AI-based recommendation systems use Artificial Intelligence and Machine Learning (AI/ML) algorithms to determine effective recommendations. AI-based recommendation systems use AI/ML to combine and interpret product as well as shoppers’ data – such as browsing patterns, purchase history, and demographical data. These systems build personalized product and visitor profiles based on all the above data. These algorithms can therefore display cross-sell and up-sell opportunities with astounding speed and precision. These algorithms make sure that the recommendations match the shopper’s intent.
Rule-based product recommendations rely on explicitly defined static rules. E-commerce managers build these rules based on what they feel is right on explicitly defined static rules. These rules cannot consider individual visitors’ preferences. As a result, rule-based recommendations are the same irrespective of whoever is browsing the site.
These rule-based recommendations lack conversion power for several reasons:
Another major drawback of rule-based systems is the effort needed to define and maintain the rules. It is very difficult to add further rules to an already large rule base without contradicting previous rules which are still active. The time spent on formulating and maintaining these rules can better be spent addressing real customer needs.
The following is the data that is collected for recommendation algorithms
The following is the data that is collected from your website.The data is automatically collected through listeners. A listener is a method that is called when the visitor does something on the site. (e.g., click on a product detail page, click add to cart etc.)
Historical data can help us in training the AI models so that you can be hyper-effective from day one of the implementations. But it is not a deal-breaker. We can start fresh too.
We prioritize data security above all. CartUp AI shall ensure that any person who is authorized by CartUp AI to process personal data shall be under an appropriate obligation of confidentiality. Firewalls are in place to restrict exposure to the internet and all communications to the outside world are encrypted with SSL certifications. Database backups are done periodically at intervals at the end of every week. Upon deactivation of the services, all personal data shall be deleted, from the CartUp AI servers.
Yes, we are compliant with the European General Data Protection Regulation (GDPR). As per GDPR requirements, we are committed to customer data protection, privacy rights, and global compliance. Our team is trained extensively on GDPR and works on implementing and continuing with the best practices for handling personal data.
Access at the platform level is protected with a password and access is granted to customers who have subscribed to CartUp AI’s solution. Database access is restricted to the outside environment while only the application and data manager are provided with access to the database. Authentication is based on user level and is encrypted with RSA standards. User activity logging is always maintained. Multiple unsuccessful login attempts result in a locked account. Our cloud infrastructure is broken into separate services, with load balancers for uninterrupted, continuous operation.
The JavaScript code that we embedded into your website automatically reads and collects your shopper’s data.
The JavaScript code that we embedded into your website reads the products pages regularly to get the product feed. We also might also request if the product feed is available with you.
We accept all the industry-accepted formats such as XML, CVS, XLS, etc.
Our algorithms have been tested and proven to deliver relevant performing recommendations without being resource hungry. The impact on site performance has been proved to be negligible, and we can demonstrate this in PoCs.
Sometimes e-commerce managers want to customize recommendations to deliver a specific desired outcome. For example, a supplier may have agreed to give additional commercial support to promote its products and give them more prominence. This is just like paying for display space in a power aisle in supermarkets. Similarly, product promotion or new product launch may mean you want to put this near the top of recommended products in certain product categories.
In these situations, you’d want to apply rules to push certain products alongside the product assortment that has come out of the AI-based recommendation system. AI-based recommendation systems have differing abilities to incorporate business rules to accommodate these special needs.
If the customer doesn’t want to continue, widgets can be turned off with a mouse click, and the uninstalling can be done in minutes without any change in their code.
We do a PoC. Until you are satisfied, we don’t make it live.
First, our AI-based recommendation systems use self-optimizing algorithms learn from the shoppers’ and product data to hone its recommendations as applicable to your e-shop. The data can be browsing patterns, purchase history, demographical data, and product feed.
Second, the system uses AI/ML to combine and interpret shoppers and product data to build personalized product and visitor profiles.
Finally, the algorithms display cross-sell and up-sell opportunities with astounding speed and precision that matches your shoppers’ intent.
Yes, we can. We can start with a small set of customers, and slowly grow to 100% of your customer base as you grow your confidence with us.
Yes, we can. We can start with a small set of product categories, and slowly grow to 100% of your product catalog as you grow your confidence with us.
Yes, we can. We can start with a limited number of recommendation widgets and slowly grow to more recommendation widgets as you grow your confidence with us.
All the algorithms are the same. The recommendations work the same on mobile site too.
Apps may require a deeper integration. It requires an SDK-based integration.
No, our algorithms are world-class. The algorithms that are used on the web are also used on the app. The only way they differ is implementation.
If it is simple customization and that does not need a lot of man-hours, we will be happy to do it. If it requires substantial and complex work, we charge the market rates.
The ranking of the product is decided by our AI-based algorithm. The algorithms rank products based on the likely hood of being purchased by the shoppers. The ranking differs from one shopper to another as it takes into consideration the shopper’s behavior.
As of now, we have out of the box support for
We have successfully worked with other platforms too such as WebSphere, Lightspeed, etc. however, it may need customized work.
The overall process of onboarding consists typically of 4 distinct steps.
Step 1) The process starts with the design of the proposed recommendation widgets
Step 2) The second step is data integration. The product catalog will be fed into our platform. The easiest way is through APIs. Also, we may need your historical data for hot-start. Historical data can be shared in XML or CSV format.
Step 3) The third step is to set up a meeting with the customer technical team. The first objective of this meeting is to place the listener file into the head of the e-commerce site. The second objective is the setup of the recommendation widgets.
Step 4) Once the customer validates the rendering of the recommendations widget in the test environment and customer, we can decide to go live.
We would like you to ask as many questions as possible.
For a new shopper, buying a specific product, the system maximizes the probability of an upsell, by recommending additional products based on data from previous shoppers that have bought that same product, and subsequent additional products.
Each individual shopper of your site will be allotted a unique user id. In the case of a logged-in shopper, he will be mapped with the customer id and unique user id. In the case of a non-logged-in shopper, he will be identified with a unique user id. In case he logged in later, the cookies are mapped back to the customer id.
Yes, it works. Sometimes e-commerce managers want to push certain products within certain recommendations. It is therefore usually important to have the option to manually insert chosen products within a widget to fulfill your business objectives.
For customers who have not signed, the system considers them to be new shoppers.
For a shopper who has signed, sophisticated algorithms are applied, recommending products, based on the shopper’s past purchases, views, and clicks, as well as their browsing patterns in the current session.
If we train the recommendation models on your historical data before the implementation you can be hyper-effective from day one of the implementations. If that is not the case then it might need around 2 to 3 weeks for our algorithms to start becoming effective.
The home page is like the shop window to your site. Having recommendation widgets that match visitors’ intent is critical in conversion and provides a great shopping experience. Some might know what they are looking for, and others are just browsing, and your product recommendations need to cater to both.
Trending items
Works well on a home page for new as well as returning visitors. Products are recommended that are most sold recently. These are based on purchases made by all visitors. The benefits are:
Likely next buy
The likely next buy algorithm takes all historical customers’ buying patterns into account. So, when a specific returning customer visits the site, the algorithm first determines the products that this visitor has bought previously. Then it looks at all other customers to determine which customers have also made the same purchases. Finally, it then determines the most likely additional purchases that customers have made, on top of the initial purchases, also made by the current visitor. In other words, if you have bought two CDs previously, and I have bought the same two CDs, and I subsequently bought the third CD, then you’d also be likely to want to buy that 3rd CD. So, that 3rd CD is then the likely next buy for you.
This highly personalized algorithm works great on the home page because it’s like the shop clerk that says to the customer that enters the store: “Hey Mr. Jones, welcome back! I know which clothes you have in your closet, that is why I think these additional clothes would go great with what you have. I know, because all our other customers have liked it as well.”
Based on your browsing history
This personalized recommendation algorithm recommends products based on previously viewed and bought products. From these products, the categories, brands, or other product attributes are identified. Products are then recommended that relate best to these attributes. This greatly improves the customer experience because the search time and effort to discover products associated with the visitor’s intent is minimized. It increases the number of products in the basket because it will trigger conversions on latent needs.
It fits well on the homepage because products are recommended based on interests the visitor has portrayed before. It is specifically relevant with a large highly varied assortment.
Table stakes recommendations
Of course, when a visitor lands on the home page, it’s a must to assist him in picking up his previous session, by offering an overview of his recent views. Also, when you’re providing your visitors to keep track of the product they like, by maintaining a wish list, it’s good practice to give them easy access to that wish list. An even better practice is using the information from your visitors’ wish list to personalize your recommendations by taking into account the attributes of these products as visitor signals in the visitor profile.
When landing on a category page, your visitors probably know what they are looking for. A category page is basically a homepage for a subset of your assortment. All the principles applied to the home page also apply to the category page, albeit for reduced product scope. Sophisticated recommendation systems allow you to filter recommendations into specific categories.
So, the best practice here is to apply the same algorithms to the category page, as you would do to the home page, and then filter the recommendations to that specific category.
The most crucial page of the site is the product page, as this is the gateway to the actual conversion. Recommendations of products that present perfect complementary selection to the current product, can lead to additional sales, increasing the number of products in the basket. Recommendations that present the right alternatives to the current product either reinforces the customer’s initial choice or give the customer options to choose from. If these complementary selections and alternatives are also matching the customer’s intent, the probability of conversion is maximized.
Frequently bought together
When a power drill is often accompanied by a specific set of drills, bought together in the same session, and when this combination is bought by a high number of customers, it makes sense to suggest this combination to other customers as well. So, on the product page of that power drill, the best practice is to recommend additional products, such as that set of drills, alongside the power drill. So, the approach to presenting complementary selections on the product page, that go well with that product, is to identify which products are often bought together.
These products can then be recommended in a bundle, like the top of the image below, or alternatively in a carousel of multiple recommendations. For example, a carousel can be applied to give the customer the choice from a set of complementary batteries to an electronics gadget. The carousel is visualized at the bottom of the image below.
Frequently viewed together / people also viewed
Proper alternatives to the product presented on the PDP can be identified in several different ways. Viewed also viewed is the first way to do this. When a visitor views products within the same session, it suggests that these products are alternatives to each other. So, when the viewing behavior of all visitors is analyzed, and combinations of products emerge that are often viewed together, these combinations are proper candidates for alternatives.
Product attribute-based recommendations (content-based recommendations)
Products that are alternatives to each other often share the same brand, specifications, product names, or other attributes. So matching products together based on these attributes is a powerful way to identify alternatives. This is specifically true for categories with products that have a lot of specifications, such as consumer electronics, or power tools.
Cutting edge Natural Language Processing and Natural Language Understanding algorithms can match products based on these attributes. The more the algorithms are trained for a specific category, the better they perform. That is why even Amazon has a hard time applying this approach because of the variety of different categories on sale on its site.
It’s also sometimes referred to as content-based recommendations, as it matches combinations of products, based on the content associated with these products.
Category based recommendations
A third approach to recommending alternatives is to simply recommend a product from the same category. While this approach is rougher and simpler than the previous two if categories are defined narrow enough it may work very well. In the example below, smart plugs and lights are the categories from which alternatives to the selected smart plug are presented.
When you check out at an outlet of a savvy retailer, you will find baskets or even a full maze of products before you reach the counter to pay for what you’ve bought. Customers, while waiting in line and while checking out at that physical store can be easily seduced to throw a couple of cash grabs into their basket. How can we achieve the same result in e-commerce?
Cash grab recommendations
To achieve this, of course, you must place recommendations on your cart page. To achieve the cash grabs temptation, you may choose to present specific categories of cheap seductive products, such as cables, batteries, or other accessories. From these categories, the cheap trending items can be recommended for maximum conversion probability.
Accessories on the cart page
It becomes even more powerful when you base the recommendation on the products that are already in the basket. So, when you recommend accessories that are cheap enough to trigger an impulse buy and are a perfect match to what you’re buying already, your conversion chances are highest. To achieve this, you can apply a “frequently bought together” algorithm and filter the results to specific categories and or to a maximum price limit. That price limit is dependent on the assortment of the shop.
Add to cart button
Now, you must be careful to not distract your visitor from his journey to the payment gateway. So, your visitor should not be driven backward in the buying journey. So, clicking on the recommendation on the cart page should not result in opening a new product detail page. Instead, your recommendations on the cart page should always be accompanied by an add-to-cart button. For fashion, this brings the additional complication that the add-to-cart button should include an option to select a size variant.
We replace your existing rule-based widgets with our high-performing AI-based personalized widgets.
The key factors for personalized product recommendations to work well are visitor traffic and product assortment.
The more visitor the better, and the more visitors come back regularly the better. Returning visitors allows the system to create and maintain an insightful visitor profile that can drive the personalization of future recommendations.
The more products in the assortment and the more variety within the assortment the better.
As a rule of thumb, e-commerce sites selling at least 1,000 products with traffic of at least 20,000 unique visitors per month benefit well from recommendations systems. Of course, this does not mean e-commerce sites with smaller numbers cannot benefit from recommendation systems, but as argued above, the more data there is, the better the recommendations become.
Implementation can take between a few hours and a few weeks, and it depends on factors like the e-commerce platform, the degree of pre-built integration with the platform, and the complexity of the product catalog.
Typically, integration addresses these four areas:
The product catalog needs to be updated very frequently to ensure a proper account of stock levels and prices. API-based integration is best for this, but a frequent sharing of a product feed is also possible.
Historic transactions can be integrated by sharing a simple feed, or through an API key.
Clickstream listening, enabling the tracking of visitor behavior on site is often done through a listener file, that is triggered by an inserted div tag within the e-commerce platform.
In rendering, there are different approaches, where JavaScript div tag insertion is a no-code easy integration, and GTM based integration of the widget rendering is an even easier way. Div tag insertion performs slightly better than GTM based integration, so that is, in the longer run, often the preferred option.
We copy the CSS files of your e-shop website and replicate the same in the recommendation widget to ensure the look and feel of your e-shop.
We prefer getting changing product assortment from you. We also use the JavaScript code we embedded into your website to reads changing products assortments.
We use the JavaScript code we embedded into your website to read the products that are out of stock and update the recommendations to not include out-of-stock products.
The following is required from you
We don’t touch your codebase. What we need is to put JavaScript on the website and wait for the magic to happen.
Our clients have experienced up to 25% increase in their revenues coming from our product recommendations, up to 25% increase in basket size, and up to 25% increased customer lifetime value.
If recommendation algorithms break down, at most, the website will revert to the default website. Failing of recommendation algorithms will not break down your website.
We provide both product recommendations as well as search. And our unfair advantage is that we have a uniquely powerful engine, in our AI solution, that will simply perform better. Also, our console and integration approach is simpler than the competition.
Breadth – both search and recommendations, for all e-commerce platforms, within recommendations we have a large number of possibilities (eg different algorithms)
Depth – our AI engine is truly cutting-edge; assisted and unassisted machine learning, deep learning, natural language processing, ML-based ranking, embeddings, neural networks, to name a few different AI technologies.
We serve our customers with obsession and a conviction that will lead us to more sales and more renewals. It means that we will go the extra mile to satisfy our customers. This shows for example in our satisfaction guarantee and our performance-based price models.
We offer a 3–months, not happy money–back guarantee, without restrictions or conditions.
We also offer ROI–based pricing, meaning we only get paid if and to the extent that the customer is making more sales.
Yes, your system comes with a console where you can manage your engagement with the system.
Yes, you can manage a lot of things. For example: which widgets to keep live on your site? Which widgets should be shut down? Customize the look and feel of the widgets, etc.
Your console tells you exactly how your widgets perform and why. It provides insights that allow you to determine where the system needs to be tweaked? Which widgets placement should be adapted to get more views? Which algorithms should be adapted to incite more clicks?
Yes, you can. Product recommendations widgets can be integrated into your search bar. It is possible to use search queries as an additional signal to make recommendations even more powerful. Recommendations can be made to appear on the search bar while the search query is keyed in. These personalized product recommendations are effective as they are related to the query. As a result, conversion increases significantly because visitors can find perfectly matched products directly from the search, rather than moving from search through several further pages before they find the ideal product.
Yes, you can. Inserting personalized recommendations into your email campaigns allows you to hyper-personalize your campaigns and boost the conversion of these campaigns. Email campaign systems segment visitors’ data for campaigning, while hyper-personalized recommendations promote the products that the visitors really care about. This powerful combination assures maximization of conversion.
Yes, you can. Customer Data Platform (CDP) aggregates and organizes shoppers’ data across a variety of touchpoints to get a 360-degree view of the shopper. A recommendation engine by mapping the 360-degree view of the shopper from the CDP to the product assortments assures maximization of the conversion.
Yes, we provide training for your technical team. Some of the things that are covered in the training process:
We humbly serve our customers with obsession and a conviction that that will lead us to more sales and more renewals. It means that we will go the extra mile to satisfy our customers.
