Detect huge analytics hits within some time range

[AbhishekDhami]

In the Adobe analytics report, we observed that for some mobile app users, we are seeing huge spike in the Video Views numbers. Its like within 1 minute seeing 900+ video views. It was random and happening very few times in a month. It was not bot activity but due to bug in the the mobile app.

I want to create a segment where I can detect such anomaly. I want to filter Visits/Visitor where within one minute, adobe report shows 300+ video views. Is it possible to create segment based on time range/period? 

[Jennifer_Dungan]

While you can create segments to try and compensate for this, I don't know if Adobe can support a rule such as "over X hits within Y minutes"... because unfortunately, segments are based on Hit, Visit or Visitor logic... and you don't want to exclude all traffic from a visit or visitor.

Also, the time boxing in a segment is more about hits that happen between certain dates... 

But, you still might be able to get creative with your segments... I don't know enough about what is happening on those inflated calls... but in the past, I have created rules that excluded inflated hits due to an "infinite reload" issue on tablets (or rather anything that was at the breakpoint between tablet and desktop - this was our issue).

I basically created a rule that looked for:

Exclude Hit [

    Reloads exists

    AND

    Mobile Screen Width equals 768

    AND

   Date Range x-y

]

This wasn't perfect.. and I would have lost some real traffic from earlier in the day before the issue started, and the end of the day after the issue was fixed... 

However, you are dealing with videos, so you might not have a clear cut value to look at 

Essentially, I excluded all the reloads for tablet within the time frame of the issue, but you may not have a value like reloads to use as a compensation... 

But, maybe you could look at Video Views and Page Views? Using the number of Page Views as a limiter for the Video Views... you might cut out some legitimate "video views" from people pausing and restarting.. or maybe you have multiple videos per page.. 

But it's something to try and get those values under control.....

[AbhishekDhami]

Thank you Jenifer for your reply and sharing your past experience.

Yes i agree, i also could not find any solution to create such time boxing segment "over X hits within Y minutes"!

In our case, we are not seeing inflated numbers in Pageviews for the visitors with High Video views numbers, so its not possible to compare with other metric.

I observed that most of the cases, within minutes seeing hundreds of Video views, and i remember after 100 hits within 100 seconds adobe create new Visit. So with this logic i have created below segment.

Within Visit,

If Media Player Instances are exactly 100.

(attached one screenshot of the report)

Again above is not prefect solution but with above segment at least able to get frequency of the anomaly at monthly level.

Thank you again for sharing your thoughts.

[AneeshT]

If you just want to know when the anomaly happened, no need to create a segment. Instead, use the minute dimension with your desired metrics. This will help you pinpoint the exact time of the anomaly.

You can also create a line chart to track hourly trends. If any outliers show up, you can easily spot them and set an alert for timely notice.

[AbhishekDhami]

Thank you Aneesh for reply and sharing your thought.

Yes right, we can create alerts for detecting future anomalies.

Actually we wanted to see, in past months how many times this issue has occurred per visitor not due to events or campaigns but because of the application bug, where within minutes seeing hundreds of video views.

with this we can remove the incorrect numbers from reporting.

[DhivyaVinoth]

We have seen this kind of spike before in one of our app implementations. Adobe doesn't really give us a way to create segments for exact time window. Instead, you can create a visit level segment where video views cross a high threshold like 300+. Add a sequential condition to detect bursts.

Curious to know if anyone else found a better way to get this at a more granular level.