Notes are observations, not a sales denominator

Repeated mentions of a small fit deserve attention, but they do not establish a high return rate. A CSV of source notes may contain neither units sold nor total orders. Without a suitable denominator, a note count cannot become a return rate. First establish whether each row represents a note, an order, or another unit, and whether one return can appear more than once.

Fitnote currently processes CSV notes with rule-based logic in the browser. Its counts describe expressions detected in the supplied notes. They do not represent all sales or all customers. I recommend using frequency to prioritize a closer reading, not to substitute for it.

Separate size, width, length, and sleeves

“Too small” does not identify the dimension that needs attention. Separate the selected size label from width, garment length, and sleeve fit. A close-fitting sensation is not itself proof of an incorrect measurement. Material and preferred silhouette can affect interpretation, but a note alone cannot establish the cause.

These examples are invented for illustration, not customer records or Fitnote results: “The body fits, but the sleeves are long” supports reviewing sleeve length rather than declaring the whole item oversized. “I sized up, but the chest is still narrow” supports both a size-selection label and a width label. It does not establish that another larger size will solve the problem.

  • Identify the named body area or dimension.
  • Retain the selected variant when it is available.
  • Leave the cause unresolved when the wording is insufficient.

Compare variants and sources on matching terms

Before comparing variants, check size, color, collection period, and source. A selected return reason and a free-text review may answer different questions. Do not silently combine them as equivalent observations. If the source or variant mapping is missing, retain that uncertainty instead of guessing.

Raw counts can make a variant with more input notes look worse. A share within supplied notes may help describe that file, but it is still not a sales-based return rate. Record the counting unit, duplicate treatment, and comparison scope. A file selected specifically for negative comments also limits what the comparison can mean.

Verify product facts before changing copy

Check the actual size chart, measurement method, variant differences, material information, and intended fit. A note saying “it does not stretch” is a reported experience, not verified material specification. Neither source should silently replace the other.

This invented copy template is conditional: “Compare sleeve length with a garment you wear, using the measurement position shown in the size chart.” Use it only if that chart actually includes sleeve measurements and the method. “For a roomier fit, compare the chest measurements across options” also requires verified measurements. The aim is to name a useful comparison, not prescribe an unsupported universal size increase.

  • Does the relevant measurement actually exist?
  • Are measurement methods and variant details verified?
  • Does the wording promise an unconfirmed product property?

Separate a revision from a causal claim

If notes change after an edit, the edit has not automatically caused fewer returns. Variant mix, collection period, or the way notes are requested may also have changed. Even linked sales and return data require a defensible comparison. Notes alone cannot establish an effect size or causality.

Keep the original observation, variant and source, verified product fact, proposed wording, and unresolved question separate. Suggested copy is a checklist or template for human review, not an AI-generated answer. A useful revision gives shoppers clearer comparison information without pretending to make their decision for them.