Watch this Ask the Expert session to learn how financial institutions can maximize collections, improve effectiveness and monitor strategies.
Watch the webinar
You will learn how to:
-
Integrate a multi-level strategy.
-
Prioritize collections cases.
-
Simulate, learn and prove value through safe innovation.
The questions from the Q&A segment held at the end of the webinar are listed below and the slides from the webinar are attached.
Q&A
What governance or monitoring is needed when using GenAI in regulated environments?
GenAI in regulated environments is typically classified as high‑risk, so it requires end‑to‑end AI governance and continuous monitoring. This includes clear guardrails using business rules, strict security and compliance checks, and full traceability of prompts, outputs, and model versions.
A human‑in‑the‑loop is essential to validate outcomes and provide feedback, ensuring the model supports decisions but does not autonomously make them. Together, these controls help to maintain safety, reliability, and regulatory compliance throughout the system’s lifecycle.
What are the key steps to integrate GenAI safely into an existing decisioning engine?
In general, in high-risk scenarios, you need to start thinking about strong governance and a trust foundation, not just as an add-on to your strategies, but as something built in from the start. From the beginning, you need to consider how to use your business rules and logic, not just as a translation of your business processes, but also as guardrails and constraints for the language model. Use business logic to instruct the LLM on what it can and cannot do.
It is very important to establish your non-negotiables from the outset. Ensure you have requirements for traceability of your decision paths and LLM responses, as well as editing and explainability. These are non-negotiable, especially in highly regulated environments. Align everything you do with language models and agents with your existing enterprise model governance policies. Do not consider these as separate from your model ops or decision ops processes, but as essential components of them.
You will need to consider building a GenAI gateway that allows you to control permissions and determine which regional and data movement rules you want to apply to these systems. Another consideration is that, when you combine deterministic models and business rules with LLM outputs, the LLM output should serve as an input into your process, not as a decision in itself. Use those rules and deterministic models to validate what the LLM-generated text produces.
It is important to work on the foundations, particularly around data and infrastructure, not having it, leads to poor outputs and compliance problems. Before going into a full deployment, it is important to start with pilots that allow an understanding on usage and business impact. When starting those pilots, consider involving different people in the process from early stages to ensure you complete a checklist where different stakeholders are prepared and confident to use their AI systems within the organization.
Also, as part of those processes, users should ensure traceability, process lineage and quickly identify failures using guardrails.
Since we have a massive investment in our CRM and banking platforms, is this intended as a replace for those or does it integrate with the systems we already have?
It is definitely not a replacement. You can think of it as an extra intelligence layer that enhances your existing platforms. Your current systems, whether CRM or any other enterprise tools, remain your systems of record and engagement. We use data from these systems and integrate with third parties easily and securely through APIs. This allows you to define your processes and business logic, apply advanced analytics and AI to that data, and return decisions both to you and back into the systems your teams are already using. It is fully integrable with your current tools and systems.
Are Agents available for SAS Viya 3, or only on SAS Viya 4?
Our current integration efforts for generative AI are focused exclusively on Viya 4. There are two main reasons for this: first, our innovation efforts in R&D are centred on Viya 4; second, certain infrastructure and architectural decisions were required. Since Viya 4 already features a microservices architecture that is cloud-ready, with an API-first approach, it is simply better suited for the agent setup we need. Therefore, our integration will be limited to Viya 4.
Integrate with GitHub?
We have out-of-the-box approaches for versioning, check-out, comments, and all of that built into Viya, especially for our decisioning components. Our decision flows and its objects is versioned directly through Viya.
What about orchestration? How much control do I have over it?
Currently, the orchestration of the strategy occurs as you define your workflow, and you have options to call different agents or sub-strategies as part of your flow. This provides a more prescribed way of using language models—while the LLM can be involved, it is not necessarily making the plan; you already have a plan for how the strategy or the paths are to be taken. These act as guardrails around the LLM responses, giving you substantial control over the possible paths during strategy optimization.
Can we access trial Viya RAM version? In my organization we have SAS Viya 4 version.
That is, I believe it's a separate process from the Viya 4A trial setup since it is a distinct application from Viya. However, you'll be able to find information on the website for the RAM application about how to access that trial.
SAS Viya 4, in which LLM model are available? Are they using all open-source model or we can use OpenAI , Google Gemini , Kimi k2 models etc. // can I choose my own models?
As part of your decision flows, you can integrate with any of the currently available models. We offer the ability to make API calls to those models. In the next few releases, we are planning to use our model repository on Viya to provide governance, allowing you to register models within your registry alongside your deterministic models, enabling traceability and monitoring. For now, integration occurs out-of-the-box through an API call, which is agnostic of the model provider.
Does Viya have its trained LLM's or it can also be used to leverage other models like Claude, ChatGPT, etc. with integrations?
Viya is LLM agnostic and can integrate with Claude, ChatGPT and other LLM models. In the next releases, you will be able to leverage Viya's model repository to ensure governance to those models.
Can I use my own MCP servers?
With the launch of SAS tools externally in the next few months, you will be able to leverage these in your own systems, combined with your own MCP servers. We will release more information on the format that our tools will be available for consumption in the next few releases in Q1. One example is that you'll be able to deploy decision flows as MCP tools out of the box, to leverage in your own multi-agent implementations.
Would I be able to also work from the cli and actually make calls to SAS (kind of headless calls)?
Yes, we have an extensive list of open APIs that you can use. I suggest that you visit developer.sas.com, where you can find a list of elements available for integration over time. As mentioned, we already have an MCP server available for code generation. In the future, we will release more of our tools on Viya, as well as tools from the analytics cycle, which can be used with other MCP servers. You can essentially use what is currently available.
Recommended Resources
Please see additional resources in the attached slide deck.
Want more tips? Be sure to subscribe to the Ask the Expert board to receive follow up Q&A, slides and recordings from other SAS Ask the Expert webinars.
