Why large volumes can be bad news
The increase in clicks and sign-ups looks convincing, but in itself does not confirm quality. Before scaling, it is useful to check whether the behavior of the audience has changed along with the volume.
Twelve signals to be checked
- A sharp jump in traffic for no obvious reason.
- Unnaturally the same intervals between events.
- Lots of registrations with almost zero further action.
- A strong skew to one type of device that does not match the source.
- Repetitive technical parameters in an unusual volume.
- Very high CTR and extremely low post-click CR.
- A sharp drop in approval rate after scaling.
- A single SubID creates almost all of the contentious activities.
- The GEO in the report does not match the campaign's stated setup.
- Behavior changes dramatically after a change in traffic provider.
- Time from click to event looks unnaturally the same.
- The cost of “raw” FTD is improving, and the confirmed one is deteriorating.
Signal doesn't equal charge
Each item is a reason to check the data, not to declare the traffic fraudulent. For example, the same devices may appear due to the features of the advertising platform, and a spike in events due to data transfer delay.
How to check by layers
| Layer | Example of verification |
|---|---|
| Source | whether the supplier or placement has changed |
| GEO | Are the countries mixed in the same campaign? |
| The device. | Is there an abnormal skew |
| Time. | Are the intervals between the click and the event normal? |
| Funnel | At what point does the strangeness appear? |
Why Scaling Increases the Problem
If the original sample contains a weak segment, an increase in the budget may not “overclock” it, but make the share of bad traffic more. Therefore, after each significant increase in volume, it is useful to re-compare approval rate, cost per confirmed FTD and SubID behavior.
What to keep for parsing
Keep clear campaign identifiers and change dates. Without this in a week’s time, it will be difficult to know exactly which source worked before the race and what has changed since.
Rules of admissibility of sources should be confirmed before launching on the page trafficker and in correspondence with the manager.
Compare it to your own standard.
There is no universal “right” time to FTD or confirmation percentage. The anomaly is determined relative to your historical behavior for a similar source and GEO. Therefore, before scaling, it is useful to have a baseline for several stable periods.
What to do in a suspicious segment
Do not stop the entire source automatically. Isolate a specific SubID, placement, or vendor, reduce the volume, and collect data for verification. This allows you to save the working part of the campaign and not mix it with the controversial.
Quality and advertising promise
Sometimes “bad traffic” is created by advertising. If creativity promises one thing and the user path looks different, people may enroll en masse out of curiosity and not move on. Therefore, the quality analysis should include not only the technical features, but also the content of the announcement.
After correction
See not only whether the approval rate has recovered, but also whether the confirmed FTD has risen too much. The goal is not to get the perfect confirmation percentage at all costs, but to find a sustainable economy.
Add control before the vendor launches
If you’re buying traffic from an external source, ask for a transparent description of the audience’s origins first and run a limited test. Don’t scale just because the first FTDs are cheap. Wait for at least basic quality assurance.
Save the raw data.
If the dispute appears a few weeks later, the aggregated dashboard may already change. Exporting by SubID and dates helps to restore the picture at the time of launch and understand when exactly the deviation began.
Red flag after creative change
If quality plummeted immediately after a new post, check to see if it’s appealing to a wider but less relevant audience. Sometimes the problem is not in the purchase, but in too “clickable” promise.
