Scoring Architecture & Evaluation Engine
Multi-vector extraction, deterministic opportunity scoring, and evidence reliability verification across AliExpress listings.
The 6-Phase Product Intelligence Engine
Every research query passes through six rigorous, automated stages before a single product is recommended to you.
Query & Deep Live Extraction
Direct AliExpress Scraping Cluster
You submit a keyword niche (e.g. "ergonomic desk gadget") or paste a direct AliExpress listing link. Our scraper cluster accesses live catalog pages, extracting wholesale price tiers, real shipping rates, and variant matrices without relying on stale databases.
Identical supplier clones collapsed into single canonical listing.
Normalization & Deduplication
Canonical Cleansing Engine
AliExpress listings are infamous for spam keyword stuffing and duplicate factory resellers. Our pipeline strips SEO clutter, converts currencies, calculates true SKU price ranges, and deduplicates identical products into clean canonical entries.
Deterministic Hard Gating
Quality & Volume Barrier
Before invoking AI evaluation, our hard deterministic filters eliminate junk inventory. Listings failing your commercial criteria—such as minimum orders sold, seller rating threshold, or maximum price limits—are immediately pruned.
Confidence Rating: 0.94 • Zero text hallucination
Multi-Vector AI Evaluation
Calibrated Qualitative Extraction
Our calibrated evaluation models analyze every passing product mode-neutrally using a rigid, monotonic 1–4 rubric governed by strict data-integrity rules (“treat missing data as UNKNOWN, do not invent data”). Crucially, quantitative demand is not guessed by the LLM—it is calculated deterministically in Python from verified sales history, while competition is computed from live catalog density.
Deterministic Opportunity Scoring (v1.2)
Weighted Execution + Non-Compensatory Gates
Normalized signals are weighted in server-side Python. Engine v1.2 enforces Non-Compensatory Safety Gates: products with fatal flaws (severe defect risk, bulky freight, or cold demand <100 orders) are hard-capped at 40–55 points max, regardless of video virality. Omitted dimensions are excluded with dynamic weight renormalization, and ranking ties are broken deterministically.
Ranked Results & Action Plan
Actionable Ecommerce Deliverable
Products are ranked via deterministic tie-breaking. You receive complete commercial deliverables: instant unit economics featuring our default 2.5x retail markup baseline (audited as assumed markup unless overridden by you), gross margin cushions, break-even CAC targets, and direct supplier links.
What the AI Evaluates Behind Every Product
Instead of arbitrary guesses, our engine scores 9 concrete commercial metrics that dictate ecommerce success or failure.
Perceived Demand
DeterministicComputed deterministically from verified lifetime orders: <100 (Score 1, capped at 40 max), 100–499 (Score 2), 500–1,999 (Score 3), ≥2,000 (Score 4). Prevents hallucinated demand.
Problem Solving
Evaluates whether the item provides tangible functional relief (pain relief, time saving, posture correction) versus useless gimmick novelty.
Video Potential
Scores visual demonstration power on TikTok, Instagram Reels, and YouTube Shorts. Evaluates 3-second hook strength and before/after impact.
Impulse Buy Factor
Analyzes whether the retail sweet spot ($19.99–$39.99) enables immediate checkout without requiring multi-day comparison shopping.
Ad Creative Potential
Measures creative diversity for media buying. Assesses whether multiple creators can film distinct UGC angles (unboxing, comedy skit, problem demonstration) without creative fatigue.
Differentiation
Measures uniqueness compared to generic shelf commodities found at Walmart, Target, or saturated Amazon Prime top results.
Dropship Margin
Calculates shipping feasibility: lightweight under 500g, non-volumetric packaging, cheap packet shipping, and a markup of at least 2.5x.
Low Competition
Catalog SignalComputed from search token overlap in the research run: ≤3 competitors (Score 4 • Blue Ocean), 4–8 (Score 3), 9–15 (Score 2), >15 (Score 1 • Saturated). Excluded cleanly if catalog context is missing.
Low Defect Risk
Safety GatedInverted & Gated: Screens against battery hazards, ingestible liabilities, fragile glass, or apparel sizing nightmares. A score of 1 triggers a hard 50-point score cap.
See How Raw Data Becomes Actionable Intel
The stark difference between manually browsing AliExpress listings versus running them through Product Score.
"2026 Hot Sale New Arrival Multifunction Car Seat Gap Storage Box Organizer Auto Interior Storage Pocket Coin Phone Cup Holder Universal Fitment..."
- ✕ Zero indication whether ads are already saturated
- ✕ No video hook or marketing angle provided
- ✕ Unknown transit damage risk or returns rate
Leather Seat-Gap Phone & Cup Caddy
✓ Normalized SKU: $4.20 Wholesale- ✓ Viral Hook: "Dropping your phone in the car seat crevice" problem-solver.
- ✓ Impulse Sweet Spot: Frictionless $27.99 buy with 2-pack bundle potential.
- ✓ Supplier Verified: 97.4% positive feedback, 7-day Cainiao delivery.
Why Product Score Outperforms Traditional Tools
Engineered for serious ecommerce operators and scaling agencies who cannot afford to waste ad budget on stale databases or hallucinated AI ratings.
| Architecture Key Capabilities |
★ Next-Gen Platform
Product Score
|
Traditional Spy Tools Minea, Dropship.io, etc. | Manual Browsing AliExpress + Spreadsheets |
|---|---|---|---|
|
🔄
Data Freshness
Real-time supplier pricing vs. out-of-date archives. |
✓ Live Real-Time Scrape
Extracts live wholesale prices, active SKU inventory, and Cainiao shipping rates directly upon every query. |
✕ Stale Databases
Days-old cached archives; out-of-stock items still show as active winners. |
⚠️ Unstructured
Fresh but manual; requires clicking across dozens of messy store pages. |
|
🧠
AI Evaluation Architecture
Decoupled rubrics vs. generative text hallucination. |
✓ Decoupled 9-Dimension Engine
Rigid 1–4 qualitative rubrics evaluated by AI, then scored via deterministic Python math. 0% hallucination. |
✕ Black-Box Vanity Score
Single fixed number (e.g. '89/100') with undisclosed origin or arbitrary calculation. |
✕ Pure Intuition
Emotional gut feel; inconsistent results across different team members. |
|
⚖️
Custom Scoring Formulas
Tailoring algorithms to specific business models. |
✓ 6 Research Modes + Custom Formulas
Switch between General / Blended, TikTok Ads, Meta (FB/IG), Niche Store, High Margin, Google Shopping modes, or build custom weighted formulas (Pro Creator & Agency). |
✕ One-Size-Fits-All
Forces the same generic ranking onto every business regardless of channel. |
⚠️ Excel Formulas
Requires building and maintaining custom spreadsheet calculation models. |
|
🛡️
Defect & Transit Risk Indexing
Protecting net margins against return rates. |
✓ Inverted Hazard Defense
Inverted scoring automatically flags fragile glass, battery freight restrictions, and high-return apparel sizing. |
✕ Zero Risk Indexing
Looks only at ad views; completely ignores supplier defect rates or breakage. |
⚠️ Hours of Reading
Requires skimming hundreds of translated 1-star reviews to detect flaws. |
|
🔗
Direct AliExpress URL Audit
Auditing specific supplier items discovered elsewhere. |
✓ Instant Link Audit
Paste any AliExpress link to immediately trigger full 9-dimension scoring, margin analysis, and video hooks. |
✕ Ad Library Only
Can only search products already detected by their ad scrapers. |
⚠️ Manual Calculation
Must manually calculate margins and guess marketing angles. |
|
💻
Developer REST API Access
Programmatic headless automation for scaling agencies. |
✓ Full Programmatic API
Generate API keys, query endpoints with custom formula IDs, and stream structured JSON into internal tools. |
✕ Expensive Add-On / None
API locked behind expensive custom sales calls ($1,000+/mo) or unsupported. |
✕ Not Available
No programmatic automation possible. |
|
🎯
Mathematical Reproducibility
Auditability and consistency across team members. |
✓ 100% Invariant Guarantee
Identical inputs and formula configurations always produce the exact same score. Full auditable transparency. |
✕ Drifting Rankings
Scores change unpredictably without clear criteria or mathematical trail. |
✕ High Human Bias
Evaluations depend entirely on who is reviewing the product that day. |
Frequently Asked Questions
Everything you need to know about the platform, data integrity, and subscription plans.
Do I need an AliExpress API key or developer account?
No. Product Score collects the AliExpress catalog data for you. You don't need a developer account, API credentials or proxies: enter a keyword or a product link.
How does this differ from asking ChatGPT to evaluate a product?
Asking a chatbot for a score out of 100 twice usually gives two different answers, and it can overlook shipping or return risks. Product Score separates the AI's job (rating each dimension on a fixed 1–4 rubric) from the maths (a deterministic formula on our servers), computes demand from real order counts, and applies safety gates that cap the score of products with serious risk, however viral they look.
Can I audit a single AliExpress product link directly?
Yes. Paste an AliExpress product URL into the research form. The engine extracts that product, rates it and returns a full Opportunity Score report with unit economics.
What are Custom Scoring Formulas and who can use them?
Custom Scoring Formulas let you set your own weights across all 9 dimensions, for example weighting video potential higher for TikTok. Every plan can use the 6 built-in research modes; Pro Creator & Agency subscribers can also build and save their own formulas.
Can I integrate Product Score into my own workflows via API?
Yes. Agency subscribers get REST API access: generate API keys in your profile, start research jobs programmatically and feed scored products into your own dashboards, Airtable or store automations.
Find Winning Products Backed By Hard Math
Join pro ecommerce operators and dropshipping agencies who test products with confidence. Free tier available—no credit card required.