Google Ads for online shops: the system from a retailer's point of view

You steer the definition of success more than the campaign

If you learn Google Ads through its interface, you see campaign types, ad groups and settings and take them for the controls. For an online shop, that is less and less true. In automated campaigns the system sets the bids, and it optimises for the number you report to it. So I would start with the question of what the system gets to see as success.

The mechanics behind this are in Google's help pages. Value-based bidding optimises for the reported conversion value, conversion-based bidding for the number of conversions. If you report revenue, the system bids on revenue. That holds even if a report in the same account shows gross profit, because a report is not yet a bidding signal.

This leads to four questions that every retailer has to answer for themselves. No campaign setting answers any of them for you, and every later detailed question about structure, budget or feed depends on them.

Which demand: what separates the campaign types for a shop

For a shop, the campaign types come down to one distinction: do you address people who are searching for your product right now, or do you leave it to the system to find further buyers? According to Google, Search campaigns reach people actively searching for your products, with text ads and keywords you choose yourself. Shopping campaigns show product listings in the search results and in the Shopping tab. Performance Max accesses the entire inventory from a single campaign, including YouTube, Display, Search, Discover, Gmail and Maps, and according to Google can be used to complement keyword-based Search campaigns.

My assessment for the typical case: a shop needs a vehicle for product listings, meaning Shopping or Performance Max with Merchant Center data, plus Search campaigns for the queries you definitely want to cover. Separate Display or Video campaigns only come in once you can name a reason the other campaigns do not cover, and a measurement that proves it. App campaigns are irrelevant without an app.

In my view, a new shop with little conversion history is better off starting with Shopping than with Performance Max, because Shopping delivers product-level data first. With stable volume and clean measurement, that argument carries less weight. What Performance Max then delivers and what it hides is covered in the article on Performance Max for online shops.

There is one hard rule for how campaigns interact: Search campaigns with an exact match keyword are prioritised over Performance Max when the query matches that keyword exactly. Without an exact match, according to OMT the higher Ad Rank decides. For me, the exact match keyword is therefore the most reliable lever when a search query should run in a specific campaign with its own target. That is only worth it if this target differs clearly from the rest, otherwise you are just creating admin work.

Which product: not every item deserves advertising money

With Search, you choose keywords. With product listings, you supply inventory: a Shopping campaign needs a Merchant Center account through which you upload your product range. In my view, this shifts the work from the campaign to the product data: I would rather put an hour into the feed than into a fifth ad variant.

The interface does not ask the commercial question behind this: whether every item deserves advertising money. Some items leave nothing after ad spend, others are only available some of the time. I recommend recording this before any debate about structure, most simply in a table listing item or product group, margin and availability. For a very small catalogue with similar margins, this step is largely unnecessary.

Which margin: why baskets of the same size are not worth the same

Once you report revenue, a ROAS target treats baskets of the same size identically. The worked example shows what follows from that. The figures are invented and rounded: suppose a shop sells two products for €100 net each, with a target ROAS of 5, which means €20 of ad spend per sale.

In the reported value the two sales are identical; in the till they are €50 apart. The formula 1 divided by margin is the lower limit: shipping, payment costs and returns come on top, so the actual ROAS target is higher. How to derive it is explained in the article on the right target ROAS.

Two routes follow from this, and both are legitimate:

• Margin groups with their own target. I would not put items with 60% and 10% margin under one ROAS target. That is easy to implement, but it splits the conversion volume across several campaigns.
• A different value in tracking. For value-based bidding, at least two different values have to be reported, and according to Google these can be real economic values such as revenue, or proxy values. If you report gross profit in the example, €60 and €10 appear in the account. With a target of 2 on this new scale, A may then cost €30 on average and B only €5. The technical side and the pitfalls are covered in the article on margin as the conversion value.

Before you build either of these, a quick check helps: sort your product groups once by ROAS and once by gross profit per advertising euro. If the ranking does not change, the switch will gain you little. That is the norm with uniform margins, for example when almost the entire range sits in a narrow margin band.

Simply raising the target does not solve the problem. According to Google, a higher ROAS target can limit the number of auctions and reduce total conversion value. You may then get a nicer ratio on less business, without the mix of A and B necessarily improving.

Which evidence: the data chain from basket to bid

This is where the real decision is made, because every earlier answer depends on this chain. It has four links, and each can break silently without any campaign stopping because of it. The first two feed the bid, the last two your item-level reports.

1. The purchase is counted. The total number of conversions and their value have to match the shop system. How to compare them is shown in the article on checked conversion data.
2. The value is genuine. A fixed value per order, or a value with the wrong VAT logic, feeds the system a false ranking. Value-based bidding needs the at least two different values mentioned above.
3. The items match. Conversions with cart data only work if the item IDs in the conversion tag exactly match the IDs in the Merchant Center feed. The diagnostics area in Google Ads lists the reported IDs so you can compare them with the feed.
4. The cost of goods is in the feed. Together with the cart data, the cost_of_goods_sold attribute provides additional gross profit metrics.

The fourth link is where an expensive misconception sits. According to the Merchant Center Help, the cost of goods is used for reporting, and a rough estimate is explicitly enough for that. The way I read it: it shows you how much profit is generated on paper, but in the standard set-up it does not change what is being bid on.

What stays with you and what Google does

The boundary is easiest to draw using Performance Max, because that is where the most is automated. The breakdown follows the Google Help on Performance Max, as of November 2026.

The left-hand column is shorter, but it is the commercial one. If you enter revenue, an unrealistic target and unchecked conversions there, you get a very efficient execution of the wrong task on the right.

In an existing account, you start at the back

The interface suggests starting with campaigns and structure. I would check the four questions in reverse, because each answer builds on the previous one. At every step the same applies: only once it holds is the next one worth doing.

1. Evidence. Which conversion action feeds the bid, which value is reported, does the total match the shop, do the item IDs match? If that does not hold, this is where repairs happen first, and the campaign stays as it is.
2. Margin. Does the current ROAS target support the margin of the items that actually generate the revenue? Calculate break-even per product group and run the ranking test from above.
3. Product. Is budget flowing into items that lose money after ad spend or are out of stock? This gives you a product selection you can justify.
4. Demand and structure. Only now is it worth asking which campaign is responsible for which search queries, and whether campaigns should be merged or split. The account structure then follows from the answers in the previous steps.

This order is my assessment and does not apply everywhere. With very few conversions, data volume is the tighter limit: for target ROAS in Search and Shopping campaigns, Google names at least 15 conversions in the last 30 days at the conversion tracking level. I regard that as the lower limit, not as a figure from which target ROAS runs stably. Below it, there is little point in splitting margin into groups, because each group would have even less signal. With uniform margins, step 2 largely falls away and the product range moves up the list. If the account is blocked right now, for example by a suspension in Merchant Center, that comes first.

Only a test measures incremental impact

The four questions make sure the system is given the right task. Whether a reported sale would not have happened without the ad is something even the best conversion value cannot tell you, because it measures attribution, not incremental impact. That needs a test of its own. Nor does the chain replace an overall calculation for the shop: if revenue in the account rises and the contribution margin in the business does not, the figure from the business is the one that counts.

Open the conversion action that feeds the bid and write down what arrives as the value. If that is not the figure you run your business by, you know where the work begins.