Getting started with Personalization Suggestions

AmpleEdge

Getting started with Personalization Suggestions

Cortex is Sitecore’s vehicle to bring machine learning to marketeers. The most practical use of this is the Personalization Suggestion functionality.

In this blog post, we will dive into the why and how you can use Sitecore Cortex Personalization Suggestions.

Personalization Suggestion is part of Sitecore Cortex, where Sitecore collects data to find out which experience gives the best result. It is available as part of the Experience Platform starting from version 9.1.

The benefits of Personalisation Suggestions are more than one. It can:

  • Help your organization in defining segments of visitor-patterns (personas).
  • Define personalization rules for you.
  • Be used for personalization rule’s strategy - you can use the outcomes to verify the rules that you already have.
  • Show content “flaws” - visitors arriving on pages with less impact, suggesting that content restructuring is required

Making personalization suggestions work requires two steps:

  1. completed A/B tests in Sitecore
  2. Big data

A/B tests

When we talk about AB testing, we first need to think of the candidates for testing. Not every page is suitable to be tested, but it is also not required to do so. Instead, you need to make a plan to identify valuable pages. Those pages are either:

  • Impactful; these are the pages that generate lots of traffic
  • Decision-driven; the page where you typically have goals & engagement values placed, such as pages containing forms.
  • Landing pages; the types of pages where visitors land when they come to your site, either via link published on social networks or via search engines.

An important thing to note is, when you want to use the Personalization Suggestions engine, you can only see it executed under two conditions: only when running a component test or page test.

To generate a page test, you can do it either via Experience Editor and Optimisation Ribbon or directly in experience Optimiser available on the dashboard.

Big data

The term “big data” is slightly different in this context to the more general term used in machine learning. In this context, big data is used to describe data as the information that has been driven and generated as a result of the executed A/B tests, and big because you need a significant amount of data to be able to get Cortex to suggest. How signification you ask, it depends!

Sitecore Cortex looks into the data and learns from it. The accuracy of the assumption correlates to the amount of collected data. The more available data, the more precision will be accomplished.

Working with personalization suggestions

So suggestions work based on the test results. Which is why it is necessary to drive visitors to your pages where A/B testing is happening. To drive traffic, you can utilize convenient Sitecore features like (email) campaigns and marketing automation.

For personalization suggestions to appear, the tests need to be finished.

Please note that not every finished (historical) test guarantees that personalization suggestions will appear. The data collected needs to be significant.

Once both conditions are applicable, the test will appear under Personalisation Suggestions under the Personalisation tab in Experience Optimiser.

How does it work?

Sitecore Cortex relies on decision trees where each tree has nodes. Each of the nodes has:

  • The rule that defines the node
  • Engagement values for all tested variations
  • Number of visits in the node

If the engagement value of any tested variation is higher than the one of the Test winner, Sitecore will present this as a suggestion. The assumption is that this trend will continue.

How to read the results?

The relative uplift of this value in comparison to Original is the “predicted effect compared to the test winner.”

When you apply this as personalization, the overall change in the uplift also depends on the segment size, which is why Sitecore weighs it with the segment size used as a weight

The weighted predicted effect = Predicted effect compared to test winners * % of total visitors/segment size (100

Screenshot Sitecore Personalization Suggestion

The total predicted effect is the sum of all weighted predicted impact.

The word “Predicted” is added everywhere to indicate that it is not guaranteed; it is Sitecore’s approximation of past trends.

Example:

In this case, this result is 121,85% better than the original component, as measured for visits within this specific segment.

121,85% * 3,55% / 100 = 4,54% is the value of weighted predicted effect.

Applying personalization

If the suggested personalization correlates towards your business goals, you can decide which of the Experiences, one or both, you want to use. This is accomplished by selecting the checkbox and clicking Apply. Sitecore will create personalization rules which target the described audience in the test. Visitors who do not fall into this category will see the original content displayed.

Personalization Flow

As with every marketing strategy, personalization requires a lot of planning, testing, verifying and adjusting.

Personalization Flow

As shown in the graph, this is an ongoing process that requires agility. You always need to think of returning visitors, since, with time, they will accumulate enough Engagement Value points. Create a plan and allocate the time which will allow you to analyze the gathered data and executed tests and personalization rules. Depending on the number of visitors you have, the advised time frame to re-evaluate your strategy is 3 months.


AmpleEdge

Una Verhoeven

Founder

Do you have questions?  Una is happy to help!

  • LinkedIn icon
  • Twitter icon