Check each row for a specific condition.

Write a condition with a clear Yes and No boundary

A useful flag answers one question about the supplied text. Does the customer explicitly request a refund? is narrower than Does this customer need help? The first condition names an action and specifies that the request must be explicit. That makes it possible to explain why a row should receive Yes or No.

Include the distinctions that matter in your workflow. A customer asking whether refunds are available has not necessarily requested one. A customer saying Please return the amount charged has requested reimbursement even without using the word refund. Define the meaning you care about rather than depending on a single keyword.

Avoid joining independent conditions with and or or unless you truly want one combined result. Does the message request a refund or mention a damaged item? cannot tell you which fact was present. Two separate flags preserve that distinction and let you filter for either condition or both after combining the exported columns.

Use flags to find specific requests and stated facts

For customer support, flags can identify explicit requests for refunds, replacements, or a human agent. These columns help someone find the messages they need to review. A Yes identifies the stated request; it does not authorize the refund, approve the replacement, or send a response.

For supplier and delivery correspondence, a flag can identify whether a message supplies a tracking number or requests an appointment. Keep the question tied to the evidence actually present. A statement that tracking will be sent later is different from a message that already contains it.

For product feedback, flags can identify a stated regression, a request for export functionality, or supplied reproduction steps. Independent flags can coexist on the same row when you run separate jobs. That is useful when the facts overlap and there is no meaningful single category that should win.

Leave room for uncertain rows

Flag uses separate thresholds for Yes and No. Results in the interval between those thresholds become Needs review. This gives you a way to keep ambiguous rows visible instead of forcing every message into one of the two labels. It is especially useful when messages imply an intention without stating it clearly.

Start by checking known positive and negative examples, including similar wording with different meanings. Please refund my payment, What is your refund policy?, and I no longer want a refund should not be treated as the same request. Negation, quoted messages, and a request made on someone else's behalf can all change the interpretation.

Use thresholds to manage review, not to hide a poorly worded condition. Raising the Yes threshold requires a stronger model signal before a row receives Yes; lowering the No threshold requires a stronger signal before it receives No. Widening the middle interval sends more rows to review for the same results. The underlying signal is not a calibrated guarantee of real-world correctness.

Should you Categorize, Score, or Flag?

The same message can support different questions. Choose the output you need before choosing a tool.

Categorize

Which kind is it? Choose one label from a set, such as Billing or Shipping.

Score

To what degree does it meet the rubric? Assign a defined level, such as 1 through 5 for reproduction detail.

Flag

Does it meet this condition? Return Yes or No, with uncertain results marked for review.

A message about a broken export might be categorized as Technical, scored for the detail of its reproduction steps, and flagged for an explicit refund request. Keep those questions separate so each result remains interpretable.

Try Flag on your spreadsheet

1. Add and preview your data. Open a tool below, then paste from Excel or Google Sheets or upload a CSV. Check the header row and column names.
2. Choose the source columns. Select the fields that contain evidence for your question. Leave out unrelated information and any existing answer column.
3. Define the result. Edit the starting definition and name the new column. Make sure the rules cover the examples and edge cases you expect.
4. Preview, review, and export. Preview a sample before processing the remaining rows. Inspect uncertain results, correct mistakes, then copy or download the results for your spreadsheet.