This is exactly the kind of problem that makes people swear off data analysis for good. The user here has a straightforward goal, build a geographic demographics report, but their database is a mess, and traditional spreadsheet tools offer no real help. We think the real story is not about the missing ZIP codes. It is about the gap between what users need to accomplish and what legacy tools actually deliver.
The user describes a common nightmare: only 20% of customers have a ZIP code in the dedicated field, while 60% have address lines that mix street, city, state, and ZIP in inconsistent formats. Commas appear or disappear. City names shift between all caps and proper case. For someone "pretty rusty with Excel," the usual advice, write nested formulas, build helper columns, pray for consistency, is a nonstarter. It would take forever, and even then, one rogue comma could break everything. The frustration here is legitimate, and it points to a deeper problem: spreadsheets were designed for clean, tabular data. They were not designed to understand the messiness of real-world information.
An AI-native spreadsheet changes the equation. Instead of asking the user to normalize the data manually, it can interpret the patterns in the address strings. It can recognize that "123 Main St, Portland, OR 97201" and "123 MAIN ST PORTLAND OREGON 97201" both describe the same location. It can extract the ZIP, the city, and the state without requiring the user to write a single formula. The tool does the heavy lifting, and the user gets to focus on the report itself. That is the practical shift: the user's job moves from data janitor to data analyst.
For this user, the path forward is not about learning more Excel tricks. It is about trying a tool that treats messy data as a starting point, not a dead end. They can start small, paste a sample of those address strings into an AI-native spreadsheet and see what it extracts. If the results are clean, they have their geographic report in minutes, not hours. If the tool needs guidance, they can correct a few examples, and the model adapts. There is no reason to accept a 20% success rate when the technology to handle the other 80% already exists. The only question is whether they are ready to explore it.