Figure 1: Understanding the stages of customer journey based on browsing signals to curate recommendation strategies based on journey stage
Here’s how we can curate recommendation strategies based on the journey stage –
- Awareness and engagement stages: Brands can adopt recommendation strategies with a higher exploration component when the user is in the awareness and engagement stage. Strategies based on what similar users like, can result in products across diverse categories that will resonate with the user. Other strategies that will be effective in generating engagement are top sellers within a brand or a given category, and “known affinity” based recommendations across diverse product lines
- Consideration stage: Brands can adopt more exploitation and less exploration at this stage and recommend items similar to the one in consideration, items frequently bought together, or even “known affinity” based product variations within the same product line
- Purchase stage: It is common best-practice to go for extensive cross-selling at this stage, however it may fail to engage with a price-sensitive consumer if their cart size is already higher than their average order value (AOV). At this touchpoint, brands can explore content-based engagement, such as “Did You Know” facts about products and brands in the cart.
Mistake 2 - Failing the explore-exploit trade off
Is recommending a best-fit product for a shopper, always a good recommendation strategy? How often should we “exploit” product recommendations from historical intelligence? In the context of ecommerce, it’s very hard to answer this question unless we capture a shopper persona’s intent, as they might be impulsive, exploratory and looking to discover new content. Let’s take cooking for an example - some of us have explored cooking to various depths during lockdown, while others just cooked for the need of it, instead finding passion somewhere else. A need-oriented consumer, trying to fulfil an immediate need of having a meal may look at a recipe of “Garlic Baked Potato”. An exploratory persona too, may also look at the same video while exploring recipes out of pure interest.
Things get interesting when they come back to YouTube, only to see their home feed is full of recommendations on how to make similar recipes based on potatoes, eggs, garlic or baking. Such content will be useful for the persona who was exploring recipes, but useless for the need-based persona who had looked up the recipe to fulfil an urgent need of cooking something to eat. He may not be interested to see another recipe video anytime soon.
Mistake 3 - Merchandise context insensitive
Another mistake is when recommendation engines are not aligned to inventory clearance policy. Consider the example of “Similar Products” or “Related Products” in a Product Detail Page, and “Frequently Purchased Together” recommendations at a Cart Page. Here, products having higher inventory should be boosted and ranked higher w.r.t products having low inventory, else high inventory items will often be hidden from the consumer resulting in poor turnovers. This gets more relevant in a post-COVID era and during Holiday seasons, when brands resort to heavy mark-down losses to promote and clear unsold inventory.
Mistake 4 - Failing to capture retail sector context
Marketers, industry experts and data scientists need to collaborate to see if algorithmic product recommendations are aligned to the industry context. For e.g. in Healthcare, product recommendations need to be strongly aligned to the buyer’s need. There are challenges with a few recommendation strategies, as explained with a couple of examples below -
- Products based on what similar users like
- Products frequently purchased together
An anxious consumer looking to treat flu, will be very confused if a “People like you also bought this” recommendation results in diarrhoea treating medicines. This can happen because consumers with similar purchase history could have purchased diarrhoea curing medicine when they had one. Machine learning driven Product Associations and User Based Collaborative Filtering, will often generate products belonging to a different category, whereas “Similar Products” will be more accurate.
Mistake 5 - Recommendations fatigue
Ecommerce companies need to be sensitive towards their customers and avoid fatigue due to recommendations. Too many recommendations can confuse consumers, resulting in a paradox of choice. Imagine a visitor who lands onto a Product Page and sees multiple recommendations like “More Like This”, “People who viewed this, also viewed this”, “Frequently bought together”, “Top Sellers” and also “New arrivals in this product line”. It’s very likely the user will bounce off from the page. According to a survey, visitors who were recommended with less products had a 30% conversion rate, whereas visitors presented with too many choices had a 3% conversion rate
New opportunities in the new normal
As we embrace the new normal, it’s the best time for retailers to revisit their recommendation strategies in order to engage with shoppers more effectively. To make product recommendations more relevant to customer journey and industry sector, retailers need to have a good mix of data scientists, marketers, and domain experts co-innovating together on how to optimize recommendation strategies. A/B testing recommendation strategies at various touchpoints is an effective way to optimize conversation rates. We will discuss the topic of “Optimizing recommendations along the customer journey" in more detail in a separate article.