Use case 1: SMART problem statements assistant

- [Instructor] Let's consider a real-world example. Imagine the following case study. Let's say a retailer is facing increasing pressure from competitive market dynamics, particularly fluctuating customer demand and rising supplier prices. They also need to become more profitable to stay relevant, and that's why they hired us, the data scientists, and gave us access to their ERP system to help them find areas for improvement. Now, how do we go from this very broad situational description into an actionable SMART problem statement? And this is where our first use case will help us, the SMART problem statement assistant. So let's dive right in. First, we'll start off in the course repo and go to branch 02_02. You can see all the resources for this example here. Let's open the first link to the GPT. I'm now here in ChatGPT and accessing the SMART Problem GPT. To give you a quick look under the hood, let's see what the instructions for this GPT look like. Quick heads up, you won't be able to see these settings like me here because I'm the owner of this GPT. You can use it right away, but if you want to see the detailed instructions, check out the config file on the GitHub and the branch that belongs to this video, just as we covered in the ChatGPT setup video. So what this GPT is doing is that it prompts the LLM to behave like a senior data analyst at McKinsey with 10 years of experience. Why? Well, because just by putting it into this definition, it's more likely we get expert-level outputs from it and not just average answers. The goal here is to help the user turn a loose situation description into a SMART problem statement. And we also point to the definition of SMART below in the triple backticks. Then we want to have it perform the following tasks. It should identify areas where the user's problem statement does not yet meet SMART criteria. Then it should ask targeted questions to get closer to the SMART problem statement. And it should also ask questions one by one, waiting for the user's answers. And finally, if it has enough relevant information, it should suggest a SMART problem statement and review it critically. Now, we also provide some details here. In particular, what SMART really stands for. There might be different definitions of SMART problem statements or SMART criteria. So we provide our own definition here. And also we provide some examples of what a SMART problem statement looks like and what a non-SMART problem statement looks like, which will help the GPT to figure out what we actually want from it. And finally, it should start by asking the user for their problem statement. So let's see it in action. What we do is we press the Start button first. And what we see here is that we get this little greeting message, asking us to provide the problem description that we have. This is where we could now paste in the situation description from our case study. And to speed things up, I've prepared a sample chat that we can follow. And again, you will find the link to this chat in the GitHub repository. Now, I just pasted the text from the case study here, and you can see that after we entered that, we get this little greeting here. You can see that it says, thanks for sharing, but it asks us to be more specific. In particular, it's asking us what specific outcome is the retailer hoping to achieve, and are there any particular areas of the business where we want to put our focus first, and we reply with, it could be both, more sales or less cost. And here it's asking us a follow-up question: If we want to double down on one of these two aspects, whether it's increasing sales or reducing costs. And we say, we want to go for a combined approach. And from there, it continues. How can we make this more measurable? In our case, measurable could mean that we want to say the profit margin must be back to 3% because currently maybe it just sits at 1.5%, and now, the ChatGPT responds, great. What we need is now an actionable criteria. Since we have access to the ERP system, what kind of data are we working with? Is it sales data, cost data, or customer behavior data? And we say, because that's the data type we are working with mostly here, sales data and also some supplier data as well. And from there it asks, okay, how is this relevant? Does this profit margin target align with a broader business goal? And we say, yes, it's about aiming to be more competitive. And finally, we get asked about a timeframe, and we say 12 months., Bear in mind that sometimes we might not have an exact answer to all of these questions, which is totally okay. Sometimes you can't fulfill all the SMART criteria, but at least now you are aware with the blind spots that you have and you can fill them in as needed or go back to your stakeholder and discuss about these things. And now that we checked off all the SMART criteria, we get a final SMART problem statement back, which in this case, reads as follows. What strategies can the retailer implement to increase their profit margin from 1.5% to 3% within the next 12 months by leveraging insights from sales and supplier data in alignment with a broader goal of improving competitiveness in the market? And this is a super actionable statement to work with. We could now save this problem statement somewhere so we can come back to it later as we need it. And it's also the statement we're going to use for the next videos.