Wasted Spend Cleanup for Shopping and Performance Max

Most cleanup work produces a negative keyword list, and within a fortnight broad match and Performance Max have regenerated the traffic it was meant to stop. The largest cut we made last year was not a keyword at all: 31 percent of one retailer's Shopping spend sat on eleven item IDs that sold reliably and lost money once a 24 percent return rate and zone-based shipping were counted. Nothing in a search terms report shows you that. Where the money goes: Three places, usually in this order. Products that convert and do not pay. Inventory and price drift, meaning items still being bid on while out of stock at the warehouse serving that province, or priced above the offer they are displayed beside. And demand you already own, where brand terms, the retargeting tail and conquest campaigns bill you for arrivals that had another route in. The margin join: We take ninety days of orders with refunds, returns, landed shipping cost by destination zone and cost of goods, compute contribution margin per item ID, and write it back into the feed as a custom label. From there bidding can tell a sale from a good sale, and the cut list stops being an argument about what ROAS threshold is acceptable. This is the step most shops skip, because it needs your finance data and a day of reconciliation rather than a platform export. Brand gets tested, not asserted: Splitting the country geographically, holding brand campaigns dark on one side for two to three weeks, and reading total orders rather than platform-attributed ones is the only way to know what that line is worth. Sometimes brand is cheap insurance against a competitor bidding on your name and the answer is to keep paying. Sometimes it is defending traffic that arrives regardless. We will not run this in Q4, we will not run it while someone is actively conquesting you in the same window, and if your order volume is too thin for the split to produce a readable difference we say so rather than sell you a test that returns a number and no information. The cuts that are not arguments: Search Partners on accounts where it has never reached target. Display expansion on Search campaigns. Performance Max running without brand exclusions or account-level negatives, quietly rebuying traffic you already pay for elsewhere. Bid adjustments left on demographic segments where they add variance and nothing else. These arrive as a list with the numbers attached in week one. Guardrails, because waste regrows: The deliverable is not a one-time list. It is the custom label structure and its refresh schedule, the account-level negative list with a named owner, a weekly n-gram check that takes twenty minutes instead of a day, and feed rules that pull an item out of bidding when availability or price breaks. In your account, documented, handed to your team. What we will not sell: A saving percentage agreed before we look. Any number promised in advance is a number we would then have to produce, and the way it gets produced is by cutting spend that was working and treating the lost revenue as somebody else's line item. We also do not work on a share of savings, because that incentive points at a smaller account rather than a better one. Who this is not for: Accounts under 40,000 USD a month in search and shopping, where the work costs more than it returns. Businesses that cannot or will not share cost of goods and return rates, since without them we would be running exactly the search terms exercise we just criticised. And lead generation accounts with no closed-won value in the CRM, where the equivalent work is real but starts in your CRM rather than your ad account.