AI Caught Fake Pricing — When 'Cheaper' Doesn't Mean 'Available'

An E-commerce client told us "our competitor is cheaper on almost everything." So we built an AI system to check automatically — and found something unexpected.

The Problem: Manual Price Checking Doesn't Scale

The client sells 500+ SKUs. Multiple competitors, each changing prices daily. Comparing products one by one across multiple stores would take a full day of manual work.

And even then, the comparison would be misleading — because a lower price doesn't mean the product is actually available.

Solution: AI Pipeline in 6 Steps

We built a single script that runs the entire pipeline in 20 seconds:

1. Scrape client's products (175 items)
2. Scrape competitor's products (468 items) via API
3. Auto-match products — brand + model + type
4. Compare prices for every match
5. ★ Check stock — is the product actually available?
6. Write report to Google Sheet with color highlights

Why "Matching" Is Hard

The same product has different names across stores:

The AI has to normalize names, extract brand names, parse puff counts from multiple formats (20000, 20K, 20000 Puffs), and match based on:

1. Brand — dictionary of 60+ brands with aliases
2. Puff count — exact match or ±10% tolerance
3. Sub-type — cartridge vs device vs starter kit (must be same type)

Out of 175 x 468 = 81,900 possible pairs, the AI found 50 actual matches.

Plot Twist: Fake Pricing Detected

The initial comparison looked alarming:

29 items where the competitor listed a lower price... scary, right?

But when we added step 5 — actual stock checking — the picture changed completely:

From 29 apparent losses → only 15 real ones (48% less panic)

All 14 "cheaper" items had zero stock — every single variant showed 0% availability. We tested 19 items manually: stock = 0 on all of them.

Red / Green / Grey Framework

We created a simple framework so the client can read price comparisons instantly:

Google Sheet Output

The pipeline writes results to Google Sheet automatically with 4 tabs:

Red rows = act immediately Grey rows = ignore (fake pricing) Green rows = you're safe

The client opens the Sheet and knows exactly what to do — no analysis needed.

Technical Notes

Supported Platforms

Shopify is much easier — the JSON API gives you products + variants + stock in one call. WooCommerce requires HTML parsing page by page, but it works.

Product Matching Algorithm

# Simplified matching logic
1. Extract brand (60+ brands, longest-match-wins)
2. Extract puff count (regex: "20000 Puffs", "20K", standalone digits)
3. Check sub-type (cartridge vs device vs kit)
4. Score: exact puffs = 100, ±10% = 80, brand only = 70
5. Accept matches >= 70

Match accuracy is roughly 85-90% — some products have names too different to match automatically. We keep adding aliases to improve accuracy over time.

Results