Paid search for an online store is not the same product as paid search for a service business, and buying it as though it were is how most budgets get wasted. A store's spend flows mainly through the product feed rather than through keyword lists, its results are judged on margin rather than on leads, and a large share of its campaigns are now automated in ways that hide the levers a buyer used to inspect. An agency selling ecommerce paid search is really selling three competences: feed hygiene, structure and bidding discipline, and honest measurement. This page sets out what each of those means, what genuinely moves the fee, and how to tell a candidate that understands your margins from one that will happily buy you unprofitable revenue.
The feed is the campaign
Shopping and Performance Max campaigns do not match on keywords you choose; they match on the data in your product feed. Titles, descriptions, product types, brand fields, GTINs, availability and price all determine which searches your products appear for and at what price. That makes feed work the highest leverage activity in the whole program, and it is also the most commonly skipped one, because it is unglamorous and lives in your commerce platform rather than in the ads account. Google's Shopping ads policies also set requirements a feed must satisfy before it serves at all, covering accurate pricing and availability, prohibited products and the match between the landing page and the listing. Ask any candidate agency what it will do to your feed in the first month. If the answer is that the feed is your developer's problem, you are buying bid management on top of a broken foundation, and no amount of bid management fixes a title that does not describe the product.
Structure, automation and what you can still control
Most store spend now runs through automated campaign types. Google's own description of Performance Max is that it serves across search, shopping, display, video and other surfaces from a single campaign using automated bidding and targeting. That buys reach, and it costs visibility: the buyer sees less about where money went and has fewer direct levers. The competent response is not to refuse automation but to structure around it, separating brand traffic from non-brand so you can see what you are actually paying to acquire, splitting high-margin and low-margin ranges so a single blended target does not quietly subsidise your worst products, and holding back exclusions and budget caps as real controls. Quality Score still matters in the search campaigns you do run, and Google documents it as a diagnostic built from expected clickthrough rate, ad relevance and landing page experience. An agency that cannot explain how it separates brand from non-brand is an agency whose reported return includes people who were going to buy from you anyway.
Measurement, margin and the numbers that matter
The most expensive disagreement in ecommerce paid search is about which number counts. A return on ad spend calculated on revenue flatters every campaign, because it ignores cost of goods, shipping, payment fees and returns. The same program measured on contribution margin can look mediocre or even negative. Before you sign anything, decide which measure you will judge the work on and put it in the reporting spec, and get returns and cancellations reflected in it. Attribution is the second argument: platform-reported conversions will always exceed what your own commerce reports show, because the platform counts view-through and cross-device paths generously. The healthy arrangement is that the agency reports platform numbers and your own order data side by side and explains the gap rather than pretending it does not exist. Stores that also invest in organic search find that the paid program gets cheaper over time as branded demand grows, which is why the two budgets are best planned together rather than argued over separately.
What moves the fee, and how to vet a candidate
Agency pricing here comes in three shapes: a flat monthly fee, a share of media spend, or a hybrid with a floor. A share of spend is the most common and the one to read hardest, because it pays the agency more for spending more, which only aligns with you if the contract also names a target efficiency. Catalogue size, number of markets and currencies, feed complexity and whether creative production is included explain most of the spread between quotes. To vet a candidate, ask for accounts it currently manages at your rough scale, ask what it changed in the first ninety days of the most recent one, and ask what it would stop doing in your account rather than what it would add. Insist that the ads accounts, the merchant centre account and the analytics property are owned by you and merely granted to them, because agencies that build inside their own manager accounts can make leaving expensive. Finally, ask how it treats brand search, and listen carefully to whether the answer protects your reported numbers or your actual profit.
Questions people ask about ecommerce paid search
How much should a store spend before hiring an agency?
Below a few thousand dollars a month in media, agency fees usually eat too much of the budget to make sense, and a competent in-house setup plus feed hygiene will do better. Above that, the difference a specialist makes on feed quality and structure generally outweighs the fee.
Is a share of spend fee a bad deal?
Not inherently, but it needs a counterweight. Pair it with a named efficiency target and a cap, or with a flat floor so the agency is not rewarded purely for scaling budget. The problem is never the model itself, it is a model with no term that constrains it.
Should I bid on my own brand name?
Often yes, if competitors bid on it, and the cost is usually low. What matters is that brand spend is reported separately, because folding it into a blended return makes every campaign look better than it is and hides what non-brand acquisition really costs.
How long before a new paid program stabilises?
Automated campaigns need a learning period and enough conversion volume to optimise, so expect several weeks of noisy results before anything is worth judging. Changing targets weekly during that window restarts the process and is the most common self-inflicted wound in a new account.