Pricing Thousands of Items Daily From eBay Sold Data: The Real Cost (and How to Cut It)

A developer building a card scanner laid out the math in August 2026: "I need to fetch daily prices of 8000 cards on eBay with different variations such as Raw, PSA 8 9 10 .. which makes it 40000 request." Another wanted sold listings "for approximately 15k different products over the last 90 days."

This is the typical scale for a card, collectibles or catalog app. Here's what it costs, and how to need far less of it.

#The naive plan: refresh everything daily

8,000 cards × 5 versions × 1 refresh a day = 40,000 lookups a day, about 1.2 million a month.

Most of those lookups re-buy an answer you already had. A card that sold 4 times in 90 days doesn't have a new price every morning.

#Strategy 1: refresh when someone looks

Store each price with a timestamp. When a user opens a card, refresh it only if the stored price is older than 24 hours. Nobody looks at most of a catalog on a given day, so lookups follow real usage instead of catalog size. Typical result: 5,000–8,000 a day for the same 40,000 card-versions.

#Strategy 2: tier by popularity

• Hot (top ~2% by views or sales volume): refresh daily.
• Everything else: every 3 weeks. Slow items rarely change, and Strategy 1 still refreshes anything a user actually opens.

40,000 card-versions with 2% hot: 800 daily + 39,200 ÷ 21 ≈ 1,870, so about 2,700 a day, roughly 80,000 a month.

#Strategy 3: skip versions nobody sells

If a card has had zero PSA 8 sales in 90 days, there's nothing to refresh. Re-check empty versions weekly at most. For many sets this removes a third of all lookups.

#What it costs

With one request per card-version per refresh, on SaleHistory pricing:

Scale includes 60,000 requests; extra requests use prepaid credits at $2.50 per 1,000. At the tiered volume that's about $0.0016 per lookup, and the same strategies work whatever data source you use. Pricing.

#Scraping it yourself at this scale

Builders who tried report the same costs: residential proxies at "at least 2€/GB" with around "50% success rates," CAPTCHAs, layout changes, and, since mid-2026, needing signed-in sessions that risk the account you sell on. And when it fails, it often fails quietly. Read why empty results are the dangerous failure.

#Implementation sketch

from datetime import datetime, timedelta

TTL = {"hot": timedelta(days=1), "rest": timedelta(days=21)}

def price(card_version):
    row = db.get(card_version)
    if row and datetime.utcnow() - row.fetched_at < TTL[row.tier]:
        return row                      # cached, no request
    data = sold(card_version.query)
    if data["count"] == 0 and row:      # a real "no sales": keep last price, mark it old
        return db.touch(card_version)
    return db.save(card_version, data["stats"], fetched_at=datetime.utcnow())

One query per version is still the rule. See how to comp trading cards for the query patterns.