Hidden From the Static Eye

Don't be Static (when monitoring your data)!

Last week, I gave an example of why Static Threshold Alerts won't cut it.

In my earlier post, I explained why is it a must to have a sophisticated machine learning alerting tool like Anodot.

To summarize, A static threshold just won't catch or will alert about an incident, when it is already too late.

Below, you will find two examples. The first example demonstrates how a static threshold won't or can't find certain anomalies.

The second example demonstrates early detection use case.

First Example:

Hidden from the Static eye:

There are times where you have an anomaly that is within the expected seasonal behavior, however, it doesn't reach to lower or upper thresholds.

In this hourly aggregation example. You can see the data in blue, covered by a sleeve that represents the predicted behavior of the data. The red lines represent upper and lower thresholds.

In these cases, it might take, not only hours but even days or weeks to catch an anomaly,

Second Example:

Early Detection:

In another live example of hourly aggregation, you can see the data in blue, covered by a sleeve that represents the predicted behavior of the data.

If our client's monitoring was based on a static threshold, then the anomaly would have caught only at 5:00 am.

Here is how it looks like:

However, if we zoom in, you can see that with Anodot, the anomaly was caught at 3:00 am, two hours before, once it deviated the predicted behavior shortening the incident detection time.

Please also note that the deviation started on the off-peak of the metric,

The following also applies to all levels of aggregation, from minutes through hours to days and weeks.

If to summarise the above,

When every second, minute, hour counts, time is literally money. Bleeding revenue or users.

Or in short, Businesses Can't Afford to Find Incidents Late!

I hope it makes sense, please feel free to contact me directly if you have any questions or if you like to have more information.