Managing Personalization Via the Dashboard
On this page
Personalization involves several distinct, but related tasks. Like a puzzle, every piece matters but they do not all serve the same function. For example, it is best to start with the corner and edge pieces than the all-blue sky ones. It’s the same with Personalization, there are three distinct pieces:
- Defining and sending events
- Knowing what you are trying to accomplish and how it works
- Creating actual user preferences
Furthermore, like puzzles, personalizing results is a collaborative process. Personalization is a combined effort of technical and non-technical people. For example, developers add code that send events, and marketers design and configure personalization strategies.
The Dashboard is the main tool for putting together personalization. For this tutorial, we will focus on this screen:
Before getting started with the Dashboard, let’s discuss what you need to do first. We’ll start with the corner and edge pieces - the events: Sending events is the natural starting point, they make personalization possible, they frame the subject.
While the actual sending of events is a coding task performed by developers, the strategy behind events is a non-technical undertaking. As a result, deciding which events to send is usually done by non-developers like marketing personnel and product managers.
Personalization events are not taken into account right away. User profiles will be updated roughly every 10 minutes.
And what is the strategy? To send Algolia every user action you consider useful in defining a user’s preference (or tendency). Typical actions include searching, clicking, adding to a cart, and buying.
Insights events (click, conversion, view) used for analytics and/or personalization do not take immediate effect. The delay can range from 10 to 60 minutes depending on how long after the search they are sent. For precise times, see our page on when Insights events take effect.
Understanding events and impact
Continuing with the puzzle, let’s look at the blue skies - those pieces that create the background understanding. You need to:
- choose the right kind of events to send,
- understand how personalization impacts the ranking and overall relevancy of your results,
- and know what actions (on the dashboard) you need to take to configure Personalization and create user preferences.
Creating reliable user preferences
Once the surrounding pieces are put in place - that is, once the correct events are being sent, and you understand how personalization works and what actions you’ll need to take - you can start to work on the central image of the puzzle, namely: your users - who they are and what they prefer, their tendencies.
From this point forward, you’ll be using the Dashboard to create user preferences and determining how much you want personalization to impact the search experience. Like many of Algolia’s features, such as relevance tuning, merchandising, and synonyms, the Dashboard is your main tool to manage personalization.
There are 3 main pieces to the personalization screen:
- Impact, where you determine how much you want Personalization to impact the search results and the ranking formula.
- Events, where you define which events to track and their respective significance in building user preferences.
- Facets (filters), where you choose which facets/filters best reflect the choices users make regarding the products that interest them.
1 - Setting the overall impact of Personalization
Setting impact decides the actual effect that user preferences will have on the contents and order of search results (i.e., on relevance and ranking).
A. 0%. Personalization is disabled, it will play no role in the ranking.
You can read more about the personalization calculation to get more insight in the impact of personalization.
B. Personalization is now at 50%. This means that it affects your ranking, but not too strong.
C. 100% impact. The impact of personalization is at its highest. This does not mean that all results are personalized. Personalization will always remain only one part of Algolia’s multi-criteria ranking formula. Setting personalization to 100% only means that, given Algolia’s method of calculating personalization, user preferences will play its strongest role in the overall ranking formula. Another way of looking at this is that personalization will play nearly the same role that custom ranking plays on the results.
Note: Changing this value does not in any way affect the ongoing capturing of events nor the calculation of user preferences. It only affects whether or not, and by how much, those user preferences are applied during the ranking process.
2 - Defining events
With this part, you’ll be doing two things:
- defining the events that need to be coded into your front end,
- defining the weight that each event plays in the collecting of user preferences.
In the above configuration, we track four events. Events that Algolia receives affects a users’ preference.
Each event is scored differently.
The configuration on the left indicates that Adding to wishlist is more important than Add to cart, and that checkout and filter toggling are the least important preferences indicators. To be precise, adding to a wish list is five times more significant than adding to a cart (100/20) and ten times more important than checking out and toggling the filters (100/10).
The configuration on the right is completely different. Now, Add to cart and checking out are both five times more important than adding to a wish list, and toggling the filters has low significance.
3 - Defining facets and weighing their importance
With this final piece, you create the precise preferences, tendencies, likes, tastes, buying habits, and so on, for all of your users, on an individual, personal basis. These settings determine what sort of products a particular user wants to see. Category-based preferences will then be mixed in with the textually relevant results.
In a book search, for example, the query term “harry” will still be the essential basis of all of the results – that is, all results will contain “harry” - but a user’s preference for politics or children’s literature will determine whether Harry Potter books or Harry Truman books show up first.
In the above configuration, we’ve chosen to track authors and categories (such as horror, sci-fi, romance, etc.) as the preference basis. This means that other pieces of data, such as year of release or media type, aren’t taken into account.
On the left side, we’ve made author five times more important than categories. Results containing a user’s favorite authors will show up higher than their favorite categories.
On the right side, we’ve switched preferences, making categories six times more important than author. A user who prefers romance novels and likes Stephen King will see more romance novels than Stephen King books, or any horror novels by other authors.
Simulating personalization to test your settings
Now that you’ve configured your Personalization strategy, you may want to test how effective it is.
You can find out with our interactive tools that let you see your Personalization choices in action.
In the Strategy section you can choose the events and facets that will have an impact on the user profiles.
In the Simulation section you see the immediate impact of Personalization by selecting an existing user token and performing search queries as if you are this user.
With a user token selected, every search will display results without Personalization on the left, and results with Personalization on the right.
Both columns contain several pieces of information:
- the rank, which is highlighted in blue if the item is personalized
- the ranking difference between the personalized and non-personalized results:
- - means the item has the same position with and without Personalization
- -2 means the item moved two positions lower with Personalization
- +2 means the item moved two positions higher with Personalization
- double-arrow means that the item moved from a position lower than 200.
You can click on a result to find out why it is personalized.
Personalization works based on a user profile, based on the events you sent during the last 90 days. We use this user profile as a set of scored filters.
The scores depend on both the quantity of actions that the user performed, and the importance of each action as defined in the Personalization strategy. Each score expresses the affinity that the user has for a given filter. The higher the score of a filter, the higher matching records will be boosted at search time.
To simulate strategy impact, you need to focus on specific users.
In the User section, you can select real users from your application and display their profile. You can either enter the specific user token you’re looking for, or select one from the displayed suggestion list of recently active user tokens.
The profile is then displayed along with the date of the last received event associated with it.