Watch this Ask the Expert session to learn how AI can help you solve marketing challenges in an ethical and sustainable way.
Watch the Webinar
You will learn:
-
How a practical and embedded AI approach in the MarTech stack can support and enhance customer communications.
-
How SAS Customer Intelligence 360 delivers AI capabilities today and what it will deliver going forward.
-
The practical first steps to using AI in your marketing approach, from audience or segment definition through content creation.
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
Heard about Viya Copilot, is this part of CI Copilot?
Currently, the Viya copilot and the CI 360 copilot are two entirely different Copilots. These Copilots currently uses different model providers; CI 360 uses Amazon Bedrock while Viya uses Azure AI Foundry. However, we are working to maintain consistency across both and could imagine a future with integrations between the two.
Do we have any approximate timeline when the AI Agents in Ci360 would be available to test for partners and for customers to use?
We're looking at an April release for the Journeys agent. The other thing that I really should have mentioned in the presentation is we are 100% looking for early adopters or testers for this as well. If there's anyone that you can reach out to from your account team at SAS or you can get in touch with me directly, I'd be happy to talk a little bit more about what that would look like.
How do you make sure that any application of AI in your tool is ethical and does not demonstrate bias?
Yes, I want to start with the second part first ensuring the application does not demonstrate bias. That is obviously very important with language models. The challenge is that there is no way to be 100% certain that bias will never occur. Internally, we do a lot to make our AI as ethical and trustworthy as possible. We use guardrails, which monitor the inputs and outputs of AI to ensure they are on topic, relevant to the use case, and not introducing bias. This approach catches many potential issues. Additionally, before any AI release, we conduct extensive internal testing, specifically testing the system's guardrails by asking the AI difficult questions to ensure it refuses to produce biased responses and results. We also collaborate with the Data Ethics Practice at SAS, who provide valuable guidance on these processes. The reality is that AI systems are non-deterministic, and we cannot fully guarantee the absence of bias. However, we strive to mitigate this risk through thorough testing and robust guardrails.
Will agents be able to fix errors in a customer journey in the future?
Yes, this is a part that I am very excited about, especially with agents and the Journeys agent you just saw. We teach the Journeys agent how to build a valid journey in CI 360, but we are not just hoping it gets it right the first time. These agents work iteratively. For example, the agent will try to submit a journey, and CI 360 might respond with an error message, such as “You can't link that task type,” or “You used the wrong type of node,” or “You entered an invalid schedule.” The agent can then take that feedback, learn from it, and try again. This is a great feature of agents—they can improve over time by learning from their mistakes. So, to answer your question: yes, the Journeys agent should be able to fix errors. It already does this as it builds out journeys for users.
Is Agent specific for a domain or it can be universal?
If we go back to where you're mapping it out, and consider the main modules as CI 360, when we talk about domain-specific agents in the context of our development, we really need an agent for audiences, an agent for journeys, and so on. We are developing domain-specific agents in that sense. However, keep in mind that it's all under one multi-agent ecosystem. For the end user, you are interacting with a single agent, which delegates tasks to the appropriate domain agents. That is our current strategy.
Would it be possible to feed a requirements file (for example, as an Excel file) to the Journeys agent, and the agent would then output the Journey?
We've given considerable thought to this and discussed it with some customers, particularly regarding Journey briefs, which are often seen as Word or PDF documents. I don't see any issue with Excel files either. We just need to ensure we support uploading that file type. The concept is solid: you provide requirements to the agent, and it parses those requirements to build out the journey for you. What's really exciting about our work with agents is that, as we introduce more agents for different tasks—such as audiences, segmentation, and journeys—you can upload increasingly large requirements files, and the system will be able to configure many objects for you and present them for human review.
Can AI-created Journeys be edited by humans and vice-versa?
That’s a great point—I should have demonstrated that in the demo. In my example, I uploaded a brief, the agent generated a plan and built out the journey. What I didn’t show is that, after the agent had created the journey, I can manually edit any part of it. Alternatively, I can continue interacting with the agent and make additional requests, such as asking it to change SMS notifications to emails for specific nodes. The idea behind our agent strategy is that you can collaborate with them manually or provide feedback to ensure we’re building campaign objects exactly the way you want them.
Can prompt engineering enhance the use of the Copilot?
I'm curious whether you mean the prompt engineering work we do, or if you're referring to tips and tricks for prompting Copilot to get better responses. I'll assume it's the latter. Something that's really helpful is that we have a variety of default prompts. As you navigate CI 360, depending on where you are, it suggests different prompts to help get the conversation started. I recommend reviewing these and observing how they're formatted—they serve as best practices for interacting with Copilot. Keep in mind that CI 360 is a large solution with many capabilities, so being specific in your questions will usually yield better responses. For example, if you're asking about a task, specify the type of task, like an email task—bulk versus triggered. Being precise with Copilot will help you receive more useful answers.
Does it work equally well in all languages, such as when database fields are non-English?
In our white paper, we discussed this in considerable detail. We develop our capabilities in English first, and that is where we conduct most of our testing. However, we also perform language evaluations for all the languages supported by our UI. Part of our decision criteria when selecting which large language model to use is assessing the model’s performance in different languages. The model family we most often use is the Anthropic Claude models, and we have found that they provide robust language support for the languages used by most of our customers. We have received positive feedback from customers worldwide regarding language capabilities, and it is certainly something we strive to maximize. Although achieving perfection is difficult, especially with regional dialects and nuances, it remains somewhat challenging for AI models.
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.
