Exploring AI Workflows Part 1: Building a Simple Campaign Budget Allocation Tool with ChatGPT o1-preview

Wednesday, Sep 18 - 04:00 - 07:00

Today, I started on a project to evaluate AI workflows for development efficiency and, hopefully, generate some useful ideas along the way. My focus was on utilizing OpenAI's new "ChatGPT o1-preview" model, and I was genuinely impressed by its reasoning capabilities and its ability to correct errors as they arose. I also tested out the new Cursor text editor/IDE, which was an impressive experience after 3 hours of use.

Starting with a bare-bones Remix.run application, we integrated OpenAI's GPT-4 and Chart.js to lay the groundwork for a simple campaign budget allocation tool as first suggested by my colleague Jean-Daniel Germain . In the start of this project, I prioritized functionality over aesthetics first—a role reversal from the old days when Ludwig Magnusson handled functionality and I took care of the design.

The brainstorming process was a collaborative back-and-forth between me and ChatGPT o1-preview. Together, we designed the application's structure and defined key components like user input forms and media channel data. To ensure the tool provides realistic and valuable recommendations, we incorporated publicly available data sources to inform the media channels and their parameters based on industry standards.

A thing to note here is that I had to force the model to use its available data here as it from the beginning didn’t want to share its knowledge with me.

Me > “This is a bit concerning, I was hoping you could help out with the parameters. I would at least like you to give me a "to your absolute best knowledge" example, source all the data you have available, looking at how other models have been implemented, open source data that you are trained on and so on to come up with an as close as possible model that you can. Re-implement the code using that.”

ChatGPT > [finally gives me an answer based on sources from Nielsen, eMarketer, Statista, Advertising Research Foundation]

One of the standout moments was when the AI recommended integrating machine learning libraries and mathematical functions to enhance the allocation process. I am not well-versed in ML, so this was a nice surprise. Specifically, it suggested using the Nelder-Mead algorithm from the fmin library to optimize budget allocations across different media channels. The AI provided sample code, which I integrated into the project. Throughout this process, I identified errors and areas for refinement, and the AI offered good solutions that greatly improved the application's performance after a bit of back and forth.

By iteratively refining the code with the AI's assistance, we established a good foundation for the simple campaign budget allocation tool. This tool now combines (somewhat) data-driven insights with advanced optimization techniques, aiming to provide valuable support for media planners and marketers.

My main goal with this project is to assess how AI can streamline development workflows and contribute to efficiency. The collaboration with ChatGPT o1-preview has been great, and I'm excited about the progress made today. I look forward to sharing more updates as the project evolves and exploring how AI can continue to enhance the development process.

#AI #MachineLearning #OpenAI #ChatGPT #CampaignBudgetAllocation #RemixRun #ChartJS #Mediatool Mediatool #SoftwareDevelopment #MarketingTech #AIWorkflow