How Do E-Commerce Retailers in Kenya Make Use of Recommender Systems?

Data is changing the way things are being done not just in e-commerce platforms but in almost every other industry. Even though that is clearly a cliched statement, most companies haven’t fully moved to data-driven support systems probably due to a number of factors that we shall discuss in a later article. It’s however commendable to see how fast we are adopting BI tools like Tableau, Power BI and Looker. However, as we highlighted in our earlier article, the true power of data is realized not by descriptive analytics but by prescriptive analytics which is the basis of decision intelligence that we champion through our Insense platform.

One such area where we have done extensive research on is retail industry and today we focus on one aspect of retail that is highly overlooked by many: recommender systems. We were having a demo session for Insense when it hit me that this is something I always thought existed in most e-commerce platforms only to be surprised that most of them actually don’t have it so I spent time to sample the top 17 e-commerce platforms in Kenya to check if they had any form of recommendations.

My brief study’s objective was to find out how many e-commerce platforms have traditional recommender systems and how many have AI-driven (modern) recommender systems. Traditional recommender systems are not data-driven and most of them base their recommendations on product brands, categories, prices, colors et al. If you view a product in electronics category, they show you other products from the electronics category or if you viewed a laptop, they’d show you other laptop brands. Typically, they use the phrase: “you may also like” to show these recommendations. AI-driven recommender systems, on the other hand, use data mining techniques to find out associations in product baskets so as to come up with recommendations for users. The data-driven recommenders have gotten favor with the big companies owing to their accuracy and personalization that leads to extremely higher conversion rates as compared to the traditional methods. In traditional recommenders, if I searched a HP laptop, they would suggest to me that I may also like a Toshiba laptop (which is highly unlikely hence low conversion) but on the other hand, the data-driven recommender would note from the buying patterns that those who usually buy Samsung laptops also buy a lap top bag and as such, suggest to me a laptop bag – the language used is “Those who bought this also bought…” or “Those who viewed this also viewed….”

So how are the Kenyan e-commerce platforms fairing in as far as recommender systems are involved

  1. None of them have implemented a mix of both the traditional and modern recommender systems
  2. Only 4 out of the 17 use the modern recommenders
  3. 6 of them use the traditional recommenders
  4. 7 of them do not use any kind of recommender system
Plot of The Type of Recommender Systems Used by Kenyan E-Commerce Platforms
Plot of The Type of Recommender Systems Used by Kenyan E-Commerce Platforms. Source: IDTLab – Insense

Specifically, only Jumia, Kasha, Shopit and Mother Baby Shop have implemented what looks like the modern recommender systems. To my surprise, Naivas, Sky Garden, Jambo Shop and Kilimall fall in the e-commerce heavy weights that do not implement any form of recommender system that I could identify.

Data-driven recommenders use data mining techniques to find out associations in product baskets so as to come up with recommendations for users

Why exactly is it important for retailers to implement modern recommender systems? By this question I do not just mean e-commerce platforms but all retailers including brick and mortar stores. Whereas only e-commerce platforms can implement recommender systems on their platforms, both e-commerce platforms as well as brick and mortar stores can use recommender engines to send out recommendations either via e-mail or SMS campaigns to their customers. does recommendations both on-site and off-site through emails. A brief look into the data from Amazon will reveal the amazing returns:

  • 35% of’s revenue is generated by its recommendation engine
  • According to Sucharita Mulpuru, a Forrester analyst, Amazon’s conversion to sales of on-site recommendations could be as high as 60% in some cases based off the performance of other e-commerce sites.
  • Surprisingly, recommendations via email convert better than on-site recommendations

Amazon’s success is pegged on two things: data-driven decisions and customer obsession. All their marketing decisions are based on data and their platform layout and content is based on data. Let’s have a look at how Amazon has implemented recommender system on their site:

  • Recommended for you, “Phelix”: This is a curated list that shows recommended products for a user based on their buying history and there is a specific page for “Your Recommendations” on their platform to access this section
  • Frequently Bought Together: This section curates a list of products that users frequently buy together into a product bundle based on the buying behavior of products
Frequently Bought Together Product Bundle on
Frequently Bought Together Product Bundle on
  • Your recently viewed items and featured recommendations inspired by your browsing history: In this section, you can see all the products that you have viewed in the recent past and is very crucial to making one to ensure get their past browsing products and add them to cart without the hustle of having to search for the product again. Platforms like Kasha have this implemented while Kilimall have “Featured Recommendations” implemented.
  • Related to items you viewed: This is the section where they implement the traditional recommenders and show products related to the ones you viewed based on such parameters as brand, sizes, prices, categories et al.
  • Customers who bought this item also bought: This is a more specialized recommendation than the frequently bought together bundled products and is one of the best for achieving cross-sells and up-sells

  • There is newer version of this item: This feature allows one to see a newer version of the item. Could be modified to include “cheaper version” or based on any other parameter
  • Recommended for you based on a previous purchase
  • Best-selling in [category name]

It’s not a surprise that Amazon’s 35% of revenue comes from their recommendations based on such a comprehensive strategy. This benefit is not only seen in Amazon but also in Netflix where the company reports that:

  • 80% of the content that people watch are based on personalized recommendations.
  • The personalized recommendations produce $1 billion a year in value from customer retention

Recommender systems have wide usages and value and is something that your business should consider; not just in e-commerce but also in any other online platform: jobs lites, gaming sites, educational sites et al. Recommenders not only offer personalized experience to users but also make product discovery easier and faster, increase cross-sells and up-sells with the end result being higher click-through rates, higher conversions, higher sales and because of the improved customer experience, recommender systems are a massive plus to improving customer retention. Success of a recommender system, however, can depend on many factors, e.g., user trust, transparency, user interface et al. A study, for instance, showed that 30‐40 % CTR increase is realized with an adaptive algorithm and 100 % CTR increase is realized after changing the screen position of the recommendation widget.

Why recommender systems?: personalized experience, easier product discovery, increased cross-sells and up-sells, higher conversions, higher sales, improved customer retention

Despite these values, modern recommender systems aren’t highly adopted due to the complexity of the set up both in resources and expertise. A robust and fast implementation requires a big data infrastructure with streaming analytics and that’s one of the things that particularly motivated us to implement the Insense platform. On the Insense platform, retailers and other online businesses and SaaS platforms can now enjoy the benefits of modern recommender systems with our Universal Recommender being implemented based on two of the best technologies implemented by the likes of Netflix and Amazon. We bring to you the power that Amazon and Netflix have without you having to worry about spending the millions of dollars that they have spent to set up their own infrastructure or getting specialized expertise to achieve this.

80% of the content consumed on Netflix are from their recommender system with the recommender earning them $1 billion a year in value from customer retention.

Data is the new oil but oil but it is of no use without being refined. Recommender systems are definitely the naphtha we get from refining our company’s raw data.

Leave a Reply

Your email address will not be published. Required fields are marked *