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.

    1. 01

      Recommendations

      Homepage · PLP · PDP · Minicart · Cart · 404 · Empty search.

    2. 02

      Promotions

      AI coupon engine. 1-to-1 offers at scale. Web push and on-site.

    3. 03

      Search

      Semantic, vector, NLP. Personalized ranking. Real-world A/B results.

    Section 01

    01

    The Engine

    Behavioral AI infrastructure · Proven head-to-head against the world's largest labs

    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.

    +61%vs. Google DeepMind
    +292%vs. Meta AI

    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 marketplace

    Fashion 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.

    1. 01

      Merchants compete on conversion and cart value, not catalog size.

      The two surfaces that move both: Recommendations and Search.

    2. 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.

    3. 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

    02

    Personalized Recommendations

    Seven placement zones · Real A/B results · Live clients

    Section 02 · Recommendations

    Placement opportunities across the customer journey.

    Most merchants use only two placements today — Homepage and PDP. Five more are open.

    1. 1Homepage — banner & carouselsPersonalized in real time, per session.
    2. 2Category / Listing page (PLP)Personalized ranking and contextual recommendations.
    3. 3Product Detail Page (PDP)"Similar items" and "Next Interaction".
    4. 4Minicart / Pre-cartCross-sell — 2.8× add-to-cart, 2.0× cart conversion.
    5. 5Cart pageLast-touch upsell before checkout.
    6. 6404 pageRecover lost traffic with relevant picks.
    7. 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 · Homepage

    Sports retailer

    The slot stays the same. The product and the offer change.

    How it works

    • The main homepage banner personalizes in real time, based on recently viewed products and behavior.
    • The system adjusts the visual and the linked offer to match current interest.
    • Same slot, every visit — fully individualized experience.
    NORTHPEAK
    User A · running affinity

    Homepage banner · this week

    Trail running picks — up to −20%

    Shop running
    User B · cycling affinitysame slot, same session

    Homepage banner · this week

    Your next ride starts here

    Shop cycling
    • +47%CTR upliftSynerise vs. non-personalized variant
    • +2.2%Conversion rate upliftSynerise vs. non-personalized variant

    Takeaway

    Driven entirely by real-time behavioral signals.

    Recommendations · Homepage

    Sports retailer

    Personalized category & brand banners.

    How it works

    • Surfaces the most-accessed sports and preferred categories, personalized per visitor.
    • Personalizes the brand section too — showing brands the individual has engaged with.
    • Performance measured after-click, against a non-personalized variant.
    NORTHPEAK
    Your categoriespersonalized per user
    For you

    Trail running

    From €49

    Road cycling

    From €89

    Hiking & camping

    From €29

    Yoga & studio

    From €19

    Men’s apparel

    From €15

    Kids’ sport

    From €12

    • +112%CTR upliftPersonalized vs. non-personalized
    • +73%Conversion rate upliftAfter-click performance comparison

    Takeaway

    The clicks are qualified — users find what they actually want.

    Recommendations · Category page

    Media retailer

    Synerise 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.
    PAGINA
    Recommended for youSynerise engine
    For you

    The Silent Orchard

    €12.99

    Atlas of Small Wonders

    €15.49

    Letters from June

    €9.99

    Nightwatch Papers

    €13.99

    Bestsellers in this category

    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 brand

    The 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.
    ATELIER N°9

    Wide-leg linen trousers

    (214)

    €39.99

    Sizes XS–XL · 3 colours

    Add to bag
    You may also likepredicts the next move

    Cropped 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 retailer

    Accessory 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.
    VOLTA

    ProBook 16″ · 1 TB SSD

    Added to cart just now

    €1,299✓ In cart
    Complete your setupaccessory cross-sell
    Fits your laptop

    Wireless mouse

    €29.99

    Laptop sleeve 16″

    €24.99

    USB-C hub · 7-in-1

    €44.99

    Go to checkout · €1,299
    • 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 brand

    A 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.
    STRIDE

    404

    This page doesn’t exist — but these were picked for you.

    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 retailer

    Precision 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.
    Reminder

    Time to restock for Luna

    Based on her usual 4-week feeding cycle

    Adult dry food · 10 kg

    Bought 6 times

    €42.99Reorder

    Dental chews · large

    Usually bought together

    €8.99
    Add both to cart
    • +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 brand

    Urgency 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.
    Only 3 left in your size

    Runner 990 · size 42.5

    €129.99

    Get yours now
    • +8.5%Conversion rate upliftSynerise vs. non-personalized variant

    Takeaway

    Same trigger, different product — made individually relevant by inferred interest.

    Section 03

    03

    Personalized Promotions

    AI coupon engine · 1-to-1 offers at scale · Web push and on-site

    Section 03 · Promotions

    From mass promotions to 1-to-1 offers.

    The engine selects the most relevant promotion for each individual customer.

    1. Input

      The available promo pool

      % discount on Product A, B, C — and N more offers.

    2. Engine

      Synerise AI

      Proprietary deep learning. Real-time segmentation incl. last purchase. Cold-start prediction. Continuous testing. No manual adjustment.

    3. 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 chain

    A 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.

    Redemption rate over time · illustrative
    Control groupAI-managed promo pool
    • 19,25%Revenue uplift — up tovs. control group
    • 450%Conversion uplift — up tovs. control group
    • Promo volume expansionAI-managed pool vs. baseline

    Takeaway

    AI promotions hold their effectiveness even as the user base grows.

    Promotions · Cart

    Sports retailer

    Prediction 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.
    Your cart · 2 items

    Velocity Run 3

    Size 42

    €119.99

    Training tee · dry-fit

    Size M

    €24.99
    ✓ Free shipping unlocked — code applied automatically
    Total+ €5.99 shipping€144.98
    Checkout
    • +8.52%Conversion rate upliftSynerise vs. non-personalized
    • +147%Average revenue per userUplift vs. control group

    Promotions · On-site

    Pet retailer

    4,466 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.
    Just for you

    −15% on your next order

    Valid for 48 hours on your favourite categories

    SAVE-15-NOW

    Apply & shop
    • +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.

    1. 01

      Personalized product ranking

      Re-orders PLP results on behavior, preferences and session context to maximize relevance and conversion.

    2. 02

      Merchandising boost rules

      Combines AI ranking with business factors (margin, stock, seasonality), aligning personalization with commercial goals.

    3. 03

      Real-time adjustments

      Automatic re-ranking the moment a product enters the feed. Zero lag.

    4. 04

      Sponsored product placements

      Flexible ad placements inside personalized PLPs — relevance plus monetization.

    5. 05

      Campaign-specific personalization

      Dynamic search and PLP adjustments tied to campaigns — promo boosts plus real-time AI.

    6. 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.

    Search · PLP

    Sports retailer

    Same 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.
    NORTHPEAK
    Selected for yousame URL · per-user order
    For you

    Velocity Run 3

    €119.99

    For you

    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 brand

    Behavioral 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 · Behavioralquery: “razor brand” + user ID

    Pharmacy chain

    Same query, two users, two entirely different results.

    Female user — personalized

    VITAFARM
    razor brand

    Women’s razor · 3-blade

    Inferred from purchase history

    €9.99For you

    Women’s replacement blades · 4-pack

    Same line

    €14.99

    Shave gel — sensitive

    Frequently bought together

    €4.49

    No gender in the profile. The engine infers it from purchase history and shows the women’s line first.

    Male user — personalized

    VITAFARM
    razor brand

    Refill cartridges · 8-pack

    Buys refills regularly

    €24.99For you

    Refill cartridges · 4-pack

    Alternative size

    €13.99

    After-shave balm

    Frequently bought together

    €7.99

    Buys refill cartridges regularly. The engine infers he already owns a razor and shows refills first.

    Zero config — just the user ID. The engine reads purchase history and re-orders results in real time.

    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.

    1. 01

      Already proven in production

      Real A/B results with multiple live merchants. Connects via API and SDK. No replatforming.

    2. 02

      Science that compounds at scale

      Every behavioral event improves every model. As the merchant base grows, accuracy, relevance and targeting improve together.

    3. 03

      Fast to activate, easy to expand

      Start with recommendations. Add search. Layer in promotions — all on the same behavioral data layer.

    We use cookies

    We use cookies and similar technologies to analyze traffic, personalize content, and serve targeted ads. By clicking "Accept", you consent to the use of cookies. Cookie Policy