The AI-assisted modeling workflow

0:05 Join us in this detailed tutorial from our expert trainer who walks you through each step. For an even more hands-on experience, you can download the course files linked below and follow along. Let's get started. We are going to walk through the AI assisted financial modeling workflow. In finance, we often rely on a linear workflow. You know this feeling well. It is the budget season at Bean and Basket. You receive raw reports from the point of sale systems of 50 different coffee shops.

0:28 It is a mess. You spend three hours just formatting dates because half the stores use European formats and the other half use US formats. You spend another two hours standardizing product names because one manager types cappuccino and another types cap. You write complex nested formulas in Excel and drag them down multiple rows. A reference break or a change in price assumption by the CEO means heavy rework. You have to backtrack to the very beginning. Recleaning and repasting data.

0:54 On the contrary, the AI workflow is different. It isn't linear. It is circular. We can visualize this process in four distinct steps namely context loading, structuring, execution, human review. The first step is context loading. In this workflow, we simply upload the raw chaotic files directly into our reasoning engine, typically a tool like chat GPT plus or claude. We will use chat GBT. Over here we have the sales record received from different cities. Let's organize them in one folder for ease of navigation and prepare the files for upload by properly naming each file.

1:41 Upload the files. At this stage, we don't try to fix the data. We're effectively handling the junior analyst the box of receipts and letting them assess the scope of the work. That leads us to the second step, structuring. This is where the magic starts. We do not ask the AI to calculate the total revenue yet. Remember, it is probabilistic. It might make a math error. Instead, we use its reasoning power to clean the mess. We prompt the AI.

2:06 You are an accounts assistant. I've attached the coffee sales records from three cities, namely New York, San Francisco, and Chicago. I want you to create a single Excel file by combining all the records in the attached files. In that file, keep one row for headings and then populate it with the data present in each file. Add a new column to the left title city and add the abbreviations for each city based on the file from which you've extracted the record.

2:28 AI has performed the said task swiftly and here we have a single Excel sheet containing all the records from the three source files. At this stage you can structure the data till the point the output matches your requirements. For instance, every branch uses a different naming convention for cappuccino. Let's structure this data into uniform terminology. Prompt the AI in the item column. Find the words cappuccino, kappa, or cap. Then replace it with cappuccino. You continue to do the same for structuring other columns like date, location or price.

3:17 Additionally, you can use the reasoning engine for data formatting like adjusting capitalization or column placement. At this stage, a prior knowledge of the potential data errors and ambiguities can be very helpful. Once done, you can extract a clean and structured data set as a CSV file or a Python data frame. Now we move to step three, execution. We take the structured data and move it into our execution engine, Microsoft Excel or Python. This is where we calculate the margins or run the variances.

3:41 Because the data is clean, your formulas are simple. Here we've calculated the item wise sale with a sum formula. The spreadsheet does what it was born to do, precise deterministic math. It doesn't have to guess what cap means because the reasoning engine had already solved that ambiguity. And finally, the most critical step, human review. At this stage, review the available data to identify opportunities for deeper analysis. For example, while the sheet is well organized, it can be enhanced by adding a discount column.

4:30 This would enable a clearer linkage between discount campaigns and sales volume, supporting more advanced insights. You provide the extra information, refine the prompt, and the cycle runs again instantly. This workflow does more than just save time. It changes the way you think. We have our engines and we have our workflow. We will look at the specific software from chat GPT to Microsoft Copilot that fits into each stage of this pipeline. One of the biggest hurdles for modern finance teams is navigating the noise.

4:55 There are thousands of AI tools on the market and every week a new startup claims that they can automate your