Traffic Arbitrage Desk Built to a Margin Floor

I will not open a traffic buy that has not shown 25 percent contribution margin at zone level, after refunds, chargebacks and the network's scrub, across at least two full weekly cycles. Most arbitrage decks fail on the second point rather than the first: a source clearing 18 percent on day three and 4 percent by week three was a bidding artefact, and the spend that followed it was tuition. This engagement stands up a buying operation whose margin you can see per placement, and it ends with your team holding the controls. Where the traffic comes from: Push and in-page push, in-app and native. RichAds, PropellerAds, ExoClick and Zeropark on the self-serve side; Taboola and Outbrain where the offer can carry a content step; Mintegral and Unity when the destination is an app install rather than a web checkout. Search arbitrage and Meta accounts are outside what I run — the account-ban economics of that game reward people who buy assets in bulk, and I am not that operator. The first eight weeks: Week one is plumbing, and it is not optional: click IDs passing to the offer, postbacks returning, cost pulled by API so margin is calculated rather than typed. Weeks two and three run a deliberately unprofitable spread across sources and creative angles, wide enough to have something to cut. From week four the work is subtraction — zone and placement blacklists, day-part and device rules, frequency caps — and the number I report is margin per thousand impressions by source, not ROI on the account. Week six is where most sources die. By week eight you have two or three that survive their own bid inflation, written down as rules a buyer can follow on Monday morning without me. What you also get: A float model. Arbitrage dies of working capital more often than of creative: networks pay net-30 or net-60, sources want prepayment, and a business scaling on 30 percent margin against a 45-day cash gap runs out of money while its dashboard says it is winning. I will build that table with your finance lead and tell you the ceiling your balance sheet actually supports. What I refuse: Cloaking of any kind, including the polite versions with a safe page for review. Nutra, financial signal offers and anything with a rebill the user did not knowingly agree to. Trademark bidding on a brand that is not yours. Creative that borrows a broadcaster's or newspaper's identity. These are not squeamishness; each one converts a margin business into a business whose main asset is an account that can be closed by someone else's compliance team on a Tuesday. Who this is wrong for: An operator wanting the source and offer combinations that are working right now. Those have a shelf life measured in weeks, they are the only thing anyone in this trade holds back, and a person selling them to you is selling them to everyone. Also wrong for a team without a buyer of their own to hand this to — I will run the desk during the engagement, but a buying operation with no owner reverts to the network's default targeting within a month. And wrong for anyone whose budget for media is the same budget as the fee; the media is separate, paid directly by you, and the smallest amount that produces a readable answer is roughly fifteen thousand euro over the eight weeks.