Products Solutions Templates Pricing Resources

Data sampling: when your reports use estimates instead of actual numbers

Home / Everything About / Everything About Analytics / Data sampling: when your reports use estimates instead of actual numbers

You run a custom report in your analytics tool. It says "This report is based on 100,000 sessions. Unsampled data: 1,250,000 sessions." Your report is using estimates from 8% of your actual data.

Data sampling happens when analytics tools can't process all your data so they estimate instead. The estimate is often accurate enough, but it's still not your real numbers.

What is data sampling and why tools use it

The problem: too much data to process

Analytics tools process millions of events per second across millions of websites. Processing every single data point in real time isn't practical. When traffic gets high enough, tools switch from counting everything to sampling a subset and extrapolating.

How sampling works

The tool looks at 10% of your traffic. It calculates metrics from that 10%. Then it multiplies the results by 10 to estimate total metrics. If the sample shows 5 conversions, the tool estimates 50 total conversions. It's a mathematical estimate, not your actual numbers.

Sampling rate varies by tool and traffic level

Some tools start sampling at 100K sessions per month. Others at 1M. Some free tiers sample more aggressively than paid tiers. The higher your traffic, the more aggressive the sampling.

Sampling is transparent (sometimes)

Good tools tell you when they're sampling. A note appears saying "This data is sampled." Bad tools don't disclose it. You think you're looking at actual numbers but you're really looking at estimates.

When data sampling happens

In Google Analytics

Google Analytics free tier samples when you create custom reports that exceed their processing limits (usually around 1M events). Premium tiers sample at much higher thresholds. If your report includes a lot of data, sampling is likely.

In real-time reports

Real-time data is almost always sampled because showing true real-time data for every event would be computationally impossible. Real-time dashboards show estimates updated frequently, not exact numbers.

In complex custom reports

The more dimensions and metrics you include in a report, the more likely sampling becomes. A simple "traffic by source" report might be unsampled. A detailed report breaking down 10 dimensions with custom metrics might be heavily sampled.

The impact of data sampling

Estimates are usually close to actual numbers

Sampling algorithms are sophisticated. A 10% sample extrapolated to 100% is often accurate within a few percent. For big-picture decisions, the estimates are good enough.

Sampling breaks down on small numbers

When you sample 10% and look for conversions from a specific traffic source, you might see zero in the sample even though you had a few in reality. The estimate becomes "zero conversions" when there really were two or three. Small numbers get distorted.

Comparisons become less reliable

Comparing conversion rates between two traffic sources when data is sampled is dangerous. You might conclude Source A performs better than Source B when the difference is just sampling error.

Trend analysis becomes less accurate

Tracking whether conversions are trending up or down when data is sampled adds noise. Is the decline real or just sampling variation? You can't be sure.

How to work with sampled data

Get unsampled data if your tool offers it

Premium analytics tiers usually offer unsampled data. Google Analytics 360 provides unsampled reports. Most paid analytics tools let you export unsampled data. If you're making important decisions, upgrade to get real numbers.

Understand the sampling rate

When you see a sampled report, look for the sampling percentage. "Based on 100,000 of 500,000 sessions" means 20% sampling. The smaller the percentage, the less reliable the numbers. 50%+ sampling is reasonably reliable. 5% sampling is not.

Use segments to reduce data volume

Sampled reports happen because you're trying to process too much data. Use a date range or segment to reduce the data set. Report on the last 7 days instead of 90 days. Report on one traffic source instead of all sources. Smaller data sets are less likely to be sampled.

Export data for offline analysis

If your tool allows exporting unsampled data, do it. Export to a spreadsheet or database and do your analysis there. This gives you real numbers instead of estimates.

Build automated reports with lower complexity

Instead of creating one complex report with 10 dimensions, create multiple simple reports. Each simple report is less likely to be sampled. Together they give you unsampled insight into your data.

Can you avoid data sampling

Data sampling is not always avoidable

If your traffic is very high and you want real-time reporting, some level of sampling might be unavoidable. You can minimize it by keeping reports simple and focused.

Premium tools offer higher thresholds

Paid analytics tools have much higher sampling thresholds than free versions. Google Analytics 360, Adobe Analytics, and other premium platforms can handle much higher data volumes without sampling.

Server-side analytics avoid sampling

If you track conversions server-side in your own database, you control the data and never sample. Your conversion data is exact. This approach requires more technical setup but gives you true numbers.