Implementation from scratch
Before the implementation
To implement personalization from scratch, embark on a trial period with a recommendation engine provider. Nowadays majority of the recommendation / personalization engine providers offer a trial or proof of concept period which enables e-commerce owners to harvest the first fruits of personalized recommendations or test them against inhouse solutions.
If your e-store is SME-sized or has a simpler structure, look for a solution that offers a performance-based pricing, where the costs will appear as a small percentage of the revenue you make with the help of the recommendation engine.
If your e-store is enterprise-sized or has already used and deployed personalization solutions, a ‘proof of concept’ - period is recommended that offers you the possibility of full customization and support. During this step you will be able to target specific KPIs, A/B test different providers or see how a personalization engine can boost your metric compared to your inhouse solution.
Upload initial data
To ensure seamless performance from the beginning your e-store has to upload different types of data to the recommendation engine provider in order to connect it into the engine.
- Item catalog - recommendation engines require the list of the recommendable products and some additional information like category, name and price. The easiest way to provide this data is to upload a .csv or .xml file, and make it available to periodically download from HTTP, FTP. In case your item catalog changes frequently (for example a classified site) then real-time item catalog updates are required using server side APIs.
- User events - the most important part of recommendation algorithms is based on collaborative filtering. This is an online process and historical data is not required but the logging user actions should be implemented.
After this the systems is ready to provide your users personalized recommendations. The only thing you have to do is display the recommendations on your site and to keep our event database up-to-date by tracking the user events and forwarding them to the recommendation engine.
There are two different way to accomplish this:
- Server side - If you are looking to implement the widget displaying the recommendations yourself, then you may opt for the more secure, server to server communication method.
After the product feed is linked in, the first item-to-item recommendations start to be presented. When user events are deployed, they also will be assessed and the collaborative filtering - based algorithms also start working and send the first recommendations.
The importance of automated data collection and personalization can not be highlighted enough: with automation overhead costs will be reduced, time is saved up for sales and marketing activity to convert prospects to quality customers.
After the first days - Recommendation types in order of appearance
Product pages: Based on the item-item similarity logics product page recommendations can be provided right after the product feed is integrated. Cart Page Recommendations can display offers with the aim of upselling or bundle-selling right after the product feed is integrated and the similarity map about the products is created, the systems learns buying habits based on the user events Home Page also can display recommendations like latest deals and discounts, personalized recommendations can be displayed for users with existing browsing history.