Why the “traditional” way of KPIs tracking is not an option (any more).
As a marketer, you are interested in how various marketing KPIs and metrics evolve over time. In theory, you should concentrate on 2 or 3 main marketing KPIs which you can track regularly and then adjust your marketing tactics based on that.
However, in practice, the amount of marketing KPIs and metrics can be massive. In addition, they can change over time based on your focus and priorities. It is very hard for a marketer to keep an overview of what's going on with hundreds of KPIs at hand.
This can make it hard to track and detect the abnormal behaviours of very important metrics. However, it is becoming more important to know and understand unusual and surprising patterns. An anomaly in one metric can affect other metrics and also directly have an impact on your bottom line or ROI. This makes the process of KPIs tracking especially important in situations where urgent action is required from the marketer.
You can monitor these metrics manually by looking at dashboards and visually checking each one of them. However, this method of marketing KPIs tracking has its clear drawbacks:
1. First of all, it is very time-consuming.
2. Second of all, it is tiring and can be really annoying and error-prone - especially if you do it every day.
3. It doesn’t allow you to scale - it is physically impossible to monitor all metrics when there are that many.
4. Less energy left for creative work.
5. When you find an anomaly, it might be already too late to take action.
6. Last but not least, it is very expensive to do this on a regular basis.
As a result, the process of tracking KPIs is not used frequently and many marketers skip certain steps or leave it out completely. Missing the important anomalies can lead to increased business costs and lost opportunities and directly have an impact on your ROI.
Monitoring your marketing KPIs at scale
Nowadays, the process of marketing KPIs tracking can be automated with the help of artificial intelligence. AI technologies can make it way faster and more effective. However, before implementing AI technologies make sure that your data is prepared for integration.
Different AI technologies provide a wide variety of solutions to this problem.
One of these solutions is “Anomaly detection” algorithms. These algorithms detect the anomalies in time-based KPIs and can notify you promptly in case of any findings.
There are three general types of anomalies: outliers
, sustained increases or variance changes.
Now, let’s have a deeper look at these common anomaly types.
For a better understanding, we have provided a simple illustration for each anomaly type.
You can easily identify outliers by looking at the graph below. Those are the data points that differ significantly from the main population. It is an abnormal observation that lies far away from other values.
As an example, these could be conversions you have made after holding an important event, or an influx of visitors on your website after making an important announcement at the press/ on social media. The higher value is only short-lived but can be very important to you. The outlier can also be negative, such as a sudden drop in revenue for your main product.
2. Sustained increase
Here, we can see an example of a sustained increase. Basically, it signifies an abnormal behaviour that lasted for a certain period.(was extended over a certain period).
It can represent both sustainable upward and downward trends.
As a marketing example, we can think about a heat wave which subsequently causes a spike in the sales of ice-cream during that time. Another example is, if you have a problem with your website for a certain period, the amount of bounces might resemble that.
3. Variance change
Variance change represents an unsustainable fluctuation; a significant decrease and increase in the metric’s value. One can say that it is a very fast changing behaviour of the number. As an example, this can happen when you sell globally and suddenly the foreign-currency starts to fluctuate heavily, which in turn has a direct impact on the sales revenue in your local currency.
How exactly does it work?
Basically, the anomaly detection can optimize the marketing KPIs tracking process by fully automating it. It reveals significant anomalies and abnormal behaviour fast and way more consistently than a team of marketers.
Anomaly detection can help you to find anomalies in your most important KPIs and immediately notify you about it.
It is important to note that anomalies can have both positive and negative effects on the business. Some anomalies signify the improvements. For example, the success of specific marketing actions, such as conversions or others like having a problem with your e-commerce platform might have a very negative impact on your revenue.
Algorithms and the specifications of the statistical models can vary depending on what your KPI is. For example, at Nexoya, machine learning and statistical models specialized in anomaly detection on time series data are leveraged heavily. Nexoya’s algorithms are built in a way that they also account for trends and seasonality, as well as for holidays and special events.