Smart Alerts: Use AI to Detect Data Anomalies When They Occur

Noticing changes in data is the key to being a good analyst and, at the same time, one of the biggest challenges. Keeping an eye on all the important metrics and dimensions takes time and is mostly boring busywork. Only once a change in the data is discovered, an analyst’s real job can begin: analyzing and finding an explanation for the shift. In order to cut down on busywork, Mapp has recently introduced Smart Alerts, an alert system that keeps an eye on selected, important metrics and dimensions and notifies users about significant changes via email.

This is a big relief for the analyst. No more tedious report scanning or dashboard monitoring just to make sure nothing has changed –instead, the analyst can focus on really analyzing the data, i.e. how a detected change was triggered.

How it works

We look at the past behavior of a time series (e.g. the number of visits), and via machine learning we try different models (e.g. exponential smoothing, moving average, or ARIMA to predict the next data point). The model that proves to be best at predicting the next value based on past data is used to predict future data points for this time series. When the next value is above or below a configurable threshold, the alert is triggered, and an email is sent. The threshold configuration is done via a sensitivity level –more sensitivity leads to smaller thresholds and quicker alerts.

What it looks like

Take a look at the image below, directly from Mapp Cloud. The blue line shows the actual value of the metric (Quantity Orders in this case). The red dot shows the anomaly. The blue shaded area marks the calculated thresholds: this is where a data point is predicted to be. The more sensitive you set the configuration, the smaller this area will be.

How you will benefit from Smart Alerts

The metrics that come to mind first for monitoring are the essential ones: Visits, Visitors, Page Impressions, Qty orders, Conversion Rate, Bounce Rate. When one of these metrics suddenly spikes or drops, it is crucial to figure out why –the faster the better. While this is a good start, such an alert does not indicate the source of the irregularity. Only looking at one metric is not the recommended way to really discover in-depth insights.

A better approach is to monitor a dimension with a metric. It’s a good idea to choose important dimensions, e.g. device class + visits. By setting up such an alert, you monitor not just one-time series for overall visits, but up to ten-time series –one for each device class (e.g. top 3: desktop, tablet, mobile). So, if one of these alerts is triggered, you immediately get an idea where the root of the problem may lie, e.g. with the mobile version.

Other interesting dimensions to select for monitoring are Entry Source Type, Campaigns, Browsers, Country, Pages or Page Categories, Product Categories, and of course any custom dimensions you set up.

The combination of a metric + dimension allows for a quicker understanding of what is happening on your site and what might have prompted the data irregularity.

Have a look at some more ideas for great monitoring combos and what they can tell you:

• Bounce Rate + Campaigns give you a good indicator of when a campaign suddenly drives low-quality traffic to your website.
• The combination of Conversion Rate + Device Class lets you know when the sales funnel does not work for a particular website version.
• Visitors + Country will inform you of local irregularities, e.g. a public holiday.
• Conversion Rates for every step in the sales funnel point you immediately to the step causing difficulties.
• Bounce Rate + Landing Page alerts you when a specific landing page suddenly stopped functioning.
• Conversion Rate + Browser/Bounce Rate + Browser alerts you about technical issues with your website.

Who you will want to alert

Finally, not every alert is equally important to everyone in the company. Make sure to send alerts that hint towards a technical problem to the technical team and alerts that are pointing towards problems with campaigns or landing pages to your marketing team. Overall metrics like traffic and conversion might even be of interest to the C-Level team.

And remember, alerts do not only hint at problems –they can also alert you when things go better than predicted. Hence, they provide a great way to let everyone know when they did a good job. We recommend you use alerts to make sure people get positive feedback and appreciation as well.

Have you discovered alerts combos you would like to share with us? Let us know on social media and we’ll make sure your voice is heard by the rest of the marketing community!

P.S. How to get the feature

Smart alerts are currently restricted to admin users, who can give access to the feature to other users. If you don’t see a bell icon in your account, get in touch with your admin or a Mapp Consultant to get access to the feature. Every account gets one access for free, so give it a try!

Want to learn more about Smart Alerts and Mapp Intelligence get in touch with us:
Get in Touch