Why does the predicted industry look wrong, and how can I improve it?

Industry Prediction is a model, and a small share of predictions will not match how you would classify the business. When one looks off, the fix is usually in the inputs.

Check what the prediction was based on. The reasoning field says what the model found. If it mentions a website or other online sources, it classified from that site’s content. If it mentions only the name and address, it had no sources to work from and classified from the name, which is where most surprising results come from.

Provide the website. Including the business website on the request is the single most effective improvement - once Baselayer verifies that website is truly associated with the business. A business classified from a live website with clear content scores high and lands on the right code the large majority of the time. If the applicant gave you a website, send it.

Provide trade names. If the legal name is generic but the business operates under a descriptive DBA, send the DBA in alternative names. The model considers all names provided. This is very common for businesses in the hospitality industry.

Look at the accuracy score. A wrong prediction with a low score (< 0.75) is working as intended: it is telling you to review. A wrong prediction with a high score is worth sharing with your Baselayer representative, because it may point to a website mismatch or a data issue on the business.

Consider the level. Some businesses genuinely sit between two six-digit codes, for example a spa that also does retail. The four-digit industry group is often right even when the six-digit choice is debatable, and the keywords will show both activities.

If you believe a prediction is wrong after providing the website, send us the search ID or prediction request ID with the code you expected. We review these individually and use them to improve the model.