February 14, 2020

Let AI Focus on Finding the Needle in the Haystack

The work of the digital marketer becomes more analytical by the day. By studying changes in customer behaviour, digital marketers can learn how to improve their future customer interactions to create long-term and value-driven interactions.

A challenging role

The role of the digital marketer is changing. Reading all of the reports and dashboards to keep an eye on the numerous metrics at play takes time. But there is another way.

We believe that marketers should be able to focus on the things that will make a difference to their business, instead of spending too much time taming the technology behind it. Instead of reviewing all the marketing reports and data available, what digital marketers should be doing is looking for reasons why significant changes have occurred.

With that in mind, the platform should highlight anomalies and outliers in customer behaviour through alerts, enabling them to focus on the important changes and tell them when things are getting interesting.

Mapp has launched Smart Alerts to carry out that very role. Let’s take a look at what good looks like in an alert system.

How machine learning can help

Machine learning is key to a successful alert system.

It allows marketers to look at past behaviour – for instance the number of visits to an eCommerce site – and try out different models to predict the next data point.

The model that proves best at predicting the next sales value can then be set up to predict future data points. When a new value is above or below the threshold, an alert is triggered, and an email is sent to the right person.

What does an alert system look like?

A useful alert system should look something like the one seen in figure 1, where the blue line shows the value of the metric – in this case order quantity – and the red dots show where an anomaly has taken place. The blue shaded area shows the calculated thresholds – ie the minimum and maximum values – that a data point was expected to have fallen between, according to the machine learning prediction. The more sensitive the configuration, the smaller the area.

Figure 1: A Smart Alerts feed reveals when anomalies have taken place.

Combining metrics with dimensions

All of the obviously important metrics should be monitored, such as website visits, unique visitors, page impressions, order quantity, conversion rate, bounce rate, and so on. When one of these metrics suddenly spikes or drops, it usually pays off to find out why. But there’s an even better approach to monitoring for metrics only. This involves keeping track of a metric alongside a dimension.

For example, you can measure website visits by device type, allowing you to monitor not just one time series for overall visits, but three time series, one for each type of device – ie desktop, tablet or mobile. So, if only one of these alerts triggers, you’ll immediately know where the problem might lie – eg with the mobile version.

Other interesting dimensions that could be selected for monitoring include lead source, campaign, browser, country, web page or page category, and product category.

Combining metrics with dimensions can be hugely powerful. It allows you to more quickly understand what is happening on your website and what might have prompted an anomaly in the data.

Some examples

Here are six examples of useful metric and dimension combinations:

  1. Bounce rate by campaign gives you a good indication of when a specific campaign suddenly drives bad traffic to your website.
  2. Conversion rate by device type indicates when the sales funnel does not work for specific devices and website versions
  3. Visitor by country prompts you when there is an issue with a particular country, eg a public holiday, or some other event.
  4. Conversion rate by each step in the sales funnel points immediately to the step where conversion issues are occurring.
  5. Bounce rate by landing page alerts you when a specific landing page does not work any more.
  6. Conversion rate by browser alerts you to problems with your website that are probably technical.

And finally, tell it to someone who cares!

Not every alert is equally important to everyone in the company. It makes sense to send alerts that hint towards a technical problem to IT. On the other hand, you should send alerts that point towards problems with campaigns or landing pages to the marketing department. Finally, overall metrics like website traffic and conversion rates might even be of interest to the c-suite.

Alerts not only hint at problems, they can also tell you when things are going better than predicted. When that happens, alerts ensure that people get positive feedback and appreciation.

Digital marketers can use alerts to completely redesign their working life so that it’s more rewarding, as they allow you to focus on what’s important and highlight what opportunities are available to improve customer behaviour and performance.

Written by

Ricardas Montvila,
VP, Global Strategy at Mapp Digital