Does FouAnalytics Sample Data?

One issue that comes up when verifying ad impressions is the sheer size of the data. When we analyze data for brands with on-site measurement tags, there's usually not a massive amount of data. But brands spending millions of dollars on advertising generate millions and billions of impressions. The corresponding data about every single impression in an ad campaign can be massive to store and analyze and can cause reports to load slowly or never load at all. In some cases, you can analyze a subset of that data to draw conclusions about the larger data set.

Size of Dataset

The size of the dataset is an important thing to consider before sampling the data. If you have a small dataset (such as a website for a small business), sampling doesn't provide a lot of benefit and can lead to inaccurate conclusions due to a couple of outliers. But if you have a large, consistent dataset such as below, sampling makes sense.

Size of Sample

The size of the sample is also important. If you sample only 1 in 10 impressions, how sure can you be that the other 9 aren't fraud? Legacy verification platforms don't measure every impression and in many cases, that's because their tags are blocked by bots. This creates a small sample that isn't representative leading to incorrect conclusions about the entire dataset.

You can see this for yourself if you look at the number of impressions that were measured and compare that to the number of impressions for which you were charged for verification services. Even if they only measured 20% of the impressions, they may still claim 100% were fraud-free.

Representative Sample

FouAnalytics measures every impression, but for convenience a sample is loaded in the browser so you don't have to wait to load millions of datapoints. This is why you always see 2 donut charts in the main dashboard - the "historic total for the date range" represents the entire dataset while the "live data loaded into the browser" is the sample. The sample is used to populate the grids of supporting data. You can compare these two donut charts to understand if your sample is representative of the entire dataset. If the relative proportion of the colors is about the same, then you have a representative sample. If you need to get more data to analyze you can "fetch" more from the server. You can't fetch more data from legacy platforms where they didn't measure it in the first place.

This representative sample allows you to explore the data and reliably find the worst offenders contributing to most of the fraud.

Does FouAnalytics Sample Data?

FouAnalytics does not sample the data as a default, but we can when it makes sense. In the example above showing 170 billion impressions the data can be sampled because it's a large dataset that's extremely consistent. This can save costs while maintaining the ability to find and remove fraud. And even without complete measurement, FouAnalytics still works.

Ask your legacy verification partner if and how they sample data. Sampling is okay if you can still get reliable, accurate conclusions from the data, but if sampling prevents you from finding and eliminating the fraud, then it's a problem.

If you want to compare FouAnalytics to your legacy verification company, let me know and I can get you set up with a trial so you can "see fou yourself". It'll only take a few minutes to add the tag to your DSP or Adserver so you can see what legacy detection misses.