Catalogue intelligence
Words customers used that matched nothing, ranked by how often — and what to do with each one.
Every time a customer uses an order-shaped word that matches nothing in your catalogue, it is written down. This screen is that list.
Viewing it is available on Pro and Business. Collecting is not gated — a Starter shop accrues the same vocabulary from day one and sees all of it the day it upgrades.
What is on the list
Words and phrases from order-shaped messages that matched no product and no alias, with:
- How many times each one came up.
- When it was last used.
- Which customers used it.
Ranked by frequency, because the word twelve customers used matters more than the one somebody typed once.
What is deliberately not on it
- Out-of-stock and over-limit lines. Those products matched perfectly well. The problem was stock, not vocabulary.
- Notes. "after 6pm", "leave at the gate", "thanks" — no alias could ever answer those, so they would be permanent noise you could only dismiss.
- Anything from a non-order message. A question about delivery is not a failed product match.
What to do with each one
Teach it. Add it as an alias to the product it meant. Every future customer using that word gets a matched line instead of a question.
Add the product. Sometimes the word is not a synonym — it is something you do not stock and several people have now asked for. That is the most valuable thing on this screen.
Dismiss it. A typo, a competitor's brand, a word that meant nothing.
The reopen rule
There is an asymmetry here, and it is intentional:
- A term you taught that comes back is reopened. You gave it an alias and it still missed, so the alias evidently did not cover the way people actually type it.
- A term you dismissed stays dismissed, however often it recurs. You already made that decision; asking again every week would be nagging.
Reading it over time
The useful signal is not any single word — it is the trend. If the list keeps growing after you have worked through it a few times, the vocabulary gap is wider than the aliases you have added, and it is worth spending twenty minutes adding several per product rather than one.
The clarifying-question count on bot health is the same signal from the other side.