When I am hungry and short on time, I want to decide what to eat quickly and trust it will be good, so I can stop scrolling and just order.
If Zomato reduced the decision to a small set of high-confidence, personalized picks and made the reason for each one visible, decision time and cart completion would rise more than any gain from a bigger catalog or a deeper discount.
- Reordering a past favorite is nearly frictionless, which serves the highest-frequency job well.
- AI Match Scores personalize ranking beyond raw aggregate ratings, a real step past one-size-fits-all lists.
- Group ordering with shared carts and clear status handles a genuinely hard social coordination problem.
- The Sushi design system lets teams reshape the UI from the backend and experiment without app updates.
The Match Score is a black box. A user sees a place ranked for them but not why, so they do not trust it and fall back to scanning ratings.
Why it matters: Personalization only pays off if it is believed. An unexplained score is ignored, and the app is back to being a long list sorted by a number.
Personalization ranks the list but does not shorten it. Dense, uncurated options still stretch far below the fold.
Why it matters: The core pain is decision fatigue. Reordering the overwhelm without reducing it leaves the hardest part of the job, actually choosing, on the user.
Ads and discounts compete with relevance in the same feed, so best for me and best margin for Zomato blur together.
Why it matters: When users cannot tell recommendation from promotion, they discount every recommendation, which erodes the trust personalization was meant to build.
There is no fast decide-for-me path for the frequent, low-stakes just feed me moments.
Why it matters: For habitual orders the app still demands full browsing, taxing the exact sessions that should be effortless.
Make the Match Score legible with a one-line reason on each card, for example you reorder biryani on Fridays, and 4.4 from people who order like you.
Turns an ignored number into a trusted recommendation, which is the whole point of personalizing in the first place.
Match Score tap-through and time from open to order
Exposing logic invites gaming and complaints. Show human reasons, not raw model weights.
Add a Decide for me mode that presents three high-confidence picks with one-tap order and collapses the long list.
Attacks decision fatigue directly by reducing choices instead of merely reordering them.
Share of orders via Decide for me and cart-abandon rate
One bad pick erodes trust fast. Gate on a confidence threshold and make swapping a pick effortless.
Visually separate sponsored from for you and cap ad density inside the personalized rail.
Protects the credibility of the recommendation, which compounds into repeat orders and retention.
Trust survey score and repeat-order rate
Short-term ad revenue may dip. Defend the change with retention and lifetime value, not daily ad yield.
Weekly sessions where a user decides and orders without doom-scrolling.
Relevance in the personalized rail must not fall below a set ad-load ceiling.
I would not add more filters or a bigger catalog. The problem is too many choices, not too few, and adding surface area makes the core job of deciding even harder.