AI-Personalized Product Recommendations in Email: Going Beyond Klaviyo's Defaults
By Loyal Send, Editorial Team5 min read
Listen to this article7:26
Why Defaults Aren't Enough
Klaviyo's built-in product recommendation blocks pull from fixed rules: best-sellers, recently viewed, related category. These work fine as a baseline but they're static. Every subscriber sees essentially the same recommendations, filtered only lightly by category or recent behavior. The result is generic cross-sell and 1-3 percent click-through rates on "you might also like" blocks.
AI-personalized recommendations use a subscriber's full purchase history, browsing behavior, declared preferences, and collaborative-filtering signals to pick products per person. Done well, the same email becomes 50 different emails, each showing 3-4 products optimized for that individual. CTR on the product block rises to 8-15 percent; revenue per send typically jumps 20-50 percent.
The Three Layers of Email Personalization
Layer 1: Rule-based
"If subscriber is in Skincare category, show skincare products." Easy, no AI needed, works as a baseline. Klaviyo natively supports this.
Layer 2: Collaborative filtering
"People who bought X also bought Y." Requires at least 10K orders of history to work well. Shopify's native recommendations do this; Klaviyo can pull it in via Shopify integration.
Layer 3: AI-powered individual personalization
"For THIS subscriber's specific purchase history + browsing pattern + preferences, the product with highest purchase probability is Z." This is what dedicated recommendation engines do.
The Tools That Do Layer 3
- Nosto: enterprise-focused recommendation engine with deep Klaviyo integration. $500-3000/month. Best for brands over $5M revenue.
- Rebuy: Shopify-native, integrates with Klaviyo product blocks. Cheaper ($100-500/month). Good for mid-market.
- LimeSpot: Similar to Rebuy, strong for cart recovery + email personalization.
- Segment + custom model: For brands with data team, build custom collaborative-filter model and pipe into Klaviyo via API.
For most DTC brands under $5M, Rebuy or LimeSpot is the sweet spot. Quick to implement, meaningful lift, reasonable cost.
How to Integrate With Klaviyo
The recommendation tool generates a per-subscriber product feed (typically 4-8 product IDs with images, prices, URLs). Klaviyo pulls this feed via:
- Webhook on email send (request current recommendations at render time)
- Profile property sync (recommendations updated daily, stored on profile)
- Dynamic content blocks (Klaviyo queries the recommendation API when email renders)
Method #1 (webhook at send time) gives freshest recommendations but requires webhooks to be fast and reliable. Method #3 (dynamic blocks) is more common and works with any recommendation engine that has a standard API.
The Content-Aware Email
Beyond product recommendations, AI can now personalize email copy itself. This is where things get interesting in 2026:
- Subject line variant tests where AI picks the variant most likely to perform per-subscriber
- Body copy tone adjusted to subscriber segment (casual for Gen Z, professional for B2B, etc.)
- Send-time optimization (not just timezone, but within timezone based on individual engagement patterns)
Klaviyo's AI features (subject line generator, smart send time) are first-wave versions of this. Expect meaningful improvements across 2026-2027.
When AI Recommendations Fail
- Insufficient data. A brand with 500 orders can't train a collaborative model. Stick with rule-based.
- Narrow catalog. If you sell 10 products, AI recommendations are noise. Show the best 3 on rules.
- Cold-start subscribers. A brand-new subscriber has no history. AI recommendations default to best-sellers; this is fine, just don't expect magic.
- Over-personalization. Showing only 1 product type forever creates boring email. Mix in 30-40% discovery content (new arrivals, category exploration).
The Privacy Concern
AI personalization relies on data. Subscribers increasingly expect transparency. Some brands now include a "Why am I seeing this?" link next to recommendations that explains: "Based on your recent purchases and stated preferences." This is excellent trust-building. It also reduces the creepy factor when recommendations feel too accurate.
Measuring Success
- Product block CTR: Baseline is 1-3%. AI-powered should hit 8-15% within 60 days of implementation.
- Revenue per recipient: Lift of 20-50% expected.
- Diversity of products viewed: AI should not narrow subscriber interest, if it does, fix the model to include exploration.
- Unsubscribe rate: Should not change or should decrease. If it increases, recommendations are likely annoying subscribers with over-repetition.
Frequently Asked Questions
Do I need a huge catalog for AI recommendations to help?
Yes, at least 50-100 products. Below that, rule-based is better. AI shines with broad catalogs and purchase depth.
How quickly do recommendations improve?
Initial results within 30 days. The model gets meaningfully better at 60-90 days when it has observed enough click-through data to learn individual preferences.
Should every email have personalized recommendations?
Most should, but not all. Brand-story emails don't need recommendations. Product-driven emails (post-purchase, cart recovery, weekly digest) do.
What about SMS?
SMS personalization is limited by character count. Include 1 personalized product link max in SMS, not 4. Lean on email for recommendation-heavy messaging.
Is this worth it below $1M revenue?
Not typically. Below $1M, the ROI of rule-based personalization exceeds AI because you don't have enough purchase depth to train models. Focus on getting segmentation + flows right first.