Zovda
Every analytics stack has a step called "remove bots" and almost none of them say where the line was drawn. That line decides both the quality of the data and which real users got deleted.
The number after filtering looks cleaner and is not necessarily more true.
Three signals and their costs
- Posting rate: catches automation, also catches the most enthusiastic fans
- Account age and history: catches new accounts, also catches people who just joined
- Text similarity: catches copy-paste campaigns, also catches a slogan everyone repeats
Each signal is cheap and each one has a false positive that lands on a real person.
False positives are users you cannot see
A filtered account does not complain. It simply disappears from the data, and if it belonged to a small audience, that audience loses representation without anyone noticing.
Report the count of removed items and the reason for each class, not one aggregate.
Keep the excluded rows
Store what was removed and why, in the same export, flagged. A team that needs the unfiltered view for a legal question or a brand safety review should not have to re-collect.
The flag also lets you reverse the decision later, when the threshold turns out to be too aggressive.
Report before and after
Every metric that depends on the filter should be reported twice: with and without. If the two numbers tell different stories, the filter is doing analysis, and that analysis should be visible.
Zovda builds these collections through Datuva, where removed items stay in the export with a reason. The social media data collection shows the flag fields.
Tell us what counts as a real user for you. We reply within one business day.