How AI Matching Achieved 99% Accuracy on 200,000 Complex Motorcycle Parts
Matching accuracy
Unreliable (frequent wrong matches) → 99% accuracy across 200,000 SKUs
Profitability
Baseline → +20-25% increase
Revenue
Baseline → +30% growth
A French motorcycle parts and equipment e-commerce seller operating on PrestaShop with Google Shopping integration. They carry over 200,000 highly specialized SKUs including brake cables, bearings, gaskets, and accessories with complex variant structures (model-specific, year-specific, specification-specific). They are also a myFulfillment user, which provides native integration for cost and stock data.
The Challenge
Product matching for motorcycle parts is a nightmare. A brake cable for a Triumph Bonneville looks similar to one for a different Triumph model, but using the wrong competitor price for repricing leads to incorrect adjustments and eroded trust in the tool. At 200,000 references, manual matching was physically impossible. EAN codes don't reliably exist for many aftermarket parts. No API provides this data. Previous automated pricing attempts failed precisely because the competitive data was unreliable. Wrong matches meant wrong prices, and wrong prices meant margin loss or BuyBox failures. The customer needed matching they could trust before they could automate anything.
Top challenges:
- 200,000 complex motorcycle parts with no reliable EAN or API matching
- Wrong product matches making automated pricing dangerous and untrustworthy
- Manual matching physically impossible at this catalog scale
The Solution
We deployed the full MyMatching pipeline alongside myPricing.
What We Configured
Step 1: 200,000 products were imported from Google Merchant Center into myPricing. Step 2: Each product was sent through the MyMatching engine, which searches Google Shopping and processes results using image comparison (product photos vs. search results), text/semantic analysis (product attributes, titles, specifications), and LLM-based arbitration (fine-tuned AI agents that make the final match decision with confidence scoring). Step 3: Confidence thresholds were configured -- above 96% auto-approved, 90-96% flagged for human review. Step 4: The customer added custom AI instructions for their product niche, teaching the model that certain attributes (like specific power ratings or material specifications) were mandatory matching criteria.
What We Custom-Built
Custom AI instructions allowed the customer to fine-tune the matching model for motorcycle part specifics. When the generic model missed an important attribute (e.g., treating a power specification as optional when it was critical), the customer added a plain-text instruction directly in the application, no technical skills required. This iterative refinement is what pushed accuracy from good to 99%.
The customer uses myFulfillment for warehouse management, giving myPricing direct access to accurate cost data (Weighted Average Cost). Combined with AI matching, myPricing gets accurate costs and accurate competitor data in the same system, so pricing decisions are based on reality. The customer also added custom AI instructions for niche product attributes that the generic model initially missed.
Key benefits:
- AI matching with 99% accuracy using image, text, semantic analysis and LLM arbitration
- Custom AI instructions let the customer fine-tune matching for niche product attributes
- Automated pricing enabled after 6 months of validated matching: +20-25% profitability, +30% revenue
The Results
Matching accuracy
Unreliable (frequent wrong matches) → 99% accuracy across 200,000 SKUs
Profitability
Baseline → +20-25% increase
Revenue
Baseline → +30% growth
After 6 months of validating matches and seeing consistent accuracy, the customer activated automated pricing. With real competitor data they could trust, and accurate cost data from the myFulfillment integration (Weighted Average Cost), pricing decisions became reliable and automated. Results: 99% matching accuracy across 200,000 complex parts, 20-25% profitability increase, and 30% revenue growth. The weekly review process for 90-96% confidence matches takes one team member a fraction of their time, making the entire system largely self-running.
Understand this case in video
Not just for motorcycle parts
Any seller with complex, variant-heavy catalogs where standard matching methods fail benefits from AI-based product matching.
- Auto parts distributors with model-specific, year-specific components across hundreds of thousands of references
- Industrial supplies sellers where technical specifications determine product equivalence
- Private label sellers who need to find competing branded equivalents for their products
- Organic/natural cosmetics manufacturers identifying competitive equivalents from other brands
Check out more
How We Built an Omnichannel B2B2C System Connecting Retail Stores to E-Commerce Inventory
E-commerce + retail storesRead storyHealth supplements & cosmeticsHow a Health & Wellness Retailer Achieved 100% Lot Traceability with FIFO Enforcement
800+ SKUs with expiry datesRead storyThird-party logistics (3PL)How We Built Error-Proof Warehouse Workflows for a Lyon-Based 3PL
Local e-commerce clientsRead storyHunting & outdoor equipmentHow a Hunting Equipment Specialist Sells the Full Warehouse Catalog From Any Store Counter
2 physical locations + online storeRead storyLifestyle & seasonal goodsHow a Lifestyle Retailer Eliminated Stockouts with Intelligent Replenishment Forecasting
12,000+ SKUsRead storyHigh-tech / Consumer electronicsHow AutoPilot AI Focuses Repricing on the Products That Actually Matter
~1,000 SKUs, multiple marketplacesRead storyParapharmacy / Health & wellnessHow Inverted Pricing Reports Turned Market Data into Supplier Negotiation Leverage
5,000-6,000 SKUs, Shopify + Google ShoppingRead storyBedroom products e-commerceHow We Replaced Scissors with Algorithms for a Multi-Country Bedroom Products Seller
5 Shopify sites + 20 marketplaces, FR + ESRead storyAutomotive body parts e-commerceHow We Automated Multi-Supplier Cross-Dock for an Auto Parts Distributor with 500K SKUs
200K-500K SKUsRead storyElectrical equipment (B2C + pros)How We Built a Custom Cable Cutting Workflow for an Electrical Retailer
B2C + professional electriciansRead storyHunting & outdoor equipmentHow a Hunting Equipment Specialist Deploys 5 POS Terminals in Minutes at Trade Shows
2 physical locations + online storeRead storyE-commerce (multi-industry)How 50% of Our Clients Automate Cross-Dock and Dropship Supplier Fulfillment
50%+ of myFulfillment clientsRead storyHome & garden e-commerceHow a Multi-Channel Retailer Replaced Excel Chaos with One-Screen Order Management
5 sales channelsRead storyGifts & gadgets retailHow a UK Retailer Turned 5 Stores Into Omnichannel Fulfillment Nodes
5 physical stores + Shopify onlineRead storyHome equipment retailHow We Guaranteed Profitability Across 10+ Marketplaces for a Home Equipment Retailer
10+ marketplaces, thousands of SKUsRead storyGifts & gadgets retailHow a UK Retailer Escaped Shopify POS and Unified 5 Stores on One Platform
5 physical stores + Shopify onlineRead storyHome & garden equipmentHow a Brittany Retailer Turned Warehouse Walk-Ins Into 25% of Revenue
€3M revenue, 10+ marketplacesRead storyBio & organic cosmeticsHow Similar Matching Gave a Bio Cosmetics Brand Competitive Visibility for the First Time
5,000+ SKUs, Magento + Google ShoppingRead storyFashion & accessories DTCHow a DTC Fashion Brand Re-Internalized Logistics from Their 3PL in 3 Days
2,000+ orders/monthRead storyCandle & fragrance manufacturingHow BOM-Based Stock Calculation Tripled Sellable Inventory for a Candle Manufacturer
4 e-commerce sitesRead storyElectronics e-commerceHow an Electronics Distributor Eliminated 90% of Label Errors with Centralized Shipping
3,000+ daily shipmentsRead storyGardening & pet supplies retailHow a Gardening Retailer Got 34% More Clicks Without Raising Ad Spend
25,000+ SKUs, 50+ year-old family businessRead storyReady to see similar results?
Book a demo and we'll show you how we can solve the same challenges for your business.
Free 20-min call · No commitment