Account structure for online shops: by profitability, not category tree

The shop menu sorts by range, bids need contribution margin

A campaign is a unit of control: one target, one budget, one message. That is why I consider the category the wrong criterion for splitting. The menu groups what people search for together. What matters for the bid is how much a euro of revenue leaves over, and the two axes cut across each other. The same category holds bestsellers with thin margins and accessories with fat ones.

Google also advises that with automated bidding, a campaign should generally have one objective. Only your own margins show where your range needs a second one.

Four tests for every planned split

In my view, a separate campaign needs a reason from the first three tests and must also pass the fourth.

1. A different target. Calculate the break-even ROAS for both parts: 1 divided by the contribution margin rate after returns. The gap must be larger than the usual monthly fluctuation of your ROAS, otherwise you are steering noise. If your ROAS fluctuates by 15% from month to month, 4.5 and 5.0 are the same target. That is my rule of thumb, not a Google requirement. How to derive the costs is explained under calculating break-even ROAS.
2. Its own budget. One part is meant to get money that the other must not take away, such as a clearance sale or a range you are deliberately building up. If you would never set its budget differently from the neighbouring campaign's, the reason falls away.
3. A different message. Language, offer or landing page must differ. A different heading in the menu does not count.
4. Enough volume. For target ROAS in Shopping, Google names at least 15 conversions per Merchant Center ID in the last 30 days, advises checking whether each bid strategy reaches that volume per month, and expects fewer, larger campaigns to perform better in general. To me, the 15 are the lower limit for running a campaign. I would only make a new split once each part keeps around 50 conversions a month afterwards, and closer to 60 for Performance Max.

Worked example: five menu items become two campaigns

Suppose a shop sells coffee machines. All figures are invented and rounded.

Following the menu, that would be four or five campaigns. Following the numbers, it is two. New machines and portafilter machines are so close together that a separate target steers nothing. Grinders, at 4.0, already differ more visibly, but on their own, with 12 conversions, they would fall below Google's lower limit, so they stay with the machines. Accessories and refurbished machines can live with half the ROAS. If they ran on a shared target with the new machines, that target would choke off revenue that has long been making money. The refurbished bean-to-cup machines show why the menu is the wrong axis: they sit in the same category as the new machines and commercially belong with the accessories.

You implement this through the feed. Custom labels are available up to five per product, and one of Google's own examples is margin, with the values LowMargin and HighMargin. They work in Shopping and in Performance Max with a Merchant Center data source. The second campaign therefore filters on a label, not on a category path. The break-even is the lower limit here, not automatically the target.

One alternative usually makes the margin split unnecessary. If you send contribution margin as the conversion value, the bid already works with the amount you keep, and the margin reason for a second campaign falls away. Budget and message can still argue for a split.

You separate brand and generic because the average is misleading otherwise

People searching for your brand already know you, and mixing them in obscures the performance of generic keywords, according to an e-commerce agency whose assessment I share. Here is the calculation, again hypothetical: brand costs €1,000 at a ROAS of 15, generic costs €4,000 at a ROAS of 2.5. Together that makes €25,000 in revenue on €5,000 in cost, a ROAS of 5. With a break-even of 3.5, the campaign looks healthy while four fifths of the budget runs below break-even.

For the split to hold, brand traffic actually has to end up in the brand campaign. That is what brand exclusions are for, which according to Google block your own brand, common misspellings and related subsidiary brands more comprehensively than negative keywords. For Performance Max, Google explicitly recommends them when brand traffic runs in a separate campaign. There they apply to Search and Shopping. Shopping can optionally keep serving for these terms; for a clean split, leave that option off and check the search terms for leaks. Whether you should bid on your own brand at all is a separate calculation, see bidding on your own brand.

Two limits. Brand here means the name of your shop or your own label. Searches for manufacturers whose products you merely stock are, in my assessment, commercially generic traffic, because the customer is still choosing between retailers. And if your shop gets hardly any brand searches, there is not enough volume for separate control. In that case, it is enough to take brand out of the numbers when you analyse them.

You separate countries when the numbers or the language differ

A country on its own is not a reason: according to Google, automated bidding does not require campaigns to be broken out by geography. The test is the same as for the range, except that with countries something differs more often. The table is my classification as a starting point; your own costs take precedence.

Hypothetically: if 30% contribution margin remains in Germany and 20% in Switzerland after shipping and customs handling, the break-even is 3.3 against 5.0. A shared target would be too low for Switzerland. There is more on this under Google Shopping for Switzerland.

Campaigns spanning several countries have a further drawback, which one agency puts like this: you lose control over how the budget is distributed between countries, and each campaign should run in only one language. For retailers, that becomes expensive as soon as the country with the highest ROAS is not the country with the highest contribution margin per euro of revenue.

How to tell that the structure is too fragmented

According to Google, segmented traffic in Search campaigns forces Smart Bidding to optimise for multiple targets separately, which can reduce efficiency. In the account, you can spot it by these signs:

• Two campaigns have the same target value and the same budget behaviour. Then nothing separates them but the name.
• A campaign does not reach the conversion volume its bid strategy needs.
• You justify a campaign with a report. Splitting purely for reporting belongs in labels.
• There are separate campaigns per device or match type without a target of their own.

A structure can also be too coarse, and that costs money too. If products with 20% and 40% contribution margin run on one target without a margin-based conversion value, the bid optimises for revenue and can push it towards products that leave little over. The campaign gets bigger and the account gets worse. How many campaigns fit in the end is covered in more depth in how many campaigns a shop needs.

When this logic does not hold

Below the conversion volume for a single target ROAS strategy, the margin split is pointless: a small shop gains more when all the data lands in one campaign, and should sort out its range and tracking first. The test also assumes that you know the contribution margins per product group. Estimated margins lead to apparent differences that are not real, so get the numbers first. Time-limited promotions are the third exception: a temporary campaign with its own budget can be right even if it would have no reason to exist permanently. The Google information is as of November 2026.