Customer Buckets
Describing a bucket instead of building it
Generating bucket rules from a sentence, and why you still have to check what it produced.
Last updated September 2, 2026
Rules take a while to build if you are not sure which fields hold what. You can instead describe the group you want — "all people who have purchased in the last year" — and have the rules generated for you.
Writing a good prompt
- Be specific about time. "Recently" means nothing; "in the last 90 days" does.
- Say what to exclude as well as include.
- Use your own words for categories and products — it matches against what you actually have.
- One group per prompt. A sentence describing three audiences produces rules for none of them.
Always read the rules it wrote
IT PRODUCES RULES, NOT JUDGEMENT
The generated rules are a draft. They may read the wrong field, pick the wrong time window, or miss an exclusion you assumed was obvious. Read them, preview the people, and correct them by hand before anything sends.
If it cannot do it
A prompt that produces nothing usually describes something the data does not hold — a fact nobody records, or a distinction that lives in somebody's head. That is worth knowing in itself: if you cannot build the bucket, you cannot automate against it until the data exists.