A lot of people are asking, how are these two things Joule and Just Ask is now connected?
So let’s say I’m an end user and I’m asking a question. Well, in the context of what we are interested in now, there are only two scenarios. Either it’s an other question. So the other question goes to another service and we don’t have to worry about it. Or it’s an analytic question. Then of course it will be forwarded to just ask. In the context of just ask, just ask is aware of the data that the customer has been indexed. Index means the customer has added them to the JustAsk configuration page and the indexing is then started automatically by JustAsk. So now what could happen that there’s only a single model that matches my query or there might be multiple models that matches my query. If there are multiple models, then we go back to the customer and say, which of these models do you think the answer could be based on? And then we’ll return a chart.
So one of the challenges um that we see is when we talk about knowledge, we like to see knowledge really as a pyramid. There’s global knowledge. Global knowledge. What is the capital of Paris? What is the capital of France? Global knowledge. There’s industry-specific knowledge. This might be particular finance knowledge. um And then there’s the knowledge that’s provided via, in the future, via the SAP Knowledge Graph or the data product.
This is curated knowledge that’s coming with the models and with the data that we are shipping via BDC. So that’s all delivered by SAP. So now let’s move to the top part of the permit. Here we have explicit customer knowledge. Under explicit customer knowledge, what we understand are the models and other semantic enrichment provided by you as partners or by your customers. So in a model, what kind of information do I have in a model? I might know that a particular dimension is a time dimension. You should be able to drill up and down in the time dimension. I might be aware that there is an account dimension that is based on a particular hierarchy. Or I might be aware of the definition of the fiscal definition that we know is different from customer to customer.
So all of this information, course, we as AI, AI is very smart, but AI does not know when your fiscal year starts. And now there two things that can happen. Either that information is being modeled in the model, then of course we can understand it, or it’s not modeled. And what we are seeing very often when working with customers is that customers are asking for the so-called tacit customer knowledge. That means, well, we talk about every day what when what does it means when a team is red? Well, that information, what it means a team is red is of course different from customer to customer. And without this information being made available to JustAsk to our AI, we cannot know what that means. So that’s why our ask for you as partners is to help your customers to collect the tacit customer knowledge and make it explicit so that just ask and other AI tools can actually leverage it.
All right, so talking about the challenge. So if I ask a question, very basic, simple, straightforward question, what is the revenue for company A this year? Well, there’s two levels of challenges or ambiguities that can arise. The first one is that at the model level, right? In your customer landscape, multiple models for different line of business or for even one single line of business, who would potentially answer that question in the sense that they include some revenue data that could be related to sales, to finance, to marketing, whatever. And so that’s the first level of ambiguity that we need to address via best practices. And the second level is ambiguities related to the underlying data model or the content of the data model, assuming the right data model has been queried. Well, then within a given data model, you may have many measures or accounts that can match or answer or be a good match for to answer that question. Imagine you have not many measures related to revenue, well, which one is the best one? Which one is the one that is expected? Is it about sales revenue? Is it about the mission revenue? Is it about total revenue? Well, that needs to be resolved. That type of ambiguity needs to be resolved, right? So the key element of the best practices is to address that challenge. In addition to that, there’s another, one.
If you think of dashboard designs, designers and stories, well, that’s typically what they do. They address that challenge, right? Because they know which model is relevant for a given audience, for a given problem. And they also obviously also know which measures, which facts uh are relevant to those users if they need additional calculations, additional context that can be provided in the story.
he second level of challenge uh is around the availability of data. So sure, customer landscape will have a lot of data, but think of end users. They have the experience of chatGPT, which is trained on public data which always provide an answer, no matter where, if it’s accurate or not, if it’s relevant or not. They will always expect an answer. With JustAsk with joule Insights, that’s not the case, right? It will not invent or hallucinate data. It will provide data and answers based on real data from customers. So you have this limit. Based on what’s available in the customer data model that is not something that can not be overcome. So that’s definitely something also to acknowledge and to take into account. On top of that, you have the processing on the data that is also a limit and that’s more on the software level. So on even on the just ask and NLQ level is what type of question can we process and what type of calculations in particular.
Can we provide? And typically, we’ll just ask, can provide some average calculations or average aggregation. However, ACC and JustAsk, by consequence, will not support that type of calculations on top of blended models. So here you hit a software limit. And again, this needs to be acknowledged and taken into account. And then the third type of challenge is around questions that contain implicit terms. So typically what we see very, very often is questions like that, or portions of questions like that.
Reference Document Below.