Resources
Attribution 101: Tracking the Performance of Product Recommendation Widgets on your E-commerce Website
Written by
Neeraj Manivannan
neeraj.manivannan
1

Attribution 101: Tracking the Performance of Product Recommendation Widgets on your E-commerce Website

Modern-day customers seek personalized recommendations at different touchpoints across their journey. Recommendations are tailor-made to their tastes and preferences. Since these customers are present on multiple channels, e-commerce brands have started adopting an omnichannel personalization approach which enables them to customize the journey of their customers across platforms, at each digital touchpoint. 

One of the major challenges for brands targeting customers – across multiple channels – is to analyze and identify which action made by the customer led to the completion of a conversion event. 

This is extremely important to measure since brands need to track key metrics like Clickthrough Rates (CTRs), Bounce Rates, Add-to-Cart Rates, Conversion Rates, and Return on Investment (ROI) for each campaign or recommendation widget deployed on the website.

This challenge becomes even more pronounced while trying to compare the performance of two touchpoints on the same channel. 

When the customer encounters two or more touchpoints in the process of completing the conversion event, how can the brands identify which action by the customer ultimately led to the conversion event? This is where Attribution comes in. 

Attribution is the practise of evaluating marketing touchpoints encountered by the customer on the path to conversion. The goal is to determine which touchpoint or event had the maximum impact on the conversion or to take the next desired step. It helps differentiate the performance of two or more elements placed together on a particular platform or channel.  

Let’s consider an e-commerce fashion and apparel website on which the marketing team has deployed numerous widgets on different web pages. To purchase a particular product, different customer journeys can lead to the final conversion event of a purchase. 

Let’s consider the following case:

The customer finds a product interesting on the recommendation widget – ‘Related to Items You’ve Viewed’ on the Home Page and clicks on the product. 

On the Product Display Page, the customer scrolls down and sees the recommendation widget – ‘Products Related to this Item’ and clicks on a product in that widget. The customer then goes ahead and purchases the product. 

The question that arises now is – how does the e-commerce brand decide which of the two widgets gets the credit for the completion of the final event (product purchase)?

That’s exactly the dilemma, we at Netcore, are helping e-commerce players that deploy omnichannel personalization address!

What are the common types of attribution models?

Since both the First-Touch and Last-Click models fail to account for the broader customer journey and the other interactions the customer would have had prior to the final conversion event, brands should ideally avoid relying solely on one of these methods.

To avoid this, brands need to look at multi-touch attribution models like the Linear model to ensure that they look at all the touchpoints that lead to conversion and give equal credit to each of them.

Build your attribution models on a solid foundation of data

To achieve the level of data granularity for effective attribution, brands need to be efficient at tracking customer-level data and also at measuring the performance of the deployed recommendation widgets.

This requires gathering customer activity data that is used by the attribution model. There are primarily two types of activity data, namely:

  1. Interaction-level Data:
    The data that is tracked includes data of events that performed by the customer –
    The parameters that are returned are the Customer ID (could be customer or session ID, which is a unique user identifier), Product ID, interaction type based on the customer’s action (view, add-to-favourite, add-to-cart, purchase), and timestamp of the completed event.
  1. Recommendation Widget-level data: This is the interaction data at an individual widget level that includes data about products that have been seen and clicked upon by the customer on a particular widget.

The parameters returned are seen/clicked type, Customer ID, Product ID, name of the widget, and the timestamp of the completed event. To understand if a particular widget contributed to the completion of an event by the customer, the interaction-level gets mapped back to the recommendation widget-level data to prove that a connected sequence of events occurred that led to the conversion. 

But how can e-commerce brands track and identify which widgets need to be attributed on the completion of an event?

Let us consider a customer on an e-commerce website to illustrate how attribution is chosen:

  1. The customer arrives on the Home Page of the website and clicks on a Home Page recommendation widget (called “Related to Items You’ve Viewed”) and clicks on the Blue Avengers T-shirt:
  1. In the resulting Product Display Page (PDP), the customer scrolls down and clicks on a product on the PDP widget (“Products Related to this Item”).         
  1. The customer then adds the clicked product to the shopping cart; i.e. the Yellow Avengers hoodie:

In this complete journey, there are two click events (the Home Page recommendation widget and the PDP recommendation widget) and one Add-to-Cart. The Yellow Avengers hoodie is the product that has been added to the cart from the PDP widget. 

The question that arises is which widgets need to be attributed for the completion of the event (purchase of the yellow hoodie) by the customer?

The Attribution Process

Attribution is a 2-step process, with parameters in each step. 

1. Selecting the candidate widget by determining if a widget played a role in the conversion:

In this step, a conversion event is mapped back to a list of widget-level click events by the same customer. If certain criteria are matched, the widget is deemed to have played a role in the conversion and is considered a candidate for attribution.

The criteria depends on the lookback period (i.e. how far back can we consider widget clicks from the timestamp of the conversion event) and the method of candidate selection of that widget, of which there are two:

  • Same-Click Attribution: If this method is selected, the widget is considered a candidate only if the Product ID of the conversion event matches the Product ID of the widget click event. If the Home Page widget configuration is set as “Same-Click”, it isn’t considered a candidate as the Product ID differs. The PDP widget will only be considered as a candidate since the PIDs match
  • Any-Click Attribution: In this model, the Product IDs of the widget click and conversion event need not match if the widget configuration is set as “Any-Click”. Both the widgets – Home Page and PDP will be considered if they had “Any-Click” configuration.

Once a list of successful candidate widgets have been drawn, we move on to the next step in which the attributes are assigned credit.

2. To determine the widget that deserves the final credit from a list of candidate widgets that were selected:

Here, in a list of candidate widgets, we determine which of the widgets gets the final credit and is added to their conversion stats. The e-commerce brand needs to choose the attribution model that they wish to consider to give credit to each of the contributing widgets.

  • First-Click Attribution: The earliest click event is considered. Here, if the Home Page widget had “Any-Click” configuration and is a candidate, the Add-to-Cart event would be attributed to the Home Widget as its click event occurred first
  • Distributed Click Attribution: The Add-to-Cart event would be attributed equally as 0.5 and 0.5 to both widgets, i.e. the Home Page widget and the PDP widget
  • Last-Touch  Attribution: The Add-to-Cart event would be attributed to the PDP widget as that was the last clicked candidate before the Add-to-Cart event. The Home Page widget would not be attributed

Choosing the right attribution model 

We’ve seen that a lot of consideration needs to be given when selecting the best-fit attribution model. As we have seen earlier, just choosing one of the models might not generate the best and most accurate results. 

To get the most reliable insights, e-commerce brands would need to use a combination of models to determine the right event or recommendation widget to attribute for the success of the conversion.

Getting the attribution model right is extremely important as it assists in assessing key metrics like conversion rates and marketing ROI.

Measuring and understanding the performance of the widgets deployed across the e-commerce website helps gain valuable insights on customer behavior. It also helps brands make better-informed decisions to achieve their business targets. To understand how your e-commerce platform can get started on your personalization journey with product recommendation widgets, get in touch with us today!

Unlock unmatched customer experiences, get started now
Let us show you what's possible with Netcore.