Real Data From 50,000 Product Screens on 1688

Before opening Scout to external users, we used it for our own operations for months. By May, total product screens exceeded 50,000.

This is not a vanity metric. This is working data, and it exposed a few things we did not expect.

Actual pass rates are much lower than expected

When we started screening batches from 1688, we expected roughly 20-30% of products to clear the first filter. Reality: an average of 8-12%.

That does not mean 1688 is weak. It means the screening process is more rigorous than we assumed. Processing 500 products to get 45-60 viable candidates worth testing is normal, not a failure.

If you are screening batches and your pass rate is above 25%, your filters are likely too loose.

Seller count is a more reliable signal than reviews

On 1688, product ratings and review counts do not accurately reflect real competitive dynamics when selling in the Vietnamese market.

The signal we found correlates better: the number of sellers listing the same product (or similar ones) on Shopee and TikTok Shop. A product with 5 sellers on Shopee, each doing 200 orders per month, typically has better margins than a product with 1 seller doing 1,000 orders per month. The latter is holding the market and will undercut as soon as a competitor appears.

We added this field to our annotation workflow after recognizing that pattern.

"Dead" products have cycles

About 15% of products we rejected in an initial batch reappeared in batches 3-4 months later, usually from a different supplier, sometimes with minor design improvements.

This is why the decision history feature in Scout turned out to be more important than we initially thought. Not to avoid revisiting products, but to know why we passed on them last time, and whether that reason still holds.

The optimal batch is not the largest batch

We tested various batch sizes: 100, 300, 500, 800 products. The result was consistent: batches of 200-350 products produced the best decision quality.

With batches above 500, we noticed the "maybe" rate climbing. Not because products were genuinely borderline, but because decision fatigue sets in toward the end of the batch. Products that clearly deserved a "pass" or "fail" started getting tagged "maybe" after 4 hours of screening.

Two batches of 250 across two mornings consistently outperformed one batch of 500 in a single long day.

What we are doing with this data

Part of building Scout was being our own test subjects: using the tool ourselves and documenting every pain point.

The result is a set of features driven directly by what the data showed: automatic batch grouping, warnings when a batch is too large, and quick filters to eliminate products by group rather than one at a time.

If you are building a product screening workflow from 1688 and want to try Scout, request access here or write to thinh@ordinex.io.