Custom Bidding and Flight Pacing for Live Display

A display flight in Israel does not pace on a calendar week. Every DSP spreads even-paced budget across seven equal days, and for a manufacturer selling to procurement, or a brokerage selling to buyers who tour apartments on a Friday morning, that pushes a meaningful share of the week's money into hours where either the audience is offline or the auction is thin enough that you are the only bidder and still pay the floor. Rebuilding pacing around a Sunday-to-Thursday week, with separate Friday-morning and Saturday-evening curves, is usually the first change we make and the one with the shortest payback. Bidding: We write custom bidding rather than selecting a packaged goal. In DV360 that means a custom bidding script scoring Floodlight activity, weighting the events that mean something and refusing to score the ones that do not, instead of a target cost per acquisition aimed at a conversion the model can reach cheaply and pointlessly. In The Trade Desk it means a custom goal held to the same discipline. Where a campaign has too few conversions a week to train anything, we say so, bid on a proxy we can defend, and re-examine it monthly rather than describing a model as learning when it is not. Auction mechanics we work on: First-price clearing behaviour by exchange and the shading each applies before your bid arrives. Floors that move once you bid consistently on the same sections. Multipliers by domain, format, position, device and hour, applied where the log data supports them and nowhere else. We read win rate alongside clearing price per supply source, because a win rate climbing at the same speed as the price you pay for it means somebody has read your bidding pattern and is charging for it. Pacing across deal types: Guaranteed, preferred, private and open-auction supply compete inside your own account unless they are separated. We split budgets so a guaranteed commitment cannot be starved by a cheap open-auction source, set flight pacing rather than daily where flights are short, and read under-delivery risk from the middle of a flight instead of the end. On short flights the arithmetic is unforgiving: an even-paced two-week flight that loses three days to creative approval cannot recover without switching to a pacing mode that ruins the price it recovers at. What we refuse: Bidding to click-through rate or to viewability as the objective. Automated bidding left unexamined, since every algorithmic change we make is checked against log-level data, and where log-level access is unavailable we tell you plainly that part of the work is blind. Retargeting pools inflated with people already returning, which flatter a bid model and teach it nothing. And we do not charge a percentage of media, so nothing in this engagement improves by spending more of yours. Who this is not for: Campaigns running under a few hundred thousand impressions a week, where there is not enough data to bid against and the honest answer is a flat bid and a better site list. Teams who want the account handed back after a month of changes with no record of what changed and why, since the change log is a deliverable here rather than an afterthought. What you receive: A bidding and pacing configuration inside your own seat, a written record of every multiplier and script with the evidence that justified it, a weekly delivery-risk read on live flights, and a hand-back document your team can operate from. Nothing built here depends on us keeping access to it.