Homepage Personalization in the Age of AI
- 2 days ago
- 4 min read
For any marketplace, the homepage is prized real-estate. Shoppers discover new items, marketplaces see highest revenues and CTRs on homepages and brands typically pay marketplaces anywhere between INR 25k - 4L a month for banners on the homepage.
Recently, I came across this excellent article on how DoorDash Used LLMs to Rebuild Its Homepage. In the case of DoorDash, the teams have revisited personalisation (on steroids) in the age of AI. Most apps fixate on personalisation as "reordering fixed categories." This is not real personalization. It is controlled reshuffling. However, the real opportunity isn’t in sorting better. It’s in redefining what gets shown in the first place.
How's Doordash killing it?
Instead of choosing from 300 fixed carousels, DoorDash built a system where an LLM now generates new carousel themes. These personalized carousels are per user and per time of day. That means breakfast, lunch, dinner, and late night can all look different for the same person.
They used: 1) Human labelers, 2) LLM-as-judge evaluation. They measured precision@10. It tells you how many of the top 10 results are relevant. They improved precision@10 from 68% to 85%. Basis A/B tests in San Francisco and Manhattan, the results showed:
Double-digit improvement in click-through rates
Higher conversion rates
Better homepage relevance scores
More merchant exploration
The system also helped smaller and niche restaurants get discovered more often. That improved the supply side of the marketplace.
In India, more recently I am seeing some fascinating and stellar examples.
How Meesho solves a cold start problem for new users
Meesho, the Bharat shopping app is unique in that the homepage is not filled with display banners but with a laundry list of products. Similar to a "bazaar" in India. Specifically, how does Meesho leverage its homepage to convert new and first-time internet shoppers in India? Especially, with zero browsing history, how do you show the right products from the very first scroll?
While they earlier started with demographic data, this wasn't enough. If the product feed on homepage is irrelevant, you have lost the shopper. The team developed the Cold Warm Net (CWN) model. The goal of modeling cold-start users is to learn effective user representation from both cold and warm user behaviors and build models that adapt as users evolve. Cold users are users with no interaction history. Warm users are new users with some interaction history.
Cold-Warm Net uses expert towers for cold-state and warm-up state of users, combined via a gate network that adapts based on user behavior. A dynamic teacher selector guides learning through knowledge distillation, ensuring high-quality personalization from the start. Rolling out these enhancements saw a notable uplift in feed engagement metrics like CTR, a sharp drop in bounce rate, indicating that users are finding more relevant content early in the feed, leading to quicker conversion, higher-quality interactions. The app also saw a notable rise in new user activation, along with a sharp drop in bounce rate.
Read more: https://www.meesho.io/blog/personalization-from-day-one-solving-the-cold-start-problem-at-meesho
Myntra leverages generative AI to help shoppers put the right look together
Myntra, the largest online fashion retailer in India took on a different approach to its personalisation in the age of AI. Its AI assistant helps shopper to put on the right look together for the gym, for a festival, etc. It goes a few steps beyond conventional search. It uses generative AI to respond to more open-ended questions like what to wear for a particular festival or a cricket match or even the trending fashion in a city.
“Users who shop using the AI shopping assistant are three times more likely to end up making a purchase,” said Amit Kumar Mondal, director of product management at Myntra. “Because it also helps users discover a complete look from multiple categories of products, we’re seeing that on average they add products from 16 percent more categories than usual.”
Zomato's Nudge Theory
Zomato’s average monthly transacting customers more than tripled from under 7 million to over 18 million.
Behind these numbers have been pivotal design and experience upgrades:
a) A full app redesign in 2017 that began shifting the platform toward a cleaner, more user-first layout
b) The 2022 launch of the Kimchi engine, which brought faster personalization and smarter in-app recommendations
c) The rollout of AI-powered Match Scores (2025), aimed at refining restaurant discovery based on individual preferences and behaviour patterns
By applying principles from behavioural science, especially nudge theory, the app helps users make faster, easier, and more satisfying choices.
Zomato doesn’t just show you what’s available. It shows you what you’re most likely to want first. By arranging restaurants and dishes based on preferences, order history, cuisine patterns, and even time of day, the app reduces choice overload and gets users to a decision faster. The “Recommended” tab acts as a gentle default nudge, steering users toward curated combinations with minimal cognitive load.
From Ratings to Relevance: Match Scores in Zomato’s UI/UX design
Zomato’s AI-powered Match Scores take personalization to the next level. Instead of relying solely on aggregate ratings, the platform calculates how well a restaurant aligns with an individual’s past behavior, considering their order history, preferred price range, cuisine choices, and even time of day. By placing high-match options at the top, the app simplifies the decision-making process. Users scroll less, choose faster, and are more likely to enjoy what they order. It’s a quiet shift from best-rated overall to best-suited for you that reflects deeper behavioral insight in action.
While the Indian marketplaces demonstrate how personalisation has lifted revenues, discovery and conversion, globally, personalisation has been powering the likes of Youtube, Netflix and even Facebook/Instagram feed to make these products what they are today.
Some great reads on the same:
Comments