How Williams-Sonoma uses Intelligent Context

Hey, I'm Alison Marr, a certified application architect here at Salesforce. Today I want to show you how AgentForce creates context Aware experiences at scale. Let's take Williams Sonoma. They're a specialty home retailer and customers love their visually stunning in store experience, which they've actually been able to translate to their website using their newly launched sous chef agent. The agent provides customers with reference, recommended recipes, even related products like this La Creuset Dutch oven. But what we've learned is in order to answer complex questions, our agents need to be trained on the right data. And enterprise data is complicated. Your data could be in the form of process flows with all this complex logic. Or product manuals with beautiful rich images and charts. All that content needs to be turned into context for an agent to really, really do its job. Let's say I ask a question, I just got a Dutch oven and I'm going to try to make braised short ribs. But I know it's really important for them to not dry out. We can see the agent's not really able to give a great answer, and that's because I haven't trained it with the context specific to this product. Let's take the manual for this Dutch oven. We can see it's full of rich images, charts, even this diagram, which is actually relevant to the question I just asked. Now, in the past, these images and charts would have been really hard for an agent to interpret because if you just copy and paste the text directly out of this image, you end up with this, which honestly looks like complete garbage. All the context of the data is lost. So no human or agent would really be able to understand this. And that's why we created Intelligent Context, a new way for agents to understand your data. Let's try it out. I'll go ahead and upload that same product manual into the builder after the file's uploaded. Notice how it gave some recommended settings. It understands the content we just uploaded. And it's going to recommend the best way to go ahead and create that search index. And having the right index is what allows the agent to quickly understand the content. So once it completes, we can ask the agent test questions and ensure the agent is answering how we'd expect. And check this out. Not only does it give us the correct answer, but it even gives the references or chunks of the file that it used to find that answer. If we look closely at this chunk, notice how it's not just copy and pasted text from the file. This is where the magic context indexing really comes in. Okay, because this is the LLM's understanding of the content, and that's what's coming out through the chunk here. Agentforce is understanding the content in that PDF. It's creating a text version of that understanding, and that's what allows us to give truly accurate answers. And if it wasn't quite right, the builder now gives you options to directly adjust that configuration. This is going to save hours of time that you previously would have spent testing agent responses, adjusting parameters, making sure that everything was interpreted correctly. So once we're happy with it, we can publish it, which will apply this configuration to any of the other data sources our agent is ingesting. Let's break down how that works. We just took this one PDF and used Intelligent Context to create vector embeddings to understand the rich content. Then it created a search index to quickly retrieve answers about the content. And since most companies have more than one product, when we publish, we're setting that same search index to apply to all the product data. But the configuration for understanding product manuals probably won't work for other types of data like sales reports or website pages. And that's why Intelligent Context makes it easy to create and publish different configurations to optimize for all of your data types. Now that we publish, let's go back to the Williams Sonoma website to try it out. I'll go ahead and ask the agent the same question about the braised ribs I'm making, and this time the agent's able to provide an informed answer and it looks great. It's grounded in the understanding understanding it has from the product manual. So now you've seen how AgentForce gives customers accurate, expert advice throughout every interaction by providing the right context grounded in your business data. To learn more, visit agentforce.com and check out the link below for more details on the data. 360 launch.