Google Ads for E-commerce: The Account Structure That Works

Most e-commerce Google Ads accounts do not underperform because of bad creative or the wrong bid strategy. They underperform because the structure hides what is actually working, so every optimisation decision is made on blended numbers. This guide covers the account structure that holds up in 2026: how to separate brand from non-brand, when Performance Max beats Standard Shopping, and how to feed the system clean value data before you touch a single target.

Separate brand and non-brand before optimising anything else

Branded search converts because the demand already exists. When it shares a campaign with cold prospecting, blended ROAS looks comfortable while the acquisition side quietly loses money — and the bidding algorithm learns from a mixed signal it cannot untangle.

The fix is structural, not tactical. Isolate brand demand so you can price it separately and read incrementality honestly. In practice that means a dedicated brand Search campaign, brand exclusions applied to Performance Max, and brand terms added as account-level negatives on every prospecting campaign.

• Run one brand Search campaign with exact and phrase match on your name, misspellings, and brand-plus-product combinations.
• Apply brand exclusion lists to Performance Max so it cannot absorb the cheapest conversions in the account.
• Add brand terms as account-level negative keywords, which also apply to PMax and Demand Gen.
• Report brand and non-brand ROAS as separate lines, never as one blended number.
• Review brand share of total conversions monthly — a rising share usually means prospecting has quietly stalled.

Performance Max vs Standard Shopping: choose by control, not by default

Performance Max is the default for a reason: it covers Shopping, Search, Display, YouTube and Discover inventory from one asset group and generally finds volume that a manual structure will not. But it trades query and placement control for reach, and in catalogues with wide margin variance that trade is not always worth it.

A workable pattern for most retailers is PMax as the core volume engine, with Standard Shopping retained for product sets that need explicit control — high-margin hero products, clearance lines, or ranges where you must see the search terms.

Split asset groups by margin or category rather than by whim. One asset group per meaningful product cluster, each with its own listing group, gives you a structure you can actually read and act on.

• Group asset groups by margin tier or category, not by supplier or SKU count.
• Use search themes as directional input, not as a substitute for a keyword strategy.
• Keep at least one campaign with full search term visibility so you retain a query-level view of demand.
• Do not run PMax and Standard Shopping on the same product set without exclusions unless you are deliberately testing priority behaviour.

Feed quality is the real targeting layer in e-commerce

In Shopping and PMax, the product feed does the work keywords do in Search. Titles, attributes and product types determine which queries you are eligible for, so feed work is targeting work — not a Merchant Center chore to hand off and forget.

Front-load the specific, high-intent terms in the title and keep attributes complete. Disapprovals and 'limited performance' warnings in Merchant Center are silent budget leaks: an excluded product simply stops competing, and nothing in the Ads interface tells you why volume dropped.

• Structure titles as brand + product type + key attribute (size, colour, material, compatibility), most specific first.
• Fill GTIN, MPN, brand, condition, availability and product_type on every item — partial feeds get partial reach.
• Use custom labels for margin band, stock level, seasonality and bestseller status so campaigns can be split on business logic.
• Audit Merchant Center diagnostics weekly and track the ratio of active to submitted products as a standing metric.
• Keep price and availability synced in near real time; mismatches cause both disapprovals and wasted clicks.

Send back conversion value, not just conversion counts

Value-based bidding cannot beat a target it cannot measure. If every purchase reports the same fixed value, Target ROAS is effectively optimising for volume — and it will happily buy your lowest-margin SKUs.

Send dynamic revenue with every purchase event, and consider passing profit or contribution margin instead of gross revenue where your margin spread is wide. Then close the measurement gaps: enhanced conversions for web, consent-mode-aware tagging, and server-side collection where you have the engineering capacity. The same discipline applies across channels — a Conversions API or equivalent server-side feed on your social platforms keeps the whole media mix comparable.

One conversion action as primary, everything else as secondary. Multiple primary actions double-count and distort the target.

• Pass transaction-level revenue dynamically; never hard-code a single average order value.
• Deduplicate with a transaction ID so client and server events do not both count.
• Use conversion value rules for genuine differences in worth, such as new versus returning customers or high-value regions.
• Mark micro-conversions (add to cart, newsletter signup) as secondary so they inform reporting but not bidding.

Ramp Target ROAS gradually and respect the learning period

Target ROAS behaves like a constraint, not a lever. Set it far above what the account currently achieves and delivery collapses; drop it too low and spend runs ahead of profitability. Start from your trailing 30-day actual ROAS rather than your desired one.

Move in steps of roughly 10-15% and hold for a full conversion cycle — for most retailers that is one to two weeks — before judging the result. Data thins out fast when you fragment campaigns, so consolidate rather than split when volume is marginal.

Use seasonality adjustments for short, predictable spikes such as a two-day flash sale. Do not use them for a general 'things feel slow' hunch; the bidding system already handles ordinary variance.

• Set the initial target at or slightly below your trailing 30-day realised ROAS.
• Change targets no more than once per one to two weeks, and change one variable at a time.
• Treat significant target or budget changes as a learning reset and avoid stacking edits during it.
• Reserve seasonality adjustments for known events of one to seven days with an expected conversion rate shift.

Negative keyword hygiene and a review cadence that survives scale

Automation widened the query surface; it did not remove your responsibility for it. Account-level negative lists now apply across Search, Shopping and Performance Max, making them the single most efficient place to enforce exclusions across a whole account.

Build shared lists once and apply them everywhere: a brand list, a competitor list, an irrelevance list (jobs, reviews, free, DIY, second-hand), and a category list for products you do not stock. Then make review a calendar item rather than a reaction to a bad week. Agencies running many client accounts usually script or centralise this step — WoAgency exists partly because that maintenance work does not scale by hand.

One caveat: negatives are for irrelevant queries, not for expensive ones. Excluding high-CPC terms that still convert profitably is how accounts slowly shrink.

• Maintain shared lists for brand, competitor, irrelevance and out-of-stock categories, applied at account level.
• Review search terms weekly on Search and at least monthly on PMax search categories.
• Keep close-variant matches in view — phrase and broad match still surface intent you did not ask for.
• Document why each negative was added, so a later reviewer does not undo a deliberate exclusion.

Questions, answered

More guides