Case study: how ScanToSold took us from 6 to 22 listings an hour

The eBay side of CB5StarSales has one bottleneck that dwarfs every other: listing throughput. Sourcing inventory is easy. Shipping is solved. But every item between those two steps needs a title, a price, a condition grade, and competitive research — and for a long time, all of that was manual.

The before picture

The old flow for a single flip: look up the UPC, open three sold-listing tabs, eyeball comparable prices, draft a title that doesn't keyword-stuff, write condition notes, publish. Ten-plus minutes per itemon a good day. Across a sourcing haul of 20 items, that's an entire afternoon — and it produced roughly 6 listings an hour at sustained pace.

The pipeline

ScanToSold collapses the research loop into a single scan. Point the phone at a barcode and the backend runs three things in parallel:

• an eBay Browse API query for sold comparables, returning a price band and sell-through estimate;
• a title generator that builds a keyword-optimized title from the product metadata and what the top-performing comparables actually used;
• a condition-grading prompt that turns a few taps into clean, buyer-readable condition notes.

The result comes back in under three seconds as a draft listing. The human's job shrinks to: glance, adjust price if the band looks off, publish.

The after picture

Median throughput went from 6 to 22 listings an hour — about 3.7×. The interesting part is what that did to sourcing decisions: items with thinner margins became worth listing at all, because the labor cost per listing dropped below their expected profit. The catalog got broader, not just faster.

The store now sits at 2,800+ items sold with 100% positive feedback, and the listing step hasn't been the bottleneck since.

What I'd tell anyone building the same thing

• Attack minutes, not features.The win wasn't AI for its own sake — it was deleting the three-tab research ritual.
• Parallelize the API calls. Sequential lookups felt instant in dev and took 9 seconds in the field. Fan-out made it sub-3.
• Keep the human on price. Full auto-pricing was tempting; a human glance at the suggested band catches the weird cases (bundles, damaged boxes, collectibles) that tank feedback scores.

ScanToSold is live at scantosold.com. If you're a reseller drowning in the listing step, get in touch.