Today’s financial professionals often find themselves in Excel purgatory — the endless cycle of exporting data from enterprise resource planning (ERP) software, juggling multiple spreadsheets, and repetitive copying and pasting before finally creating pivot tables and charts. But by harnessing the power of Generative AI (GenAI), financial professionals can break this cycle.

Gauthier Vasseur, Founder & CEO of Data Wise Academy, explained during an AFP Member Meet-Up that by integrating Python and GenAI into their data workflow, financial professionals can process data in structured tables and leverage AI-driven automation to reduce errors, improve efficiency and unlock advanced analytics capabilities.

Getting started

Python, once seen as a tool only data scientists could wrangle, is now accessible to finance teams thanks to large language models (LLMs) like ChatGPT, which can generate clean, documented Python code based on natural language prompts. Even without coding expertise, you can automate data transformations, run large-scale analyses and create dynamic visualizations.

To get started, Vasseur recommends the following components, all of which are free and open-source:

• Python: Python is a programming language that is known for its readability and ease of use.
• Code editor: Vasseur recommends Visual Studio Code as it is free and likely already approved in corporate environments that use Microsoft products, but any code editor would work.
• Jupyter Notebook: Jupyter allows users to easily combine Markdown text and executable Python code in an environment called a notebook.
• Access to an LLM: This is needed for code generation. Either internal or external options work, but be sure to follow your organization’s AI policy.
• Data sources in a structured format: This could look like CSV files, database tables or Excel files with plain tables.

Once you have all the components, follow this workflow shared by Vasseur:

1. Extract data in well-structured formats, i.e., CSV files or database tables.
2. Ask an LLM to generate Python code for your specific data processing need.
3. Copy and paste the generated Python code into your preferred application to seamlessly process the data.
4. Feed the cleaned data into visualization tools like Tableau or Power BI.
5. Use the processed data for machine learning and advanced analytics that Excel can't do.

Use case: AI-powered data transformation

What does the above workflow look like in practice? Vasseur demonstrated a use case of how a financial professional could take a large dataset, clean it, process multiple transformations (e.g., currency conversions, text formatting, geolocation calculations) and generate interactive data visualizations.

Challenge

Let’s say you’re working with a massive CSV file containing thousands of real estate records. You need to clean up the data, remove unnecessary columns, perform unit conversions, calculate distances and reformat text fields. Doing this manually in Excel would take hours, if not days. Even with formulas and macros, the process would be slow and tedious.

Solution

Use a simple prompt in an AI-powered assistant like ChatGPT: I have a CSV file on my computer. I want to remove a column, convert the land size from square feet to square meters, calculate distances based on geolocation data, reformat names and adjust pricing by dividing values by 1,000. Can you generate the Python code for this?

Within seconds, it generates the Python script, breaking down each step with clear, readable comments. The code can then be copied and pasted directly into a Jupyter Notebook — a coding environment that works like a smart document where you can execute Python commands — and the code can instantly process thousands of rows of data. The same script can be run on an even larger dataset with the same efficiency.

By giving additional prompts to ChatGPT, you can work with the data any way you want. You can generate interactive 3D charts to visualize financial patterns; map thousands of properties based on location and category; or run automated data profiling to detect missing values, outliers and correlations.

Impact

It is commonly estimated that 70% of the time spent in all data research projects is simply preparing the data for use; in this example, those hours are reduced to seconds. The process saves a lot of time, as tasks that once took hours now take seconds. It’s also scalable because the same code can be used for thousands or millions of rows.

Addressing common concerns about using AI in finance

But don’t LLMs hallucinate?

While AI can sometimes generate incorrect or misleading results, referred to as “hallucinations,” Vasseur explained that Python is a well-established language that LLMs have been extensively trained on, resulting in very few hallucinations of Python code. When you use a step-by-step approach to prompt LLMs for code generation, rather than requesting complex solutions all at once, you greatly reduce the chance of errors.

But couldn’t Python-based data processing be considered shadow IT?

Vasseur shared that Python-based data processing doesn't qualify as shadow IT because the approach maintains data freedom by starting and ending with tables. As a result, the data is never “imprisoned” in propriety formats. Furthermore, Python is already a standard programming language used in many organizations.

Once finance has proven the value of the Python solution, finance can partner with IT to make improvements, implement the logic directly in databases, create a formal IT solution based on the prototype and industrialize the dashboard.

But isn’t there a data privacy risk with using AI?

Vasseur explained that because AI-generated Python code runs locally on a user’s machine, it eliminates the risk of sensitive financial data being exposed to external LLMs. When using tools like ChatGPT, the AI never sees the actual data; it only sees the pattern or structure needed for code generation. You can always anonymize the data by using dummy names or placeholders, keeping sensitive information private.