Intersections Inc. · Brand protection · 2007
The other thing the crawler learned to see.
Minimum advertised price is the deal underneath authorized retail: the dealer gets the product and the margin, and agrees to compete on service and shipping rather than on undercutting the list price. Policing that deal across a whole market of resellers was manual and hopeless. At Intersections I designed the MAP Detection System, an automated crawl-and-compare architecture that watched reseller pricing continuously and produced the evidence a brand needed to enforce its own agreements. It's the second of the two patents I filed in 2007, and the sibling of the counterfeit-detection work.
The short version
- Same architecture family as the counterfeit system: crawl the market, normalize what you find, compare it against a rule, and emit evidence somebody can act on.
- The rule here is contractual rather than criminal: advertised price against the MAP floor, per product, per reseller, over time.
- Patent filed 2007 (MAP Crawling System, Intersections Inc.), no longer in force.
01 · The system
Watch the market, compare against the floor, keep the receipts
A brand with an authorized dealer program publishes a MAP: the minimum price a dealer may advertise. Dealers who honor it compete on customer service, shipping speed and expertise. Dealers who quietly undercut it poison the program for everybody, and in 2007 catching them meant an intern with a spreadsheet checking storefronts by hand.
The system I designed automated the whole loop at the architecture level: crawlers that walked reseller sites and marketplaces on a schedule, product matching to resolve the same item across storefronts that named it differently, price extraction with the advertised price captured in context, and a comparison engine that flagged every listing below the floor with a timestamped snapshot as evidence. Violations rolled up per reseller over time, because the difference between a pricing error and a strategy is repetition, and enforcement needed to tell those apart.
It was the second time that year I pointed a crawler at a market and asked it to notice what shouldn't be there. The counterfeit system looked for products that were lies; this one looked for prices that were breaches. Both taught me the same lesson I still apply to agentic systems at Oracle: an automated accusation is only as good as the evidence trail behind it.
An automated accusation is only as good as the evidence trail behind it.