Careem (Uber) · Super App · Google Gemini · 2025

Global
AI
Search

Integrated Google Gemini to rebuild Careem's broken global search from the ground up, restoring customer trust, achieving 94% intent classification accuracy across 10+ markets and multiple languages, and driving a 67% improvement in search-to-action conversion.

Role

Head of Product Design, Pay & Platform

Timeline

2025

Markets

10+ across MENA

AI Partner

Google Gemini

+67%
Search-to-action conversion
94%
Intent classification accuracy
-54%
Time-to-transaction from search
Unified Search, One Query, Every Vertical
INDEXING6 Verticals, 50M+ Items
INTENT MODELGemini 2.0 Flash
๐Ÿ” "birthday party for my daughter" Gemini
Query Parsing
Tokenization, spell-check, synonyms, language detection
โ†’
Intent Classification
Multi-vertical scoring, confidence per category
โ†’
Parallel Retrieval
Real-time index scans, signal weighting
โ†’
Confidence Ranking
Score aggregation, UI prominence logic
Intent Routing, Confidence Scores
๐Ÿฐ
98%
Food
Birthday cakes, party platters
3 results
๐Ÿ›’
87%
Groceries
Baking supplies, juice, snacks
5 results
๐ŸŽˆ
92%
Shops
Decorations, party supplies
4 results
๐Ÿงน
34%
Cleaning
Post-party cleanup
2 results
๐Ÿš—
8%
Rides
Guest pickup (low relevance)
,
๐Ÿ’ณ
5%
Pay
Not matched to intent
,
Confidence Thresholds
95%+ Confidence: Prominent position, featured results
75-94% Confidence: Secondary results, "Did you mean?" context
<75% Confidence: Fallback to keyword search + manual browse
Edge Cases Handled
โ€ข Empty results for any vertical
โ€ข Ambiguous multi-intent queries (โ†’ carousel)
โ€ข Non-English/regional market queries
โ€ข Typos, slang, dialect variations
94%
Intent Accuracy
6
Verticals Indexed
50M+
Searchable Items
4
Languages
10+
Markets
+31%
Search Engagement

Audit.
Integrate.
Rebuild
trust.

I led the most comprehensive search redesign in Careem's history, from cross-market research through Google Gemini integration to a progressive trust model that visibly improved with use.

01

Cross-market search audit, Directed research with 200+ participants across UAE, Saudi Arabia, Jordan, and Pakistan, documenting every failure mode, frustration point, and workaround customers had developed.

02

Google Gemini integration architecture, Worked with engineering leadership to evaluate, select, and design how Gemini would interact with Careem's service catalogue, customer context, and transaction history.

03

Cognitive load reduction, Redesigned the results architecture to prioritise clarity: AI's best interpretation of intent first, confidence-ranked actions (not listings), and transparent fallback paths.

04

Progressive trust model, Designed a system where search accuracy visibly improved over time per customer. The AI learned from successful completions, with subtle UI signals communicating personalisation without being intrusive.

05

Multi-language intent classification, Ensured 94% accuracy across Arabic, English, Urdu, and Turkish, with cultural nuance handling for regional service expectations.

"AI in product design isn't a feature, it's infrastructure. Rebuilding search on intelligence didn't just fix a feature; it restored trust in the entire super app."

Trust
restored.
Search
transformed.

The Gemini-powered search restored customer trust in Careem's primary navigation tool and established a new AI platform capability that every vertical could leverage.

Conversion

67% improvement in search-to-action conversion, customers found what they needed and acted on it.

Speed

54% reduction in time-to-transaction from search initiation across all verticals.

Trust Recovery

38% increase in repeat search usage, customers trusted search again.

Cross-service

29% increase in cross-service discovery, customers finding services they hadn't previously used.

Next Case Study

Crypto Neo-Bank:
AI-Enhanced Design Operations

Want in?

Score 100% on the Design Eye test to earn access. Or enter the password if you have it.

4 rounds. No partial credit.