Sampling and thresholds in reports: when numbers hide people

Abstract orange illustration on a dark background: an eye. Article topic: “Sampling and thresholds in reports: when numbers hide people”
In brief

To protect privacy and control costs, many reports apply thresholds (showing data only above a minimum) or work with samples. In small accounts this can hide whole segments or distort conclusions. Understanding these mechanisms avoids reading the absence of data as the absence of results.

The essentials in a definition

Sampling and thresholds in reports can hide people and distort numbers: when data are large, some tools calculate from a sample; when small, they hide rows to protect privacy. Knowing when this happens, and using raw data, for example in BigQuery or GA4 data, avoids wrong conclusions. The logic of transparency about limitations is close to NIST’s.

Privacy thresholds

To avoid identifying individuals, some platforms do not show segments with few users: for example demographic data, small audiences or low-volume search terms. The numbers appear as “other”, or simply missing.

This mainly affects accounts with little traffic and very detailed analyses.

Sampling

In large or complex queries, some tools analyse only a part of the data and extrapolate. The result is an estimate, not the exact total. Generally there is an indication that sampling occurred. The more you cross and filter, the greater the risk of distortion.

In small accounts the effects are usually less visible, but they exist.

Illustrative diagram with 4 numbered steps joined by an orange line, the last one marked as done. It represents: Reading reports with thresholds and sampling

Reading reports with thresholds and sampling

  1. Check sampling and threshold notices
  2. Compare totals with and without filters
  3. Use longer periods and fewer dimensions
  4. Validate with more than one source

How to recognise and work around

Check reports for notices on thresholds and sampling. Compare report totals with and without filters: if they do not add up, something is hidden. Use longer periods, fewer dimensions or exports, which usually have fewer restrictions. Combine sources: platform, analytics and CRM.

Never conclude that a segment does not exist just because it does not appear.

Large orange quotation marks on a dark background with abstract lines of text, illustrating the quote: “A missing number is not a zero.”
A missing number is not a zero.

Good decision practice

Prefer aggregated analyses with few filters. Validate important conclusions with more than one source. Record limitations in client reports. Avoid decisions on very small, unstable segments. Accept some uncertainty.

For sensitive analyses, use tools that offer access to raw data, with respect for privacy.

Decision table

SituationEffectResponse
Sampling in large reportsApproximate numbersReduce the range or use raw data
Privacy thresholdsHidden rows in small segmentsAggregate segments and periods
High cardinalityValues grouped into “other”Simplify dimensions
Data explorationRow and query limitsExport and analyse elsewhere

Common mistakes to avoid

  • Treating numbers as exact. Check whether there is sampling or hiding.
  • Comparing reports with different methods. Use the same source and definition.
  • Over-segmenting. Small segments disappear or become unstable.

How to measure the result

Record when a report uses sampling or thresholds and compare with raw data in critical analyses.

Limits and exceptions

Raw data also have limitations (consent, gaps) and require skills to analyse. Not all questions need exact data; for order-of-magnitude decisions, sampling may suffice. Define the degree of rigour each decision needs and use the right tool.

Illustrative example

Illustrative example: a team notices a report with many filters shows slightly different numbers on different days. It finds the tool uses sampling. It reduces the range, simplifies filters and starts analysing exported raw data. Numbers become stable, and the team documents the limits.

In practice: checkpoints

Check whether your tools sample data or hide small groups for privacy, and read the notes on each report. Use longer periods or broader segments when numbers are small, and avoid drawing conclusions from thin slices. For key decisions, confirm with the database or an export of raw data, and record the settings used.

Frequently asked questions

Why is data missing in small segments?

Because of privacy and volume thresholds in reports.

Does sampling invalidate analysis?

No, but it makes it an estimate. Validate what is important.

How do I avoid being misled?

By comparing totals and using more than one source.

How do I know if there is sampling?

Tools usually indicate it in the report; check the documentation.

Are privacy thresholds a problem?

They protect people; reduce granularity or aggregate to see the essentials.

Summary to quote

Confirm whether reports use sampling or thresholds, simplify segments and use raw data in critical analyses, documenting limitations.

Sources and recommended reading

Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.

Keep reading

Quick contact

Want to talk about your campaigns?

Leave a short note. I'll reply within one working day.