tROAS and tCPA Tuning With a Change Log You Can Audit

Most accounts I open are not badly bid. They are constrained, and nobody has said by what. So the first thing delivered is a one-page diagnosis splitting your campaigns into budget-constrained and target-constrained, read from search lost impression share due to budget against lost impression share due to rank. The fixes point in opposite directions. Raising a target on a budget-constrained campaign buys nothing except a fortnight of relearning, and it is the change I undo most often. Where the target comes from: A target return on ad spend is only as honest as the value you send. For a manufacturer selling through distributors, revenue is the wrong number. Gross margin per line, or a modelled value per qualified enquiry built from your own close rate and average order value, is what the bidder should be maximising. I will ask finance for that number and I will not invent it. If nobody in the business can say what a converted enquiry is worth, we run a target cost per action against a qualified-lead event instead of pretending to a value target, and that decision goes in writing rather than a 400 percent target being set quietly because it sounds ambitious. Conversion lag sets the reading schedule: Before a target moves I pull the days-to-conversion distribution. If seventy percent of conversions land inside three days, a change is readable in a week. If your industrial enquiries convert at a median of nineteen days, anything read at day seven is noise and I will not let a target be judged on it. Every change carries the date it becomes readable, written on the same line as the change itself. One variable at a time: Targets move in steps of roughly twenty percent. Targets and budgets never move in the same week. Structural changes wait until bidding is stable. Anything larger than a fifteen percent target shift on a campaign carrying real spend goes through a campaign experiment rather than straight into the live campaign, so there is a control to read it against instead of a before-and-after that a competitor's budget change could explain. Two tools that get confused: Seasonality adjustments tell the bidder to expect a conversion rate change across a short, known event, a two-day sale or a trade show week, and they are the wrong instrument for anything longer than about a week. Data exclusions tell it to ignore a past period because your tag was broken or your checkout was down. Papering over a tracking outage with a seasonality adjustment is a common and expensive mistake, and undoing one is usually week-one work. Reading it outside the platform: Platform-attributed return on ad spend is a modelled number produced by the party being paid. Where budget allows, a geo holdout or a paired-market test sits underneath the largest bid change of the quarter, so at least once a quarter you hold an incrementality read that Google did not generate. Where budget does not allow it, the number is called directional and left that way rather than dressed up. Not included: Ad copy and landing page work. Tag and server-side implementation, where I specify the events, values and identifiers and work with whoever owns your container or backend, but do not build it. Performance Max asset group management. Shopping feed work. Attribution software procurement. Not for you if: Your campaigns run under about thirty conversions a month each, where the bidder has nothing to learn from and structure rather than tuning is your actual problem. You need a target moved to make a month-end number. A Google representative is also making changes in the account, because I will not keep a change log against a party whose incentive is your spend. Or you want a weekly call where targets are discussed and nothing has been left alone long enough to have produced an answer.