Personalization for Indian eCommerce: What Actually Lifts AOV
Most "personalization" features ship as demos and never move a number. The ones that move AOV in Indian eCommerce are specific and tend to be unsexy. Here's the field-tested view.
Key takeaways
- Recommendations on PDP and cart: real AOV lift (5-15%).
- Search result ranking personalization: real lift, harder to A/B isolate.
- Personalized email subject lines and product picks: real lift.
- Dynamic homepage by behavior: marginal lift; high engineering cost.
- Personalized pricing: mostly hype in retail; tested in subscription.
Why this matters
Every personalization feature has engineering cost. Choose the ones that move the metric; skip the ones that don't.
What works
Product Detail Page recommendations
"Customers who bought this also bought" or "Similar items", proven, measurable. Use collaborative filtering plus content embeddings for cold-start.
Cart-page upsells
"Frequently bought together" or "Add for free delivery", common, effective.
Email and notification personalization
Personalized product picks in transactional and lifecycle emails consistently lift conversion. WhatsApp and SMS personalization the same.
Search ranking by behavior
Users who buy electronics see electronics-relevant results ranked higher. Hard to A/B isolate but cumulative effect is real.
Browse history nudges
"You were looking at X" placement on homepage on return, small lift, low cost.
What doesn't work (yet)
Dynamic homepage rebuilding
Completely personalized homepages don't outperform well-curated category-anchored homepages enough to justify the engineering.
Personalized pricing in retail
Discount banding by behavior risks PR and trust issues. Use sparingly if at all.
Voice/face-based personalization
Cool demo. Hasn't moved AOV in Indian retail at scale.
The AI angle
Modern recommendations use embedding-based similarity (much better than the old collaborative filter alone). Hybrid (collaborative + content + popularity) beats pure embedding. Add behavioral signals (recent views, cart abandons) for cold-start.
For Indian eCommerce specifically, language signals matter, surface descriptions in user's preferred language, even if catalog is multilingual.
Common pitfalls
A/B testing for too short a window. Personalization lift compounds over weeks; one-day tests show noise.
Over-personalizing on early signal. Cold-start matters; don't over-fit on 2 page views.
Recommendation system that recommends what's out of stock. Inventory awareness is non-negotiable.
Filter bubbles. If you only show what users have viewed, they never discover anything new. Mix exploration with exploitation.
What we recommend
Start with PDP recommendations and cart-page upsells. Measure AOV lift. Add search ranking personalization next. Skip the heavy dynamic-homepage work until everything simpler is exhausted.
FAQs
Algolia Recommend vs custom? Algolia is faster to ship; custom is more flexible at scale.
Real-time vs batch? Batch (overnight) is fine for most recommendations; real-time only matters for cart-page contextual.
How much lift can we expect? 5-15% AOV lift is realistic; 30%+ claims are usually selection bias.