Do you think of tools such as OpenAI’s Codex or Anthropic’s Claude Code as developer tools, built only for writing software? Not the case. A recent project at SmarterX shows how these tools can be repurposed for one of the most common (and tedious) marketing tasks: making sense of messy data.
The Problem
The goal was to understand how a specific piece of content connected to revenue. The data existed, but the answer wasn’t sitting neatly in a single field. It was buried inside a sprawling export: 144,000 rows and 1,000 columns. The file was so large that simply opening it in a spreadsheet program crashed the computer trying to load it.
The Approach
Rather than manually wrestling with pivot tables or dropping a static file into a chatbot for a one-off summary, the entire dataset went to Codex, used like an analyst sitting alongside the work. Using a fully anonymized export, Codex was tasked with:
- Inspecting the data and identifying which fields looked revenue-related versus attribution-related
- Flagging fields that were too noisy or duplicative to trust at face value
- Running sanity checks against smaller cohorts to validate its approach before scaling up
- Narrowing 1,000 columns down to a focused, relevant set
The result was a clear path toward modeling revenue attribution, without writing a single formula or manually sorting through columns by hand.
Why This Matters for Marketers
The value here isn’t that Codex wrote code. It’s that the tool could be handed a goal, “find what’s connected to revenue,” rather than a list of manual steps, and the agent ran its own multi-step investigation. It identified its own next steps, corrected its own errors, and kept going until the analysis was complete.
That’s a meaningfully different way of working than typing questions into a chatbot one at a time and waiting for single responses. It’s closer to delegating a project to a capable analyst than prompting a search engine.
Marketers don’t need to be developers to benefit from this. Any team sitting on a messy CRM export, a tangled campaign performance report, or an attribution dataset nobody has had time to dig into can put agentic tools like Codex or Claude Code to work on that investigative process. The starting point isn’t a perfectly clean spreadsheet or a precise formula; it’s a clear objective and a willingness to let the agent figure out the path.
To hear the full use case explained and more on what’s happening with AI, check out Episode 222 of The Artificial Intelligence Show podcast.
