A seasonal drop or fraud: how to tell them apart
Contents
Share
The checked guides never name a season as the cause of a conversion drop. Fraud is not assigned from the single word “it fell” either. A drop you can see comes from the date the report uses, from a mix of campaigns the guide forbids comparing, and from a bounce threshold you are allowed to change.
The order “not fraud first” is covered in click fraud or a bad ad. This is no longer the fork about the ad. It is three measurable causes that look like a slump and are not a season.
Two dates for one action
The report wizard attributes a conversion to the click date. The pay-per-conversion report attributes it to the conversion date. In Yandex’s example a click on 3 July and a conversion and charge on 5 July land on different days. A week looks empty in one report and not in the other. That is not a season and not an attack. How the calendar makes leads disappear is covered in clicks without leads.
Before you sign the drop to a month, open both reports on the same dating rule. If the “slump” exists in only one, the cause is the date rule. Comparing search and networks on that same calendar is covered in how to compare search and network conversion.
A mix you must not compare
Clicks and conversions of display campaigns on search are not compared with performance campaigns: the demand is different. A combined line that mixes those types falls or rises without a diagnosis. Split the types. A drop in the mix is not proof that the networks, or a season, took the conversion.
A bounce is not a conversion and not fraud
A visit in Metrica is a bounce only if three conditions hold at once: no more than one page view, a duration shorter than the bounce calculation time, and no service event “non-bounce”. The default threshold is 15 seconds, and it can be replaced. The browser sets the non-bounce event after that time. Because of clock skew and send delay, a visit shorter than 15 seconds may not be a bounce.
If what “fell” is the bounce rate, first check whether the threshold was changed. That is a counter setting, not a calendar of demand and not the invalid-traffic column. Fifteen seconds is not a fraud threshold. What Direct removes from spend is covered in invalid clicks. A bounce is not carried there.
A real person with no interest in the offer is not removed by Direct’s filter. A conversion drop among such visits stays in spend. Calling it fraud means expecting the filter to take what the guide deliberately leaves.
Under ten achievements a week the strategy is under-trained
A strategy needs at least 10 achievements of the goal a week. If there are fewer, the guide suggests a goal higher in the funnel, another budget or another conversion price. A drop that takes the week under that line is the named learning problem, not a season. Ten is the guide’s line. The guide does not name a month in which “it is always like this”.
The documents of the on-site check describe the decision on a visit and a pass-through if the script did not get an answer. They do not set a way to measure an effect on ad spend before and after. The check does not label a conversion drop as a season, and it does not count money.
While the word “season” is not in the facts, the check is this:
Take the drop in two reports. If it exists in only one, the click date or the conversion date is the cause.
Split display campaigns on search from performance campaigns. Do not read the combined line.
If the bounce rate moved, check the threshold. The default is 15 seconds, and that is not fraud.
Read a week under 10 goal achievements as an under-trained strategy, not as a season.
Do not carry a real refusal to buy into the invalid-traffic column.