Use case 2: Issue tree builder assistant

- [Instructor] Next up is issue trees. How can generative AI help us to create these? We'll start again in our repo. Select Branch 02_03 from the menu here. And then you can see all the resources related to this use case. Click the first link to open the Issue Tree GPT. Now, let's see what this Issue Tree GPT is doing. If we take a look at the instructions here, you can see it's, again, prompted to behave as a senior McKinsey consultant, just because that's an area where these trees are really familiar, and that's an area we want to extract the knowledge from from the LLM. The goal here is to guide the user in breaking down a SMART problem statement into manageable non-overlapping tasks using a MECE, mutually exclusive collectively exhaustive, issue tree. We emphasize that adherence to the MECE principle is extremely important. Now, the steps of this process. First, we request the user's SMART problem statement. Then the GPT should provide reasonable options for the first split, and then we provide different split types, which are further explained in the instructions. Then we let the user pick the preferred option. And then we add the selected option to the tree. This loop should be iteratively followed until the user says that the tree is complete. And once the user approves the overall tree, we want to have a Markdown version of the tree and also create a PowerPoint file representing the tree as a hierarchical list. Additionally, I specified that the AI should be super brief here in its language, what I call Spartan style, meaning a super-short output format. And you can see that in the definitions, we further explain what different splitting criteria are available and how they actually work in practice. There's also one example here of how a issue tree could look like, given a SMART problem statement. Now, let's see this in action. When we click Start, we get an initial greeting message. We can now paste in our SMART problem statement from the previous video here. And you can see that this GPT provides us with a few splitting options. Again, for the sake of brevity, I have prepared a sample chat for you. You can find the link to this chat in the course repo. In this case, I choose option four, which breaks down the tree using opposite words at the first level. Internal factors within the company versus external factors. The GPT then applied this structure and asked, "Which node do you want to break down next?" I selected internal factors. Now it provides further split options. The algebraic structure to break it down to revenue and cost, or a process structure to break it further down into key internal processes, like operations, sales, marketing, et cetera. Or to use a conceptual framework, for example, a framework like the value chain or the four Ps. Or using opposite words, we could divide it even further into efficiencies versus opportunities. I choose algebraic structure, revenue versus cost, because I think it's straightforward and logical. Now we have a second level in our tree. Next, I choose to explore costs even further. And again, the GPT provides some further splitting options. Now I decided to use a process structure to examine cost-related processes, like procurement, operations and logistics. At this point, our tree has already three levels, and it's becoming increasingly detailed. And then I continued this whole process until I built a really detailed issue tree. For example, in the procurement division, our tree suggests different approaches, negotiating with suppliers, selecting different suppliers, renegotiating existing contracts, optimizing order management, improving order placement. And that is something that I found particularly exciting because it enables data-driven approaches, such as demand forecasting, which would help us to reduce waste from unsold inventory, preventing unnecessary cost. So at this point, you can continue exploring other branches of the tree or finalize the structure. In this case, I said, "I'm done with the tree," and now the GPT delivers the entire tree in two formats. It gives me the whole breakdown of the tree that I have just created, and also creates me a little PowerPoint file, which I can then download. If I open this PowerPoint file, you will see that it doesn't really look pretty, but that doesn't matter because thanks to that hierarchical list, I can easily create it into a so-called SMART art visual. So if I select this option here and turn it into a hierarchical diagram, it looks like this. And if I now select the Horizontal Multi-Level Hierarchy visual, it looks even better and really comes out as a nice visual representation of an issue tree. Now, using generative AI for issue trees really enables us to systematically break down problems into clear MECE-aligned categories, and also helps us to iteratively refine our tree step by step. We can quickly generate structured outputs in Markdown, and also PowerPoint, which helps us to visualize the results for better stakeholder communication. This method ensures that we are exploring the right areas, while keeping our analysis structured and aligned with our problem statement.