Kaspi.kz in numbers:
- More than 450,000 products on a single platform
- More than 8,500,000 monthly active app users
Kaspi.kz is Kazakhstan’s largest innovation company, built around the country’s most popular mobile super app. Inside it, people make payments and transfers, manage their personal finances and shop online with delivery. Today Kaspi.kz is the largest e-commerce platform in Kazakhstan, with roughly $600 million in turnover over the first nine months of 2020 and +129% year-on-year growth. This case study explains how Kaspi.kz achieved a 19.4% uplift in product recommendation conversion in three months on its marketplace.
The customer journey
Although the platform is available on desktop and on the mobile web, the main entry point into the “Shop” section is the mobile app itself: users have already installed the SuperApp, are logged in, and can place orders right there.
Buying a TV set, for example, looks like this:
- In the “Shop” section, the user types the sub-category name — “TVs” — into the search bar
- Finds the right product page using filters
- Sees offers from different sellers (merchants)
- Picks the offer they like and checks out, with the option to buy on credit or in instalments
To lift conversion on marketplace product recommendations, the Kaspi team chose Stellar Labs as its main partner — a company whose expertise includes advanced personalization of mobile apps, something few personalization vendors can offer.
Stellar Labs is an IT consultancy specializing in personalization and conversion rate optimization for online businesses.
Its team of senior experts helps large brands improve the way they engage customers and build a distinctive user experience on top of leading technology platforms (Dynamic Yield, Algolia, Persado and others).
Personalization inside the Kaspi.kz SuperApp
One of the main goals for Stellar Labs and Dynamic Yield was to personalize the home screen of the “Shop” section so that every user could find and buy what they needed as quickly and easily as possible.
The home screen is the primary touchpoint with the user, so it had to rebuild itself for each visitor individually, surfacing only the product offers that matched their interests.
How the marketplace home screen is built
The “Shop” section has a home screen with a fixed set of widgets: a banner, a category block and a product recommendation block. This is what the home screen looks like for a new user: a block of recently viewed and recommended products. With the Dynamic Yield platform the pool of widgets expands — alongside recently viewed recommendations, users can also see popular products from different categories or a “You may also like” widget.
The “You may also like” block runs a mix of strategies: personalized selections based on the user’s Affinity profile, on the behaviour of similar users, on the most probable next purchase after previous orders, and so on.
Affinity is a composite signal of user interest. It takes into account the attributes of the products a user viewed, added to cart, purchased or otherwise interacted with — category, brand, colour, material and many more — weighted by recency.
The goal of the test was to show how far Dynamic Yield could personalize the home screen. The test therefore compared a home screen built with product recommendations from the previously used personalization solution against a home screen built on the Dynamic Yield platform.
What the test was built on to hit the main goal — more conversions from the home screen
1. A personal product selection for every user
If you switch personalized recommendations on for everyone at once, the home screen will most likely fill up with smartphones and gadgets. Why? Because those are the best-selling items on every marketplace that carries electronics.
The Stellar Labs team had to configure personalization on the Dynamic Yield platform so that smartphones or any other category were shown only to users genuinely interested in them at that moment, not to everyone. On top of that, the pandemic made it important to recommend seasonal and other in-demand products from categories such as home and garden, auto parts and so on.
2. Personalizing the storefront around recently viewed products
“Recently viewed products” is a well-known recommendation strategy — and no less effective for it. Viewed at least one product card in the auto parts category? Then we will show you products from that category on the home screen. Looked at a women’s watch? On your next visit the viewed item shows up in the “Recently viewed” widget, and a selection of comparable watches appears in a recommendation block below it as an alternative.
3. Refreshing the widget set on the home screen often
Say you looked at an item from the auto parts category today. The question is: how long should you stay in the segment of people interested in that category? 30 days, 14, 7, 3, 1? The idea is that the home screen changes often enough that you do not see the same recommendations from one category for a whole month.
The pain point of every marketplace doing personalization is recommending something the user has already bought. To avoid pushing a purchase the customer may have already made elsewhere, segments were given a short lifetime, which was gradually reduced during the experiment while performance metrics were monitored.
How performance was measured
To understand how users with an interest in specific categories react to one home screen variation or another, the Kaspi and Stellar Labs teams built a large number of segments and audiences with Affinity to categories at different levels and to views of specific products.
Splitting the traffic made it possible to watch how each variation performed. As soon as a result reached statistical significance, the winning strategy was rolled out to a larger share of traffic.
The result: the first 19.4% conversion uplift came within the first three months — including the time spent integrating the Dynamic Yield platform.
One of the key things about this test is that we did not just build the functionality and get an impressive result — we did it very fast. That is the core advantage of Dynamic Yield. The platform lets you launch dozens of campaigns a month with branching logic, and few vendors can match that flexibility. And speed of launching and testing hypotheses is speed of getting value out of the project.
Andrey Tyshchenko, ex-Regional Director for Russia, CIS and Eastern Europe, Dynamic Yield
