Deckers is a DTC growth engine — on a foundation you're rebuilding.
01
Legacy SFCC monolith.Heavily customized, expensive to extend — and mid-replatform.
02
Four data warehouses, no single source of truth.You can't scale what can't be fed clean data.
03
The foundation rebuild owns the budget.Most spend and focus, rightly, goes there.
04
Loyalty just launched.But login rates are low — the site stays transactional and anonymous-heavy.
02 — Why it matters
Malachyte × Deckers
Personalization drives the things Deckers cares about most.
01 — Margin
Full-price sell-through
Matching the right product to the right customer at the right time maximizes full-price sell-through — fewer markdowns, more first-look conversions.
02 — Membership & lifetime value
Loyalty that compounds into LTV
Relevant, personalized experiences drive success in the loyalty program you just launched — turning anonymous intent into known members, and making DTC the engine of customer lifetime value rather than just first-order revenue.
03 — Performance
High-performance DTC
Better search and personalization move every e-commerce metric in the right direction — conversion, AOV, and revenue per visitor — turning the storefront into a high-performance growth engine.
How Malachyte works
Like a language model — but for shopping intent
The analogy — predicting the next token
Language model Next word“She”“wore”“a”“long…”?“coat”
MiQ treats each shopper interaction as a signal — predicting intent, not just similarity.
Personalizes from the very first click — no login, cookies, or purchase history needed
Understands intent, not just similarity — every click reshapes the next prediction
Works for anonymous and logged-in shoppers alike
Malachyte
04
Why current technology falls short
Conventional search and personalization can't keep up
Built for a different reality — they leave three blind spots on the traffic that matters most.
01
Batch retraining lags
Models train offline and sit static until the next run. A visitor who shifts intent mid-session keeps seeing yesterday’s guess.
02
Search and recs, siloed
Different vendors, different models, no shared view of intent. A search and a click never inform each other.
03
The anonymous majority
Collaborative filtering needs an identity. 90–95% of traffic doesn’t have one — so it defaults to a generic shelf.
Malachyte
05
The architecture advantage
One user vector. One brain. Every surface.
MiQ
SearchRecommendationsPLP rankingCartEmailMerchandisingAgent API
Legacy stack
With Malachyte
3 vendors, 3 models, contradicting signals
1 vector, 1 brain, all surfaces
Batch updates (overnight)
Continuous learning (milliseconds)
10–15% of traffic personalized
100%, from the first click
Black box
Merchandiser control surface
Malachyte
06 / 22
MalachyteDeckers
DTC has to re‑accelerate
without spending more on paid acquisition.
The unworked lever is revenue‑per‑visit across the surfaces shoppers actually land on - the homepage, PLP, and PDP - moving the LTV side of CAC : LTV.
STATIC DISCOVERY
~60+%
of Deckers' monthly visits come direct or via search (organic + paid), per SimilarWeb. Most land from a Google result, cold.
Where they land
Homepage, PLP & PDP
Direct, search, and Meta DPAs land shoppers across the homepage, category pages, and product pages — every high‑intent entry surface, not just a search bar.
The lever
RPV
Lifting revenue‑per‑visitor on traffic you already pay for moves the LTV side of CAC : LTV — drives full-price sell-through and lifetime value.
Three possible starting points
Each unlocked by the same underlying generative user vector.
01 — Taste onboarding
For loyalty members.
Personalize onboarding flows and surface custom components by loyalty status, points, and program tier. Roughly half of Deckers' revenue lives here.
Example — a "10% off all trail shoes" perk for loyalty members surfaces as a tailored module that displays only for the right cohort, with the right products, at the right moment.
02 — COLD START ACTIVATION
For new visitors
Cold start — activate a profile in seconds using traffic source, landing context, device, time of day, and geo. Adapt homepage, PLP, and PDP to that visitor's intent from the first click.
Output — an enriched per-visitor profile that grows with every action and persists across sessions.
03 — Attribution engine
For merchandising and ops.
Simulate and forecast the impact of personalization rule changes before deploying via offline - online correlation (unique to Spotify). Increase A/B test velocity, win rate, and magnitude of impact across the board.
Lighthouse opportunity — Q4 capability where Deckers could be the design partner.
Malachyte
One profile · every visit · every surface
In-Session Personalization that activates before the first click.
No sign-in. No third-party cookies. No waiting for cohorts to materialize. The moment a shopper lands, a generative user vector starts forming — and sharpens with every action they take.
Trusted by brands such as:
Activates for
100% of traffic
Personalized from the first pageview — every visitor, no sign-in or cohort wait.
Continuously learns from
Every action
Search, browse, click, add — every signal counts toward re-rank.
Persists between
Visits
They return; the store remembers where they left off.
Decision latency
<200ms
SLA from launch — re-ranks fast enough to feel intuitive, not engineered.
Malachyte
Where the visit actually begins
Most paid visitors start on a product page — cold.
hokaSponsored
HOKA Skyward X · Everyday
Shop now
DPA / Google Shopping
hoka.com / skyward-x
HOKA Skyward X
$180 · Everyday · Men's
Add to cart
Above fold
Below fold · generic
Bestsellers
01
Straight to the PDP.
The ads that actually perform are Dynamic Product Ads — hand Meta the whole catalog, it sends her straight to a product page. The PDP is the first impression.
02
Zero history, generic page.
Above the fold is the product; below the fold is generic bestsellers — identical for every visitor. Nothing reflects who just arrived.
03
One shot at relevance — missed.
She bounces. The single highest‑intent moment of the visit — the page she was sent to — was generic.
Malachyte
What's possible · the cold‑start PDP, re‑ranked live
Same landing page. Two visitors. Personalized in <200ms.
hoka.com / skyward-x
Above fold · the ad's product
HOKA Skyward X
$180 · Everyday · Men's
Add to cart
For your everyday rotationPersonalized
Skyflow
$128
Skyward X
$180
Stinson 7
$175
Challenger 8
$155
Under the hoodPDP re‑ranked for her
This visitor
Google Search → lifestyle shopper
Lifestyle / everyday0.91
Cushioned comfort0.76
Race / tempo0.24
Revenue / visit
$0.48
+7.1% vs generic
Activates before the first click
No sign‑in, no third‑party cookies, no waiting for cohorts. The vector forms the instant she lands on the PDP.
Re‑ranks the page underneath
Above the fold stays the ad's product. The open canvas below re‑orders to her taste — live, per visitor.
Lifts revenue per visit
The highest‑intent moment of the visit finally reflects who arrived — on traffic you already pay for.
Malachyte
A user vector that warm starts every session before the first click.
An anonymous visitor lands on a PDP from a Google Search result — the vector adapts in <200ms.
New visitor
LIVE
visitor_5f8a·2e91
first session | en‑US | iOS | Brooklyn, NY referrer: google.com · Search
Trail‑kit affinity
0.91
Style · trail
0.83
Surface · technical trail
0.76
Gift intent
0.22
Price sensitivity
0.34
Within seconds, the user vector is adapting — re‑ranking her PDP toward trail socks and hydration vests, the gear nearest her intent, ahead of generic best‑sellers before her next click.
Vector space · 2D projection80 dims
Session:4 signals
Landed on · Speedgoat 6 PDP
via Google Search
Referrer · Google
organic search result
Device · iOS 17, 390w
mobile · high‑DPI
Geo · Brooklyn, NY
Tue 9:42p ET
Malachyte
What the vector knows — without an account.
An adaptive profile that represents every anonymous visitor preference and intent. Your merchandisers control it; the customer experience is continuously powered by the user vector underneath.
Visitor profileLive · anonymous
first visit · live session
Style
Everyday trail runner
confidence 0.83
Cushion preference
Max‑cushion foam
over minimal, plated
Price tier
Performance · $145–230
self‑purchase · performance intent
Category affinities
Trail shoes
0.91
Hydration vests
0.74
Road shoes
0.29
Purchase likelihood
0.58
↑ from 0.16 at session start
Malachyte
RPV with significance. Attribution per surface.
Aggregate lift tells you nothing about what to do next week. Direct, per‑surface attribution shows exactly where the value comes from — the cold‑start PDP, its below‑fold merchandising, and lifecycle email.
Revenue per visit · personalized PDP (cold‑start)
$3.42
Control $3.13 · +9.3% RPV
Revenue per visit · PDP below‑fold merchandising
$2.31
Control $2.16 · +6.9% RPV
Revenue per visit · lifecycle email (Klaviyo)
$1.07
Control $1.00 · +7.0% RPV · keep testing
Direct attribution
Credit goes to the surface that actually changed the purchase path — where the visitor interacted, not the last banner they scrolled past.
01Lands on a PDP · Meta DPA, no historyPDP landed
02PDP re‑ranks below‑fold for her · Skyward XPDP · personalized
03Clicks a piece surfaced for her · adds $180BELOW → +ADD
04Drops off · vector‑driven email returns herEmail
05Checkout · $180 · attributed to PDPDirect
Every rec surface reports its own RPV, CTR, and add rate. A rec only gets credit when the visitor actually engaged with it.
Malachyte
Where Malachyte is built differently.
No vendor category solves all of this. Malachyte covers what cross‑vertical engines, point‑recommendation tools, search vendors, and LLM startups each only partially address.
Buyer evaluation criterion
Malachyte
Premium engines
Bloomreach, Constructor
Recommendation vendors
Dynamic Yield · Nosto
Search vendors
Algolia · Coveo
LLM vendors
Marqo · Cimulate
Cold start + in‑session
User Vector that adapts in milliseconds
✓
✕
✕
✕
✕
Unified discovery
User vector that ranks/re-ranks across search + recs
✓
✕
✕
✕
✕
Merchandising
Full platform suite of controls
✓
✓
✓
✓
✕
Speed to value
Integration & onboarding
✓Weeks
✕Months
✕Months
✕Months
✓Weeks
Marketing automation
Copilot, real‑time optimization
✓
✓
✕
✕
✕
Malachyte
Multi‑Agent Continuous Learning Architecture
Custom trained and integrated, it continuously learns across every data source — replacing the slow, manual CDP query loop with attention mechanisms for in-session adaptation in milliseconds.
System of Record
Salesforce Commerce Cloud · Klaviyo · BigQuery
Real‑Time Data Processing
Event streams · sub‑200ms
Transformer‑backed World Model
LLM‑backed Catalog Encoder
(Deckers SKUs)
Low Latency Retrieval
Relevance Microservices
(e.g. Search, Recommendations)
Multi‑Objective Balancer
Unified Expansion
(e.g. Merch, Marketing)
Deckers Applications
web · app · CRM · paid social
Every Deckers interaction feeds the vector. We're ready by the next click.