language models, shows its limitations when it needs to operate in regulated processes and in those with high domain complexity. There is a lack of industry-specific data, a lack of vertical models trained on years of industrial practice, and above all, a lack of ability to trace and justify every step of the decision to the regulator. This is the historical ground of SAS, which presented a series of innovations at Innovate 2026 within a billion-dollar plan dedicated to industry solutions.
Supply Chain Agent: From Monthly Planning to Continuous Planning
The main announcement is the SAS Supply Chain Agent, now available in private preview to a select group of clients and soon to be released globally. The agent tackles one of the most challenging processes in retail and manufacturing, sales and operations planning, typically executed once a month and requiring weeks of work on spreadsheets that span multiple corporate departments. SAS transforms it into a continuous process. The agent balances demand, availability, and logistics in real time, allowing business users to build hypothetical scenarios through a conversational interface: one can ask how planning would change in the event of a 15% drop in demand and receive not only the result but also the rationale behind the decisions made by the model.
Kathy Lange, Research Director of the AI, Data, and Automation Software practice at IDC, summarizes the significance of the announcement: "Pre-packaged agents currently available generally address basic processes; here SAS is working on a very complex process, and the results can be significant." Lange's judgment is supported by the company's history. SAS has been working for years on supply chain analysis and already possesses the library of models and domain knowledge necessary to support a planning agent of this magnitude.
Industry Copilot and Real-World Cases in Production
Alongside the Supply Chain Agent, SAS introduces the first industry Copilots. The SAS Asset and Liability Management Copilot guides financial risk analysts in configuring and interpreting asset liability management scenarios. The analyst formulates requests in natural language, and the assistant converts them into an operational setup of the already governed analytical models, without needing to write code or manually intervene on the underlying technical parameters.
The SAS Health Clinical Data Discovery Copilot accelerates the exploration of clinical data, cohort creation, and quality checks for the healthcare world. The 2026 roadmap plans to extend the Copilots to financial crime in banking and to optimize planning and supply chain in manufacturing.
Regarding cases already in production, the State of Nevada and other American states are using SAS Payment Integrity for Food Assistance to reduce errors in federal food assistance programs, avoid penalties introduced by new regulations, and improve the delivery of subsidies to those who are actually entitled to them. Banks and insurance companies rely on SAS Fraud Decisioning for Payments, trained on a dataset constructed through a consortium of large global financial institutions and covering fraud on credit cards, debit cards, ATMs, digital wallets, applications, and more recent schemes like money mule, the practice through which criminal organizations transfer illicitly obtained money through a network of often unwitting intermediaries, recruited online or via social networks and used as a conduit to move funds between different accounts. In the industrial sector, SAS Worker Safety combines digital twins built in Unreal Engine, synthetic data, and computer vision to train models that detect in real time missing or incorrectly used protective equipment, even in rare scenarios like collisions between forklifts, for which there generally aren’t enough real footage for training.
Manisha Khanna, Global Market Strategy Lead Applied AI at SAS, summarizes the company's approach: "When companies piece together frameworks and experiments in AI in a fragmented way, they end up not achieving the competitive advantage they sought from the investment. We build industry accelerators to solve real and defined problems in highly regulated environments, with agents and models ready for production that work on data companies already have." This stance is consistent with SAS's history, built on closed products calibrated to specific vertical processes, opposed to the logic of 'build everything from scratch' characteristic of generalist ML stacks.
The innovations in verticals are seen in continuity with the other major line presented at Innovate 2026, the opening of the Viya platform to agents based on external LLMs through the Model Context Protocol Server. Once SAS functions can be called upon by agents built on Claude or other models, the Supply Chain Agent or Fraud Decisioning becomes usable services even from within non-proprietary stacks. The direction is consistent: SAS aims to no longer base its value on possession of chat and orchestration, but to leverage its expertise on domain data and governed models.