The framework
One engine. Three revenue levers.
One engine learns from every customer interaction. Three revenue levers run on top of it — recommendations, promotions, search — sharing the same behavioral data layer, so each gets smarter as the others do.
The engine
Synerise AI
The science behind every result. Real-time behavioral signals. Benchmark-proven against the world's largest AI labs.
- 01
Recommendations
Homepage · PLP · PDP · Minicart · Cart · 404 · Empty search.
- 02
Promotions
AI coupon engine. 1-to-1 offers at scale. Web push and on-site.
- 03
Search
Semantic, vector, NLP. Personalized ranking. Real-world A/B results.
Section 01 · The Engine
Behavioral AI, proven head-to-head against the world's largest labs.
Synerise builds behavioral AI infrastructure. The models behind this playbook run on it — and beat the published models from the largest AI research labs in real-world conditions.
Head-to-head against published recommendation models.
Challenge placements
- 1stRakuten Data Challenge
- 2ndTwitter RecSys AI Challenge
- 2ndBooking.com AI Challenge
- 3rdKDD Cup
Synerise is also the organizer of the RecSys Challenge.
Takeaway
Research-grade AI, deployed in production. The same engine that wins academic benchmarks runs every result in this playbook.
Section 01 · The Engine
Real numbers from live tests.
Synerise vs. control group · A/B-confirmed across the funnel
Test setup
- Synerise tested head-to-head against another solution, measured on a live control group.
- Funnel measured end-to-end: CTR (clicked / viewed), click → cart, cart → purchase.
- Personalization wins at every step — and the advantage compounds down the funnel.
- +136%CTR1,96% vs. 0,83% control
- +23%Click conversion → cart19,3% vs. 15,7% control
- +26%Cart conversion → purchase17,8% vs. 14,1% control
Takeaway
Every step of the funnel improves — more clicks, more carts, more purchases — confirmed against a live control group.
Section 01 · The Engine
Fashion marketplaceFashion marketplace — personalized showcase test.
Intelligent showcase vs. standard version · A/B test
Key results
- A/B test of the personalized (“intelligent”) showcase against the standard version.
- Add-to-cart — the main KPI of the test — improved across the board.
- Mobile delivered the biggest gains across all metrics.
- +3–5%Add-to-cart — main KPIImprovement overall
- ~+10%PurchasesOverall lift
- +6.3%RevenueOverall · +10.2% on mobile
Conclusion
The personalized (“intelligent”) showcase outperforms the standard version across nearly all key business metrics — especially on mobile — leading to higher engagement, conversions, and revenue.
Section 01 · The Engine
Why this matters.
- 01
Merchants compete on conversion and cart value, not catalog size.
The two surfaces that move both: Recommendations and Search.
- 02
Most tools never see the customer end-to-end.
Search and recommendation vendors — and native solutions — don't cross behavioral data with real-time purchase signals. Synerise does.
- 03
Synerise turns Recs + Search into a revenue lever.
Connects to any commerce stack via API and SDK. No replatforming. A/B measurable in weeks, not quarters.
Section 02 · Recommendations
Placement opportunities across the customer journey.
Most merchants use only two placements today — Homepage and PDP. Five more are open.
- 1Homepage — banner & carouselsPersonalized in real time, per session.
- 2Category / Listing page (PLP)Personalized ranking and contextual recommendations.
- 3Product Detail Page (PDP)"Similar items" and "Next Interaction".
- 4Minicart / Pre-cartCross-sell — 2.8× add-to-cart, 2.0× cart conversion.
- 5Cart pageLast-touch upsell before checkout.
- 6404 pageRecover lost traffic with relevant picks.
- 7Empty search resultsSubstitute a dead end with a behavioral fallback.
Takeaway
The same engine, seven placements. Each one is independent revenue you're not capturing yet.
Recommendations · Category page
Media retailerSynerise engine vs. native engine, on identical traffic.
How it works
- Goal: measure the Synerise engine against the client's native engine on identical placement.
- Recommendations personalized on behavioral history, scoped to the "books" category.
- 10% of traffic to each engine — same placement, same context, side-by-side.
- Books
- Music
- Film
- Games
- Kids
The Silent Orchard
€12.99
Atlas of Small Wonders
€15.49
Letters from June
€9.99
Nightwatch Papers
€13.99
A Season of Glass
€11.49
The Cartographer
€14.99
Green River Diaries
€10.99
Bread & Salt
€16.99
- +92%Revenue from placementvs. native engine
- +144%Conversion rate1.22% vs. 0.5% native
- +58%CTR16.82% vs. 10.64% native
Takeaway
+92% more revenue from the same placement — on identical traffic.
Recommendations · PDP
Fashion brandThe carousel predicts the next move, not the last one.
Model & scope
- A "next interaction" model predicts the products the customer is most likely to engage with next.
- Adapts to current buyer context: gender, category, product type.
- Filters by last selected or purchased size; excludes the exact product being viewed.
- Avoids duplicate color variants, with broader category matching as fallback.
- Women
- Men
- Kids
- Home
Wide-leg linen trousers
(214)
€39.99
Sizes XS–XL · 3 colours
Add to bagCropped blazer
€59.99
Straight denim
€34.99
Silk-blend top
€27.99
Leather belt
€19.99
- +44%CTR upliftSynerise vs. platform default
- +35%Add-to-Bag rateSynerise vs. platform default
- +7.2%Conversion rate uplift0.74% (Synerise) vs. 0.69% platform default
Recommendations · Pre-cart
Electronics retailerAccessory cross-sell, the moment the cart fills.
How it works
- After a high-value product enters the cart, the engine recommends relevant accessories in real time.
- Reads browsing and purchase history to predict which accessories complement the main product.
- Personalized per user — same TV, different accessory priorities by profile.
- Computers
- TV & audio
- Phones
- Gaming
- AGD
ProBook 16″ · 1 TB SSD
Added to cart just now
Wireless mouse
€29.99
Laptop sleeve 16″
€24.99
USB-C hub · 7-in-1
€44.99
- 2.8×Add-to-cart liftvs. previous algorithm
- 2.0×Cart conversionvs. previous algorithm
- +170%Conversion on high-margin productsBehavioral AI driven
Takeaway
Up to 2.6× higher revenue per user in pre-cart cross-sell — driven entirely by behavioral AI.
Recommendations · 404 page
Sports brandA bounce becomes a conversion.
How it works
- When a customer hits a non-existent URL, Synerise personalizes the 404 page in real time.
- Shows recently viewed products and personalized recommendations instead of a dead end.
- The same behavioral engine, a different placement.
- New
- Men
- Women
- Kids
- Sale
404
This page doesn’t exist — but these were picked for you.
Velocity Run 3
€119.99
Terra Trail GTX
€139.99
Court Classic
€89.99
- 4.27%CTR — +402% vs. 0.85% baseline
- 2.33%Conversion — +104% vs. 1.14% baseline
Takeaway
CTR up 5×, conversion doubled — recovering traffic that would otherwise be lost for good.
Recommendations · Dynamic On-site
Pet retailerPrecision at the level of an individual pet's name.
Mechanism
- A complex ruleset, configured and live in under one day — every edge case covered.
- Shows the pet's name only when it's in the profile. No awkward fallback.
- Shows only pet names matching the food type — multi-pet households handled correctly.
- Target: no purchase in 6+ weeks, drawn from 82,773 transactions.
Time to restock for Luna
Based on her usual 4-week feeding cycle
Adult dry food · 10 kg
Bought 6 times
Dental chews · large
Usually bought together
- +70%After-click conversion rateHyper-personalized vs. control
- +28%Average Order ValueUplift vs. control group
- 14%CTR — one-day setup346 active users targeted
Recommendations · Dynamic On-site
Footwear brandUrgency isn't price. It's availability.
Mechanism
- Synerise infers each customer's most likely size from aggregates.
- Recommends only products available in that size — not generic "limited stock" messaging.
- Right pop-up, right user: a running enthusiast sees running shoes; a casual user sees lifestyle sneakers.
Runner 990 · size 42.5
€129.99
- +8.5%Conversion rate upliftSynerise vs. non-personalized variant
Takeaway
Same trigger, different product — made individually relevant by inferred interest.
Section 03 · Promotions
From mass promotions to 1-to-1 offers.
The engine selects the most relevant promotion for each individual customer.
- Input
The available promo pool
% discount on Product A, B, C — and N more offers.
- Engine
Synerise AI
Proprietary deep learning. Real-time segmentation incl. last purchase. Cold-start prediction. Continuous testing. No manual adjustment.
- Output
A personalized coupon set per customer
Ranked most-relevant to least. Delivered on-site, web push, email. Re-ranked as behavior evolves.
Takeaway
The same promo pool, ordered for the individual — and re-ordered the moment behavior changes.
Promotions · Smart Offers
Grocery chainA larger promo pool holds effectiveness as the base scales.
When the promo pool was cut, spend and frequency dropped — a larger, AI-managed pool holds campaign effectiveness as the user base scales.
- 19,25%Revenue uplift — up tovs. control group
- 450%Conversion uplift — up tovs. control group
- 7×Promo volume expansionAI-managed pool vs. baseline
Takeaway
AI promotions hold their effectiveness even as the user base grows.
Promotions · Cart
Sports retailerPrediction picks the offer; the ecosystem delivers it.
Scope
- An AI coupon system predicts behavior and offers personalized codes at key stages — PDP, PLP, Cart.
- First test: free-shipping coupons for large bikes and equipment, targeted at high-churn customers.
- Full-ecosystem test: on-site modules, segmentation, automation, alerts and analytics as one.
Velocity Run 3
Size 42
Training tee · dry-fit
Size M
- +8.52%Conversion rate upliftSynerise vs. non-personalized
- +147%Average revenue per userUplift vs. control group
Promotions · On-site
Pet retailer4,466 users. One in three site-wide redemptions.
Context
- A tightly targeted audience of just 4,466 users generated outsized impact through precision.
- 21.4% CTR — this segment drove 1 in 3 coupons redeemed site-wide.
- The coupon carried loyalty-program framing: member benefit, time-limited urgency.
−15% on your next order
Valid for 48 hours on your favourite categories
SAVE-15-NOW
- +96%AOV upliftvs. control group
- +24%AOS upliftAverage Order Size
- 33%Synerise contribution to promo revenue31.7% of total coupon redemptions
Takeaway
Precision beats reach.
Section 04 · Search
7Modalities
Full spectrum of search capabilities.
One engine. Multiple modalities. All personalized.
Textual search
Finds items relevant to the entered term.
Full-text search
Handles complete queries; understands intent beyond exact-match keywords.
Autocomplete
Predicts and suggests terms as the user types; fewer query errors.
Visual search
Search by image upload — no words needed.
Non-textual / listing
No query required; returns results by filter attributes.
Personalized ranking
Orders product cards with merchandising rules on top of AI relevance.
Geo search
Finds products within a set distance — click-and-collect or local delivery.
Takeaway
One engine covers keyword, intent, image and location — and personalizes all of it.
Section 04 · Search
Six dimensions of personalized ranking.
Business rules and AI signals combined — every result ordered for the individual.
- 01
Personalized product ranking
Re-orders PLP results on behavior, preferences and session context to maximize relevance and conversion.
- 02
Merchandising boost rules
Combines AI ranking with business factors (margin, stock, seasonality), aligning personalization with commercial goals.
- 03
Real-time adjustments
Automatic re-ranking the moment a product enters the feed. Zero lag.
- 04
Sponsored product placements
Flexible ad placements inside personalized PLPs — relevance plus monetization.
- 05
Campaign-specific personalization
Dynamic search and PLP adjustments tied to campaigns — promo boosts plus real-time AI.
- 06
Targeted promotional listings
Intensifies promotional exposure only for high-churn-risk customers. Precision retention.
Takeaway
Commercial rules set the floor. Behavioral signals order everything above it, per customer.
Section 04 · SearchFashion retailer
Synerise vs. incumbent search — conversion rate.
European markets · A/B tested vs. the incumbent general-purpose search engine
+12%Avg. CR improvement
DE · Multibrand fashion
- Session window
- 30 min
- Incumbent CR
- 5.18%
- Synerise CR
- 6.32%
- Impact
- +22%
DE · Multibrand fashion
- Session window
- 2 hours
- Incumbent CR
- 5.58%
- Synerise CR
- 6.81%
- Impact
- +22%
DE · Multibrand fashion
- Session window
- 24 hours
- Incumbent CR
- 8.16%
- Synerise CR
- 11.12%
- Impact
- +36%
IT · Multibrand shoes
- Session window
- 24 hours
- Incumbent CR
- 4.88%
- Synerise CR
- 5.28%
- Impact
- Improvement
Other markets
- Session window
- —
- Incumbent CR
- Similar
- Synerise CR
- Similar
- Impact
- Comparable
Takeaway
The client moved to Synerise search across all markets — driven by improved KPIs and lower total cost of ownership.
Search · PLP
Sports retailerSame URL, different product order per user.
How it works
- For each user, the engine generates a product listing selected on their on-site behavior.
- Accounts for product views and transaction data — not just clicks, but purchases.
- Same category-page URL — no site-structure changes.
- Running
- Cycling
- Outdoor
- Fitness
- Sale
Velocity Run 3
€119.99
Aero jersey pro
€64.99
Terra Trail GTX
€139.99
Studio flow leggings
€39.99
Kids’ trainer 24
€34.99
Court Classic
€89.99
- +16.85%Conversion rate upliftSynerise vs. control
- +283%Avg. revenue per userSynerise vs. control
- +26.79%Avg. session durationUsers engage longer with relevant results
Takeaway
+283% revenue per user. Personalized ranking doesn't just lift clicks — it changes purchasing behavior.
Section 04 · Personalized ranking (listing)
Sports brandBehavioral ranking vs. rule-based listing.
Head-to-head test on one sports PLP · 50/50 split · same catalog
Index configurations, side by side
A · AI listing — personalized
★ Winner- Page views — session intent60%
- Item purchase — brand & size40%
- Personalization — full re-rank90%
At 90%, personalization re-ranks the entire result set per customer — two shoppers in the same category see meaningfully different orderings.
B · Listing — collection-rules baseline
All behavioral signals off. Order set by collection rules only — no behavioral data applied.
Results — Variant A vs. B
- +41%Conversion rate1.14% vs. 0.67%
- +44%Revenue per profile
- +6.3%Click-through rate12.02% vs. 11.26%
- −13%Avg. click positionRelevant items surface higher
Takeaway
Collection rules set a logical starting point. Behavioral ranking made it personal — worth +41% conversion and +44% revenue per profile.
Search · POCquery: “slimming pen”
Pharmacy chainContext-aware search, beyond keyword matching.
✕ Native search (current)
- Medicines
- Supplements
- Beauty
- Baby
Cosmetic contour pen
Matched keyword: “pen”
Lancet pen set · 25 pcs
Matched keyword: “pen”
Marker pen — skin safe
Matched keyword: “pen”
Intended product does not appear
Locks on the first word, "pen" — returns cosmetic pens and lancets. GLP-1 / Ozempic-class products never appear.
✓ Synerise — no tuning, no rules
- Medicines
- Supplements
- Beauty
- Baby
GLP-1 injection pen · 1.5 ml
Understood intent
GLP-1 analog · starter kit
Semantic match
Needles for injection pens
Complementary
Reads the intent and returns GLP-1 analogs (Ozempic-class equivalents). Zero config.
Search · Behavioralquery: “razor brand” + user ID
Pharmacy chainSame query, two users, two entirely different results.
Female user — personalized
- Medicines
- Supplements
- Beauty
- Baby
Women’s razor · 3-blade
Inferred from purchase history
Women’s replacement blades · 4-pack
Same line
Shave gel — sensitive
Frequently bought together
No gender in the profile. The engine infers it from purchase history and shows the women’s line first.
Male user — personalized
- Medicines
- Supplements
- Beauty
- Baby
Refill cartridges · 8-pack
Buys refills regularly
Refill cartridges · 4-pack
Alternative size
After-shave balm
Frequently bought together
Buys refill cartridges regularly. The engine infers he already owns a razor and shows refills first.
The compound effect
Platform-wide impact.
Based on 100+ enterprise and mid-market clients · Recommendations + Promotions + Search combined
- +7–9%Revenue increaseShowing customers more of what they're looking for
- +4–17%Average order ticketFrom more relevant product offers
- +5–30%Conversion ratevs. control, from highly personalized offers
- −4–30%ChurnPredicting and preventing before it happens
- +12–19%Customer LTVCompounding across the lifecycle
Three levers, one behavioral data layer — compounding across the funnel.
One platform.Three revenue levers.Measurable in weeks.
- 01
Already proven in production
Real A/B results with multiple live merchants. Connects via API and SDK. No replatforming.
- 02
Science that compounds at scale
Every behavioral event improves every model. As the merchant base grows, accuracy, relevance and targeting improve together.
- 03
Fast to activate, easy to expand
Start with recommendations. Add search. Layer in promotions — all on the same behavioral data layer.