Explore Less Data with Sampling
Sampling means examining a smaller group before analyzing everyone.
What Is Sampling?
Think of checking a few boxes from a delivery to understand what arrived. You can learn something useful without opening every box, although a small sample may miss an unusual item.
In zinzu, sampling selects EventOwners, rather than picking scattered individual records.
How It Works
Suppose your selected data contains 100,000 customers.
| Sampling setting | Approximate customers selected |
|---|---|
| 100% | All eligible customers. |
| 50% | 50,000 |
| 10% | 10,000 |
These are approximate counts, not exact quotas.
For selected owners, zinzu retains their eligible records across the selected sources and time range. It does not keep only 10% of each customer’s actions.
Example: A selected customer has Product Viewed → Added to Cart → Payment Completed. Those records remain available together, so sampling does not break the sequence by selecting only the payment.
Selection is consistent for the same owner identifier. Keeping the same settings and owner population makes repeated exploration more comparable.
Set Sampling
- Build your query.
- Click Submit Query.
- Find sampling percent in the submission dialog.
- Choose a value from 1% to 100%.
- Confirm and run the query.
When to Use It
Explore a large dataset:Try a smaller percentage while adjusting the events you want to find.
Refine a question:Check whether your query finds useful behavior before running against all eligible owners.
Validate important results:Use 100% when you need all eligible owners, particularly for rare behavior.
Important Notes
- Smaller samples can miss rare patterns.
- Counts describe the analyzed sample; do not present them as full-population totals.
- 10% sampling does not guarantee a 90% cost reduction. Reading source data and other query work can still incur costs.
- 100% sampling includes all owners eligible after other exclusions; it does not restore owners excluded by data skew protection.