tROAS and tCPA Tuned Against Margin, Not Platform Revenue

A target ROAS set against platform-reported revenue is a target set against a number your finance team does not recognise. Take an apparel catalogue running 42 percent gross margin at full price and 8 percent on markdown: a single tROAS of 400 percent has told the bidder those two orders are worth the same, and it will buy the cheap one all day, correctly, because that is what it was asked for. The first half of this engagement replaces the value the account sends with something closer to contribution. Only then does anyone touch a target. Where the value comes from: Margin bands mapped to product groups and carried through conversion value rules or the feed. Refunds and chargebacks netted off on a lag you choose. Offline orders imported back against their GCLID inside the import window, so the phone and the shop floor stop being invisible to the auction. For lead businesses the same logic runs on stage weights instead of revenue: an enquiry worth one, a qualified opportunity six, a signed contract its actual value, uploaded on a schedule matching your sales cycle rather than the reporting calendar. Enhanced conversions for leads closes the identity gap where consent allows. None of this is interesting and all of it decides whether the rest of the engagement means anything. Choosing a strategy, and paying for the change: Every strategy change restarts learning, and learning is paid for in inefficiency, so I change one thing and wait — roughly a week, or two conversion cycles if yours is longer than a week. Campaign-level targets where conversion volume supports them, portfolio strategies where it does not and consolidation is the only honest fix. Seasonality adjustments are for short known events, a two-day sale or a holiday weekend, and are not a remedy for a soft month. Data exclusions are for tag outages and nothing else; I have watched accounts excluded into fiction one bad week at a time. Experiments that end in a decision: Before a drafts-and-experiments split goes live I work out what effect the campaign's conversion count can actually resolve over the test window. If the answer is that only a 40 percent swing would be detectable, I say so, and we either run it longer, run it on a larger campaign, or do not run it. Tests that cannot conclude are the most expensive habit in this discipline, because they end with somebody's preference wearing a confidence interval. What I refuse: Nudging targets weekly. Third-party bid management layered on top of Smart Bidding, which is two systems arguing about your money. Reporting success in platform ROAS to a board that reads a P&L. Setting a target before I can see how value reaches the account — if the measurement work is out of scope, so am I. Not for you if: Your account cannot produce enough conversions for the bidder to learn at all. The answer there is consolidation and demand creation, not tuning, and you will get that in the audit rather than a project. Your margin data is unavailable, or available in a form nobody will stand behind in a meeting. You want targets moved this week because last week was disappointing. What you end with: A written value model your finance team has signed off, the import mechanism running on a schedule, one strategy change made and held, and a short list of which numbers to watch and which to ignore for the next quarter. The last of those matters more than it sounds, and it is the part that survives after I leave.