Automated Personalization in Sitecore - AI, Cortex, and more
Today, a client of mine turned off the Personalization on their website, even with great engagement metrics, and it has pushed me to create this post faster than ever. The reason for switching off the Personalization? They find it too difficult to maintain. This has been a common story for those of us who convince clients to go to Personalization, but Personalization requires care and maintenance, and very often sees the light of the day for only a few days. In comes Automated Personalization. The ultimate goal of Automated Personalization is to reduce human intervention with care and maintenance.
What is Automated Personalization?
The new trends for website Customer Experience has been strongly moving towards AI and Machine Learning. To be more like the Amazons, Netflix, and even Ubers of the world, AI in some form on websites has become a necessity. Companies like IBM (with Watson) , Saleforce (with Einstein), Adobe (with Sensei) and Sitecore has commoditized AI with Cortex and now Sitecore AI.
Sitecore’s tools, unlike the others are mainly focused on personalization and the automation around it to make it scalable and maintainable. Automated personalization is using AI to in some form automate the rule sets that would have to otherwise be manually created and maintained by content authors. The AI would supposedly show users the most applicable variant of content to them to get the most engagement, click throughs, and ultimately purchases, ROI, and loyalty. Examples of automated personalization include the recommended products and videos shown to you by Amazon, Netflix, and Youtube, but can also include the search suggestions in Google.
Sitecore has two main tools for Automated Personalization - Sitecore AI and Cortex which both use Machine Learning to display the best variant of content to the user. With the introduction of Sitecore AI, I have heard some confusion in the community in terms of what is it, how to use it, is it right for you, and how does it compare with Cortex. I hope to clarify some of these questions with the below blog post
Sitecore AI
What is it?
Sitecore AI is a SaaS ad-on onto any version of Sitecore above 9.0. It is expected to be released before the end of the year, as per Sitecore, but it’s not yet on the market. The first release of Sitecore AI will be similar to Cortex:
First, it will use your existing xDB behavioral data to create 8 clusters. of users. In the first release this number is static
Next you will need to create variants of the component where you would like personalization
The AI will put each user to the site in their appropriate cluster
The AI will automatically pick the variant that has the best chance of getting the most engagement the user, based on which cluster they fall under
Additionally Sitecore AI has a dashboard that gives you more info about each cluster such as: how many visitors fall under each cluster, where do they come from, what does their profile card look like, which pages and components engage them the most, etc. An example of this dashboard is shown below:
Screenshot from Sitecore AI session at Sitecore Symposium
With the first release, Sitecore AI will only be using Engagement Value Score as the success criteria, which means your user will see the variant that is the most likely to get the highest Engagement Value Score.
A working case study
Sitecore touted their work with the Microsoft Partner Network, where they premiered Sitecore AI. Microsoft has 6.6 million unique visitors/year, where 29% are authenticated users to the Partner Network portal. These users generate 56 million page views/year, and come from 100+ markets viewing the site in 11 languages. Before Microsoft launched the AI pilot, they had personalization on 6 pages with 38 variants. This complexity was getting difficult to maintain with 11 languages and 26 locales in the mix as well.
After the pilot, Microsoft saw a 16% lift in their Engagement value score. Additionally, they were able to gain operational efficiency, which is rare to hear about with Personalization. They saw a 30% reduction in design and configuration time as well.
Is Sitecore AI right for me?
A few things of note, for at least the first release of Sitecore AI:
Sitecore AI is an ad-on product to Sitecore, so most likely will have additional cost, and while the data is yours, you do not control the algorithm. That piece will come from Sitecore.
You must have at least 12 million visits per year to your website in order to make use of Sitecore AI at this time
You must be on Sitecore 9 and above
If you have tried Sitecore Personalization, are struggling with the maintenance or have turned it off because you’re unable to maintain it at this point, Sitecore AI is for you.
The Future of Sitecore AI
Sitecore intends to invest further in this technology, and make it content and journey based eventually.
Sitecore Cortex
What is it?
Sitecore introduced Cortex in 9.1 with 3 flavors.
1) Personalization Suggestions after running an A/B Test. This is out of the box Sitecore functionality.
2) Automated tagging of your content by plugging into Open Calais. All this requires is entering your Open Calais key into your Sitecore config file, and works well on non-domain specific content.
3) A framework to process your xDB data for a machine learning algorithm. Jason Wilkerson, also known by his Sitecore stage name Longhorntaco, created a demo of this that can be downloaded here: https://github.com/LonghornTaco/tacoflix-demo
If you want to hear more about the 3 flavors of Cortex, here is the webinar that I conducted earlier this year, along with Petr Anderle from Sitecore’s SBOS team: https://www.youtube.com/watch?v=Eut2xEna0NY.
Automated Personalization with Sitecore Cortex
The first flavor of Sitecore Cortex “Personalization Suggestions” is the focus of this blog post. This feature is really similar to Sitecore AI. The main difference between Cortex and Sitecore AI, is that Cortex is part of your Sitecore package with 9.1 and above, but it requires a bit of setup. The way it works is that you setup an A/B or Multivariate test on a component you would eventually want to personalize. Rather than just picking the winner, like traditional A/B Tests, Cortex will further provide “Personalization Suggestions” selecting segments where the winner works better, but also segments that the loser may work better.
In this image by Pietr Brinkman of Sitecore, the test loser shows a 234% better engagement as compared to the winner for a certain segment of users on the site, who come from Denmark on a tablet, and have seen the 10% discount campaign. In this case, it’s a no brainer to continue to show the loser to that segment of user, even though in general the A/B Test winner would perform better.
Una Verhoeven presented at Sitecore Syposium on this topic. The following blog post by her is a good summary of her presentation: https://www.ampleedge.com/blog/getting-started-with-sitecore-cortex-personalization-suggestions. The following is her suggested process flow in implementing Automated Personalization with Cortex. For example, unlike manual Personalization in Sitecore, you would setup profile and pattern cards and keys, after you run your A/B Tests.
Is Sitecore Cortex right for me?
Sitecore Cortex is a great first step into Automated Personalization as you wait for Sitecore AI. It is available out of the box in Sitecore 9.1 and above today! Especially if you are already running A/B Tests, with enough users, Cortex will give you those personalization suggestions today. While not as robust as Sitecore AI, Cortex is a perfect stepping stone, before you decide to spend the money. Cortex also requires a large number of visits to be successful, but is based on the number of variants you have running at the time.
Remember, Sitecore Cortex has an additional 2 flavors:
1) If you have issues with tagging your content and if your content is not too specific within a domain, the second flavor of Cortex with Open Calais tagging is for you!
2) If you need custom personalization (not present in Sitecore today) such as a Netflix or Amazon-like recommendation engine, you can use Cortex as your processing engine to send your xDB data to an ML algorithm to then send you the recommended products or content.
The Future of Sitecore Cortex:
Sitecore suggests that the framework portion of Cortex will continue to exist, and Cortex can be used for creating your own ML algorithm or recommendation engine on top of Sitecore.
In Summary
Sitecore AI = brand new automated personalization, potentially releasing this year; will be turn key, a Saas ad on to Sitecore, and requires 12 million+ visits/year
Sitecore Cortex = 3 flavors: automated personalization in Sitecore 9.1 and above, Open Calais for tagging content, and a framework to process xDB data and apply your own ML algorithm (bring your own AI)
I have seen it time and time again that even with understanding the power of Personalization on digital engagement, clients have a difficult time getting started, and when they do turning it off quickly thereafter. Being a Sitecore client myself, I tried to turn this on 3 times, without success in maintaining it. It was turned off within the year every time. The promise of AI in automated personalization is to reduce the burden of maintenance, and get to increased engagement and ultimately increased ROI and loyalty from personalization.
