Question

A marketing analyst writes a prompt asking a foundation model to extract product names, prices, and discount percentages from unstructured promotional emails and return them as a table. Using a plain instruction with no examples, the model produces tables with inconsistent column ordering and occasionally misses the discount field. The analyst cannot change the underlying model and needs a quick, low-cost fix within the prompt itself. What should the analyst try first?

Accepted Answer

Add a few worked examples showing the exact input email and the correctly formatted output table directly in the prompt

Answer

Fine-tune the foundation model on thousands of labeled email-to-table pairs

Answer

Connect the model to a RAG pipeline that retrieves past extraction results

Answer

Raise the model temperature so it explores more formatting variations