Skip to content
PMcademy
Template

The signal log

One row per quote. A structured way to hold raw feedback without losing the exact words.

When you reach for it

Any time a pile of reviews, tickets, or call notes needs to become something you can sort and defend.

  1. 1

    Source

    Keeps every claim traceable back to where it came from, so nobody has to trust your summary.

    What it asks

    Where did this come from? Channel, date, and who said it if you know.

    Filled in

    App Store review, 14 Aug, iOS, 3 stars

    How it goes wrong

    Skipping the date. Six months later nobody can tell if a complaint is about the current release or one three versions ago.

  2. 2

    Quote, verbatim

    The exact words carry information a paraphrase throws away: tone, specificity, whether they blame themselves or the app.

    What it asks

    Copy the sentence exactly. Do not summarise, do not fix the grammar.

    Filled in

    "it keeps saying my bank isnt supported but its literally chase, one of the biggest banks in the country"

    How it goes wrong

    Paraphrasing it into "user reports bank search issue," which is exactly what a model does when asked to summarise instead of extract. The mention of a specific well-known bank is lost, and with it the fact that this is a search index problem, not a coverage problem.

  3. 3

    Label

    Groups the same underlying problem said different ways so you can count it.

    What it asks

    What's the underlying problem, in two or three words? Use a short fixed vocabulary, don't invent a new label per row.

    Filled in

    bank-search-fuzzy-match

    How it goes wrong

    A new label for every row because every sentence is worded differently. Cap the vocabulary at 8 to 10 labels before reading a single row, and force each one into the closest fit.

  4. 4

    Who it's from

    Not every user is equally informative. A churned power user and a day one signup are different signals.

    What it asks

    New user, existing user, or churned. Guess if you have to, and say you guessed.

    Filled in

    New user, signed up but never linked a bank (guessed from context: "I just downloaded this")

    How it goes wrong

    Treating every row as equally weighted when a churned power user complaining is worth more than ten day one signups hitting a rough edge they'll happily forgive.

  5. 5

    Severity

    Not every complaint is equally costly, so rank before counting.

    What it asks

    On a 1 to 5 scale, how much does this block the person from getting value, not how angry they sound.

    Filled in

    5, they cannot complete signup at all

    How it goes wrong

    Scoring by how angry the review sounds rather than how blocking the problem is. A calm review about a total blocker matters more than a furious one about a font size.

Take the whole thing

Paste into your doc
# Signal log

One row per distinct quote. Never edit the words.

| Date | Source | Quote (verbatim) | Label | User type | Severity 1-5 |
|---|---|---|---|---|---|
|  |  |  |  |  |  |

Label vocabulary (fix this before you start, 8 to 10 max):
1.
2.
3.
4.

Rules:
- One row per quote, not per person.
- Copy the quote exactly, misspellings and all.
- Severity measures how much it blocks value, not how angry it sounds.
- Guess user type if you must, but write "guessed" next to it.
More templates