Surplus Acquisition Pricing Engine

Client: Internal • Industry: Industrial Surplus / Asset Trading • Completed: June 3, 2026
Python Machine Learning Auction Data Unit Economics Web Scraping Pricing

Challenge

Surplus and liquidation equipment is a margin business: the money is made on the spread between what a lot clears for at auction and what it resells for, net of every cost in between. The hard part is that those costs are invisible at bid time. A buyer’s premium can run 7–19% depending on the venue, sales tax depends on the seller’s state and whether a resale certificate is honored, freight depends on weight and distance, and the resale side gets clipped again by marketplace fees and returns. Bid off the hammer price alone and a “good deal” quietly becomes a loss. Across thousands of live lots spanning IT hardware, electrical gear, test equipment, and more, no broker can manually price each one fast enough to act before the listing closes.

Solution

Built the buy-side intelligence layer for a surplus acquisition operation — a local-first engine that finds underpriced lots, prices them per vertical, and tells the operator the exact maximum bid that still clears a target margin.

  • 26-source scraper network running as a daily batch across government auctions (GovDeals, GSA, PublicSurplus, Municibid), DoD/DLA surplus (GovPlanet/IronPlanet), corporate and returns liquidation (Go-Dove, B-Stock, Liquidation.com, Heritage Global, PropertyRoom), thousands of regional auctioneers (Proxibid, HiBid), and eBay sold comps — with hard freshness gates that fail the pipeline before stale data can reach a bid decision
  • 17 per-vertical fair-value models (quantile gradient-boosted regressors) producing low-bid / fair-value / high-ask bands across GPUs, servers, IT networking, electrical, test & measurement, UPS, HVAC, enterprise SSD, optical transceivers, InfiniBand/NVLink, PLC/automation, and more — each tuned to how its category actually depreciates
  • Full profit-model cost stack that layers buyer premium, seller-state sales tax (with resale-certificate handling), inbound freight, marketplace final-value fees, a returns reserve, and outbound shipping onto every lot, so a deal is scored on what lands in the bank, not the hammer price
  • Max-bid solver that inverts the cost stack to back out the highest acquisition price preserving a target profit multiple — flagging lots as infeasible when fees and freight would eat the margin even at a $0 hammer
  • Server part-out calculator that estimates component-level resale value (CPUs, RAM, drives) against acquisition cost, catching lots worth more disassembled than whole
  • Auction close monitor that watches the highest-conviction lots, polls live bids on a tightening cadence into the final hours, and alerts the moment a bid crosses the computed max — keeping a deliberately small, high-confidence watchlist rather than chasing everything

Impact

  • Every scored lot carries its full landed-and-resold economics, so bid decisions reflect net margin after premium, tax, freight, and fees rather than a headline discount
  • The max-bid solver turns a margin target into a single actionable number per lot, letting the operator bid with discipline across hundreds of simultaneous auctions
  • Underpriced lots surface daily across IT, electrical, and test-equipment verticals from sources a single-platform broker never sees
  • 70+ test files and over 1,200 passing tests guard the classifiers, staleness gates, and false-positive filters, holding the acquisition signal to an auditable standard
  • Serves as the acquisition counterpart to the published SpotWire priced indices: the same per-vertical pricing discipline that powers the public market signal also governs what the desk will pay