As an e-commerce manager or digital merchandiser, you'll all too often encounter the challenge of choosing the optimal recommendation type for your commercial objective. Because let’s face it: it’s a jungle out there!
The good news is Apptus eSales has a recommendation algorithm for every step of the customer journey. For example, options include recommendations based on individual customer, cart contents, abandoned cart contents, and top-sellers.
In this blog I want to give you a brief overview on how to pick the right recommendation algorithm. The cheat sheet below offers a summary of the decision process you’d usually face:
Now let me explain:
Apptus eSales delivers recommendations at the product level
The product level can be seen in Search or Navigation listings - “Skirt” for example. Apptus eSales will list the products in the order they are most likely to convert, given the context, example: “skirt”, “Jeans”, “Shirt”. The variants returned for each product will be listed in bestselling order, example: red, floral, black.
What is a cached area and why should I care?
If the area is cached, you are reusing an eSales result multiple times, which means that you should avoid personalisation. If you want to personalise an area, that area needs to be fetched fresh for every customer.
Personalised recommendations vs recommendations based on products
This depends on context: On a start page, personal recommendations make the customer feels that ‘this page knows me’. However, a product details page is a customer’s current context and at this moment you do not want to digress too much from that context. Therefore, you want recommendations that add here to the context as much as possible.
On a product detail page, recommendations tied to that specific product are more relevant because this is the customers context. Personalised recs should be used when you want to recommend from a vast number of products and Recommend based on Product is when the context is tied to that specific product.
When should I consider retargeting?
On the start page or in a “my pages” context. When you want to give a user a friendly reminder of the products they have shown interest in, but haven’t bought. This is quite easy, and it does convert well in the right context.
When is the right time to use a Top Seller strategy?
When you have a trend-sensitive audience and want to make sure that your site makes the most of sudden mini-trends but you want something smarter than just a statistical top-list of which items have sold most in the last week. Trends and customers move faster than your statistical top list, so use a top seller strategy to catch those trends site wide, app wide, in specific categories, across all campaigns.
To help you put this into action, I have devised a three-step checklist:
Fine tune, verify, measure!
A) Fine tune
I recommend you try to answer these questions:
- Which products do you want to exclude from the recommendation area, e.g. items on sale, items out of stock, etc.
- Is there a need for deduplication with other panels on the speciﬁc page? Deduplication means avoiding having several copies of the same item on different panels.
- Are other products shown on the page, which need to be excluded?
- Not sure which algorithm to use? Perform an A/B test using competing panels and a split test. Remember to select one of the panels when you have enough data to pick a winner.
Do one or more exercises on your own to verify the panel setup:
- Click on product in the recommendation area
- Add the product to cart
- Purchase the product
- See the result in eSales Manager a few hours later in the Site tab.
C) Measure impact
- See Click Through Rate (CTR), Add to Carts and Purchases for a speciﬁc area in Apptus eSales Manager Site tab. A purchase will be tracked here if this is the area where the visitor ﬁrst interacted with the product.
- In the dashboard in Apptus eSales Apps, you'll see what proportion of total Conversions, Revenue and Profit comes from Recommendations.
I hope this blog has helped to clarify the choices for of recommendations.
 For example, in the fashion domain Product level connects to shirt, while Variant is given by colour (red, black, blue).