top of page

How to Read the PPC Ninja Anomaly Detection View



Learn how to troubleshoot anomalies in your PPC campaigns by analyzing the data in the anomaly chart and identifying changes in bids or impressions associated with a drop in sales. This guide provides step-by-step instructions and examples to help you optimize your PPC campaign with PPC Ninja's new anomaly detection feature.



Getting Started


To prioritize the most actionable and impactful anomalies, we will start off by applying some filters. First, ensure that only the most recent anomaly for each keyword is shown by filtering the Keywords Most Recent Anomaly column to True.


Since an anomaly may only be actionable for a short period of time after it is identified, click on the Date column to add a descending (arrow down) sort which makes the most recent anomalies appear first.


Next, we will refine our ordering by adding a secondary sort. While holding down the shift key, select either the Anomaly Impact (WoW Sales) or Anomaly Impact (Total Sales Last 90 Days) column and add an ascending (arrow up) sort.


The end result should look similar to the example below. We see the rows have been ordered starting with the most recent anomaly at the top, and when two anomalies appear on the same day the one with the more severe Anomaly Impact (WoW Sales) score comes first.

Now we are ready to go into troubleshooting mode!



Troubleshooting an Anomaly


Let’s begin troubleshooting by taking a closer look at the data using the anomaly chart. We will first investigate whether or not there was a loss of sales traffic associated with our anomalous point. We can do this by plotting Sales 7 Day Avg along with either Impressions or Clicks, as shown in the chart below.



From the above chart we see that an anomaly occurred on April 3, 2023 (black dashed vertical line), and that when the anomaly was flagged our Sales 7 day Avg had been steadily declining for several days. We also see a sudden drop in impressions about a week prior to the anomaly, which means we should investigate if this reduced traffic could have been brought on by a change in bids. Removing Sales 7 day Avg from the plot and adding Bids results in the chart below, where we see a large bid drop coincident with the loss of impressions.


Now that we have identified a drop in bids was associated with our anomalous sales, we may choose to take action by manually adjustings the bids back to their previous values. This is as simple as following the link to the Keywords view and editing the bid value! Prior to manually adjusting bids, you may also want to verify that at the previous bid level your ACoS was still within an acceptable range. This can be done right within our charting system with the ACoS 7 Day Avg toggle, shown below.


In the previous example, we saw a case where an anomaly was the result of reduced impressions and took action to increase sales traffic by manually optimizing the bids.

Now let’s look at an example where an anomaly can not be explained by reduced impressions which are associated with a change in bid value. In the chart below we once again show Sales 7 Day Avg and Impressions, with a steep drop in Sales 7 Day Avg on March 27, 2023 triggering an anomaly.


Similar to our first example we see that Impressions had been declining at the point the anomaly was flagged. However, unlike the first example, we note that the Sales 7 Day Avg line was flat for exactly seven days prior to the anomaly, which is an indicator that the moving average contained just a single non-zero sales day. We can visualize what this means in the chart below, where we show Sales along with Sales 7 Day Avg and have expanded the time frame to show more of this keywords long run sales characteristics.


From the chart above we see that a spike in sales occurred on March 21, 2023 which was retained by the Sales 7 Day Avg signal for a week afterwards. Once the March 21st sales spike was no longer contributing to the Sales 7 Day Avg, an anomaly was triggered. While this anomaly did correctly alert us to the fact that a week had passed without this keyword generating any sales, the chart above illustrates that a zero sales week is not out of the ordinary for this keyword and therefore the anomaly could be considered a false alarm.


At this point you may be wondering - but what about those declining impressions that we noted at the start of this example? Once again, we can investigate this by charting impressions along with the bid value over time, as shown below.

The chart above illustrates that the bid remained constant in the days leading up the anomaly. This quick check shows that, unlike our first example, the observed variability in impressions is not associated with a change in bid value.


Conclusion

In summary, what constitutes ordinary performance can vary considerably from keyword to keyword, which means there could also be a variety of reasons behind the anomalies they generate. An anomaly is most actionable when it is accompanied by a loss of sales traffic which is also associated with a change in bid value, as we saw in the first troubleshooting example. Keep in mind that the anomaly detection algorithm works better when it has more data to utilize, which is why PPC Ninja currently only tracks anomalies for the profiles top 40 keywords as determined by total sales over the last 90 days. Limiting the number of keywords helps reduce the number of anomalies like the one we saw in troubleshooting example 2, however, these types of false alarms still do occur, particularly on smaller accounts or profiles in smaller markets. The best way to understand the story behind a particular anomaly is to dive in and start investigating, and with PPC Ninja’s built-in charting system, those insights are just a few clicks away.


Definitions


Anomaly: A point in time where a keyword's recent (7 day) average sales performance has fallen significantly below its longer term (90 day) average sales performance.


Anomaly Impact: Provides a relative measure of the effect an anomaly could have on overall sales performance. Impact scores are treated as rankings, where lower values indicate a more adverse effect on sales.


Anomaly Impact (WoW) Sales: Measures the impact of an anomaly according to the keywords week over week sales performance. Lower values indicate a larger drop in sales between the week that the anomaly occurred and the week prior.


Anomaly Impact (Total Sales Last 90 Days): Measures the impact of an anomaly according to how successful the keyword has been over the previous 90 days. A value of 1 indicates that the anomaly occurred on the keyword with the most total sales over the last 90 days.


Week over Week: Values specified as “week over week” represent the difference in the current and previous weeks performance, where current week is defined as zero to six days prior and previous week is defined as seven to thirteen days prior.


Week over Week Sales: This value is calculated by first adding up sales for the current week (0, 1, 2, 3, 4, 5 and 6 days prior to the anomaly) then adding up sales for the previous week (7 to 13 days prior to the anomaly). Subtracting the previous week total from the current week total gives the week over week sales. A negative value indicates a loss in sales since the previous week saw more sales than the current week.


Week over Week Impressions: This value is calculated by first adding up impressions for the current week (0, 1, 2, 3, 4, 5 and 6 days prior to the anomaly) then adding up impressions for the previous week (7 to 13 days prior to the anomaly). Subtracting the previous week total from the current week total gives the week over week impressions. A negative value indicates a loss in impressions since the previous week saw more impressions than the current week.


Moving Average: A time series dataset where each point is calculated to be the average of a fixed number of most recent points from another time series dataset. This has the effect of “smoothing out” short-term fluctuations and highlighting longer-term trends in the original time series.


Sales 7 Day Avg: The seven day moving average of a particular keywords sales time series.


Sales 7 Day Avg (adj): A moving average calculated using a modified version of the keywords sales time series, where large outliers have been limited to some maximum value. In many cases this series will be identical to Sales 7 Day Avg, and it is most likely to differ from it immediately following a large spike in sales (e.g., Prime Day).



Want to learn more? Checkout this video featuring Ritu Java, CEO of PPC Ninja talking about keyword management.


PPC Ninja is a powerful software helping agencies and brand owners increase their advertising efficiency with cutting edge techniques such as the Bid Sufficiency model, Zero Sales Algorithm, Placements X-Ray and more. Sign up for a 14-day free trial here.

If this post sparked your interest, leave us a comment or a question!


Recent Posts

See All

Comentarios


bottom of page