eCommerce Blog

eCommerce Blog – Opportunities in US and India (by Darpan Munjal)

Personalization & Product Recommendations

Posted by Darpan Munjal On April - 28 - 2008
1 Star2 Stars3 Stars4 Stars5 Stars (4 votes, average: 4.75 out of 5)
Loading ... Loading ...

The unorganized retail business in India thrives on personalization. You walk into a local kirana (grocery) store and the store owner greets you with your name, asks about the family and then gives you a customary 5% discount for being a long time customer. He also knows the brand of tea you like and recommends you a new brand that just came out and then chats with you for few minutes about last night’s 20/20 cricket match. You end up making some impulse purchases while you are chatting and walk away happy with the overall experience. This level of personalization is a dream for the nationwide organized retailers, and although it may seem simple in a neighborhood grocery store, scaling it to a big-box retailer, with hundreds of employees and thousands of customers a day, is a different story.

But retailers are trying. As the retail environment gets more competitive, physical retailers have started shifting from a product centric to a customer centric mentality to connect with their customers. A large number of retailers now make the assortment decisions for what products to carry in their stores based upon the local demographics of the consumer visiting the store – as opposed to having a “one-size-fits-all” assortment for all the stores across the nation. Despite the serious constraints that are working against them such as supply chain and difficulty in capturing customer insights, the physical retailers are spending significant effort and resources in personalizing their strategies.

What about the online retailers?

One would think that things would be better in case of online retailers – where the customer insights are flowing in with every single click and the technology is available to act on these insights almost on a real time basis. While personalization seems easier to accomplish in that scenario, there are only a handful of online retailers that have been able to come close to offering a personalized and unique experience for their customers. Personalization is not a new technology, but it is generally treated as a step child by most online retailers. A recent survey from Forrester Research found that just 16% of e-retailers use personalization techniques effectively for merchandising. And No, by personalization, I don’t mean the welcome message that you see when you visit an online retailer’s home page or the “My account” section where you can check your order history. I am talking about the capability which goes above and beyond the obvious – where a visitor’s implicit or explicit interactions with the website are used to build a unique shopping experience and relevant product recommendations for her. Easier said than done, you say. We all would like to get to that level of nirvana in personalization – but the question is does the technology investment and time required to get to that state justify the results in the end? May be. May be not. I certainly don’t recommend that online retailers start making large investments in personalization technology and then see if it sticks with the customers. However, I do think that it is important to take incremental steps so that the customers feel more connected with the e-tailer.

Building relationships..

The key objective of personalization should be to build deeper relationships with the customers. In today’s environment, a large number of online retailers interact with their customers on a transactional basis. This kind of interaction with the customers reminds me of the movie “50 first dates”. Retailers almost start over fresh with their visitors everyday, and completely forget about the interactions with the same visitors during previous visits. Then there are a large number of online retailers, who do take the extra step of making product recommendations to their visitors. However, unfortunately, a lot of times these recommendations are pushed based upon internal rules defined by merchandise managers instead of the customer insights. When we make a recommendation to the visitors, we are almost making a statement that we know something about their behavior and our recommendations are a reflection of that knowledge. Therefore making wrong recommendations is a risky proposition, because that could rub individuals the wrong way – case in point who had to issue a public apology and temporarily shut down its entire cross-selling recommendation system when customers who looked at a boxed set of African American films that included “Martin Luther King: I Have a Dream” and “Unforgivable Blackness: The Rise and Fall of Jack Johnson” were told they might also appreciate a “Planet of the Apes” DVD collection, as well as “Ace Ventura: Pet Detective” and other irrelevant titles.

On the other hand, helping customers find the right product at the right time has always been the dream of online retailers because they know that achieving that level of capability can significantly improve the conversion rates and the stickiness of the site. has taken on this mission and has certainly come a long way in offering recommendations based upon the collective insights captured from the consumers. The question is – are they effective? I have a lot of friends who really love Amazon’s recommendations – a lot of them think it is cool to see what else other people have purchased after buying a particular product online. I do think it is pretty cool from technology perspective, and I do think the recommendations are helpful – but I personally feel that Amazon’s recommendations have too much of “group think” factor which falls short of taking into account individual’s unique tastes, likes and dislikes. I also think that Amazon has gone a bit too far in flooding the product page with different types of recommendations– which can overwhelm and possibly further confuse visitors in their decision making.

Recommendation Strategies

So what are some of the things online retailers can do to connect with their customers?

First of all, there are rule based recommendations. This is where the online category managers build rules to associate different items that go with each other. For example, if you are looking at a DVD burner, a rule could be a setup to recommend you a 50 pack of Sony DVD-R. This is one of the most common cross-sell recommendation techniques used by online retailers – the only downside is it requires significant manual effort to setup and it is based upon the internal decisions of a single category manager. The technique might work effectively for a highly niche retailer with small assortment, but is not quite scalable and effective if you have relatively large catalog.

The next level of recommendations start taking into account the “aggregate wisdom” collected from all the customers. One common recommendation technique is to start showing “Top Selling Products” in all categories, sub categories and search pages. A lot of time customers need help with item discovery process and although they might not trust what a retailer is pushing to them in the form of recommendations, they do take other visitors’ opinion seriously. Seeing a product listed high in top selling items can give the visitor a vote of confidence – because they know they are not making a really bad decision if most other visitors have purchased the same item.

An equally effective but less frequently seen capability that relies on “aggregate wisdom” is “Top Rated Items”. This is where visitors are recommended products based upon the ratings received from other customers. The products that receive highest aggregated ratings in any given category or sub category start appearing as recommendations for visitors when they visit those categories or subcategories. Some e-tailers have actually started adding more predictive attributes in the ratings, to help other individuals make better decisions. For example lets visitors provide additional insights into ratings that can be helpful for other visitor’s decision making. In the below example, 24 users have not only rated the product but have provided tag based information on their ratings so that the information is much more meaningful to someone else who is looking at this aggregated information. For example 16 out of 24 reviewers think that this toy is suitable for imaginative play, an insight that might greatly help other visitors who are looking to purchase toys for imaginative play.

The next extension of “aggregate wisdom” recommendations is collaborative filtering. This is where a highly advanced algorithm does statistical match across different product combinations – and the recommendations are made purely based upon the affinities across various products. The good thing about collaborative filtering is that it minimizes the level of manual involvement from category managers and automatically finds patterns of product combinations based upon aggregate purchase behavior. This is what powers “People who purchased this product also purchased….” type of recommendations at Amazon. Although these type of recommendations require a relatively high implementation cost, they minimize the need for setting up manual business rules and can run on a “Auto pilot” mode to some extent. The downside with collaborative filtering recommendations is that they tend to produce “Averaging effects” which causes the overall most popular items to be recommended more often which means that they will be consumed and rated more frequently as a result.

Stages of Recommendations

The Expert Factor

What is missing in all above types of recommendations is the notion of expertise or authority. All visitors are treated equally in determining the recommendations. Although it is good to see statistically relevant recommendations based upon a democratic way of collecting insights, humans by nature would take recommendations more seriously if they are coming either from experts in the category or from other individuals who have similar tastes. For example, a serious audiophile visiting may not be interested in the best selling home theater systems on the site. This is because the best selling home theater systems are more likely to be the ones that are purchased most by the average demographics, who tend to prefer mid-range audio equipment. This person would rather like to connect with and accept recommendations from other individuals who are experts in high end music gear. This is where the online retailers should start thinking about adding more context to the recommendations and reviews. The key is to identify as many predictive attributes as possible (e.g. Age group, interests etc.) and then connect the visitors with recommendations from other individuals who have similar attributes. This is the true power of online recommendations – where customers are able to get advice and recommendations from others like them – who they can trust more than the average crowd. The ToysRus example above accomplishes some of that through customer ratings, but there is significant opportunity to do more in this space. As pointed out in some of my earlier articles, social shopping sites such as are recognizing this aspect of context based recommendations and are helping visitors obtain recommendations from style leaders or experts in different categories. This, I believe is the future of online recommendations, and those retailers who are truly able to unlock the potential will evolve as leading destinations for customers in their journey of online purchases.

Like this post? Share with your network:
  • Twitter
  • Facebook
  • LinkedIn
  • Google Bookmarks
  • Digg
  • Blogplay
  • Add to favorites
  • IndianPad
  • StumbleUpon
  • Technorati
  • Reddit
  • Sphinn
  • Tipd

4 Responses to “Personalization & Product Recommendations”

  1. Srinagesh Eranki says:

    Very insightful article. Been enjoying this series.

  2. Insightful!

    Couple of questions though:

    – How much of an impact do each of the above recommendation techniques mentioned really have? Is it an expensive distraction or a true value add?

    – If each e-tailer starts to implement similar looking recommendation system, what is the differentiator from an online shop perspective and customer experience or value.

    – How far have the major indian online shops gone to have such a system in place?

  3. Sandeep says:

    Interesting article Darpan.
    Another area which is closely linked with online commerce is Multichannel. Currently most of the retailers are busy setting up their online ventures with basic features because of which there is hardly much attention payed to multichannel experience.
    The various channels can be Web, IVR, Kiosk and the Brick & Mortar Stores. A seamless multichannel experience is by which a user cannot diffrentiate any channels in terms of his user experience or the data presented to him.For example if I go my local Subhiksha store and find my vegetable oil brand to be out-of-stock, the same can be checked on the web itself. Another example is if I find a good camera on a site like and then talk to a Customer Service Rep for the same he should be able to recommend cross-selling of items like camera bag just the way the website will.
    To achieve multichannel integration is definitely not an easy job and it calls for a integrated common IT landscape which is accessed by all channels. The site which comes closest to achieving a perfect multichannel synergy is Even though it is rated as the best site it has still some way to go – When evaluated, it does not still support the basic user goal of buying and checking status across all the channels.

  4. salonee says:

    I am a student pursuing my MBA & wanted to know of an E-Commerce Initiative you would like to create,it can be a new company, a web site or a new way for an existing company to use E-Commerce.It would be great to receive feedback from your end.
    salonee shukla

Leave a Reply

Get Adobe Flash playerPlugin by wordpress themes