Shopee Data Scraping Services

Voucher stacking makes effective price genuinely hard

This is the defining difficulty in Shopee data, and most datasets simply ignore it. Several discount layers can apply to one listing:

• Listed price after any seller-set discount.
• Platform vouchers issued by Shopee, often with minimum spend conditions.
• Shop vouchers issued by the individual seller.
• Coin discounts from the loyalty mechanic.
• Shipping subsidies and free-shipping vouchers, which are effectively price reductions.

We capture each visible layer separately with its conditions, and compute a best_visible_price under stated assumptions rather than a single "real price" we cannot verify.

The honest limit: whether a shopper can actually stack a particular combination depends on their account, claimed vouchers and basket composition. We report what is publicly displayed and what conditions attach, and we flag where a voucher's applicability could not be determined. Presenting one confident effective price here would be inventing a number.

Shopee Mall and marketplace sellers are different populations

Shopee Mall listings come from brands and authorised retailers under stricter platform requirements. Ordinary marketplace listings come from anyone.

Merging them corrupts almost every analysis you would want to run. Price distributions are wider on the marketplace side, counterfeits and grey stock concentrate there, and a brand's competitive picture looks very different depending on which population you measure.

We capture listing_type and seller identity, so brand monitoring can focus on the marketplace population while price benchmarking can focus on Mall listings. For brand protection specifically, the marketplace population is where the work is, and it is the population that a naive category crawl averages away.

Flash sales are scheduled states, not price changes

Shopee runs heavy campaign and flash sale activity on scheduled windows. A price drop during a flash window is not the same commercial event as a permanent reprice, and treating both as "price change" produces a dataset that reports constant volatility with no pattern.

We capture flash sale as a state with its window, so flash frequency, depth and timing by category become measurable. Campaign dates — the double-digit date events that dominate SEA ecommerce — are recorded as context on every record collected during them.

That context field matters more than it sounds. A price index computed across a campaign date without flagging it will show a market-wide crash that is a scheduled event, and comparing it to a non-campaign period is comparing two different market states.