Learn all about the new
Update in the i2i-algorithm
All our customers will now experience a comprehensive upgrade to one of the key components in our recommendation engine, the i2i-algorithm (also known as item-to-item).
The new upgrade offers you more logics working together to ensure better and deeper recommendations. You don’t have to do anything; the upgrade will happen on its own.
Below, we will dive deep into the most significant improvements and how they affect your personalization onsite.
Remember, you can always reach out to our Professional Service team if you need further information on the new algorithmic adjustments.
There are many ways to utilize this feature, so we have made some examples of how you can use it for your business below.
What is the i2i algorithm?
Target customers that only buy products that have a discount
With the new update, all the underlying calculation has been moved to a different platform which ensures a more powerful engine that come to fruition faster with more relevant recommendations delivered. In other words, customers will experience faster and more frequently updated recommendations on products.
What has changed?
1) Automatic Serendipity adjustment for a better balance between products
Serendipity makes sure that your recommendations include a mixture of relevant and surprising items. If you set a high serendipity score, you suppress popular products from the recommendation modules, for example, milk and bananas in an online grocery store. Without serendipity, you will recommend overall popular products, even if these recommendations might be trivial and uninspiring. The products are there because they are common items in customers’ baskets. See the example below.
With the new update, the serendipity adjustments are no longer a manual procedure but built into the component itself. The items that are used to create “noise” in the module will be removed automatically. This creates shorter call time and more related recommendations as popular products are excluded if they don’t fit the context of the module.
Take for example milk, tomatoes, eggs, and cucumber in an ordinary online grocery store. Previously, such products would always dominate customers’ personalized modules, as these are often purchased together, statistically speaking. While you might be shopping for a Friday splash of Gin and Tonics, a carton of milk would – without this adjustment –create noise in your search for Gin and tonic-related products, even if you usually buy milk from this store.
Related items for this bottle of gin:
Without the serendipity adjustment:
As you can see, milk and eggs appear in the module, despite how they are not related to the gin and tonic-theme.
With the serendipity, only relevant products related to gin appear in the module.
2) More accurate relations between items
Another essential part of the new update includes 1) how tracking data can be differentiated in only a few seconds when finding relationships between products 2) the ability to find rules across tracking parameters.
What does that mean?
It means a more narrow and more profound level of personalization in our modules. The recommendation algorithm detects the relationship between purchases. Before, the algorithm would find relations between items a customer had added to their basket and what other people had added to their baskets to present the items most frequently bought for future visitors.
With the new changes in tracking data, it is possible to find the most relevant products from what others have previously searched for and not just bought (like before). Let’s say people who searched for “socks” more often ended up buying “pantyhose” instead.
Having a search parameter included gives you a much deeper level of insights into the actual outcome of a visitor’s search intention. The algorithm registers both past and future tracking events – meaning that it considers the customer’s actions both before and after the event. That has three advantages:
1. Based on customers’ previous interactions, you will know who initially searched for socks but ended up with a different product in the end. This means you get a much more accurate idea of visitors’ intentions from browsing around your site.
2. The algorithm can “predict” future purchases that the customer is likely to make. If other customers first made an order for paint and later another one for varnish (as a part of a home improvement project), other customers will get similar, relevant recommendations.
3. You get new possibilities for enriching keywords in search engines. Take, for example, a fence. Popular search results for fences include ‘slatted fence’, ‘wicker fence’, ‘garden fence’, ‘railing’, ‘wooden fence’, ‘plan work’, etc.. All these terms can now be considered to improve visitors’ likelihood of finding the product they are searching for. This feature also connects misspelled search words with the product they ended up buying, which also helps you improve product search.
3) More gateways for customers to find what they are searching for
The update also gives your customers a better experience by providing them with options related to their search. If they searched for “grill” they will be presented with options such as ‘gas grill’, ‘cleaning tools’, ‘grill cover’, etc. By adding more parameters, visitors are more likely to find what they are searching for.
What should I do?
The changes in the i2i-algorithm will happen behind the scenes. If you run personalization from Raptor Services, you will experience a much faster algorithm that includes more logic and as a result, provides your customers with better recommendations all around.
We have picked out some of the biggest advantages of the new i2i-calculation, but overall, the update allows a faster and better experience in our customers’ personalization solutions.