Google Shopping ads: the complete operator's guide.

Google Shopping is a feed problem and then a margin problem, in that order, and almost every account that has plateaued got there by skipping to the third thing. There are no keywords to choose and no headlines to write, so teams conclude the only lever is the bid — which is the one lever that cannot fix either of the first two problems. Everything you would normally control through keywords and copy, Shopping controls through the product feed. Everything you would normally control through campaign targets, a mixed-margin catalog quietly breaks.

Before anything: how much of your catalog can even compete

The first number to pull is not ROAS. It is the share of your revenue-carrying catalog that is actually eligible to serve — approved in Merchant Center, in stock, correctly priced against the site, and not excluded by a listing group nobody remembers creating. This is the cheapest possible win and the most frequently skipped, because it looks like data hygiene rather than strategy.

It is not hygiene. A disapproved product is invisible at any bid, so every ineligible SKU is a bid strategy you are paying to run against a smaller catalog than you think you have. If a meaningful share of your top-selling products is ineligible, nothing further down this page will move the account as much as fixing that will. Check it before you touch a target.

The feed is the targeting layer

When someone searches, Google scans eligible product feeds and decides which products are relevant enough to enter the auction. Relevance comes overwhelmingly from your product data: the title first, then description, product type, category, and attributes like brand, color, and size. Your product title is your keyword list. A title that reads like your warehouse system — "SKU-2841 Oak Tbl 180" — matches almost nothing. A title that reads like a search — "Solid Oak Dining Table 180cm, Seats 8" — matches everything it should.

This is why we treat feed optimization as its own discipline: it is not data hygiene, it is targeting, and the highest-leverage hours on most Shopping accounts are spent there rather than in campaign settings. The mechanics of title construction, attribute coverage and supplemental feeds live in that piece. What matters here is the sequencing consequence — a bid change against a bad feed is an optimization applied to the wrong problem.

What one blended target ROAS actually costs you

Now the margin problem, which is where most of the money is. Smart Bidding optimizes toward the values you report, so if you report revenue it will chase revenue — including revenue you lose money on. On a single-margin catalog that is harmless. On a mixed catalog it is two opposite errors running simultaneously, and the arithmetic shows exactly how much they cost.

Breakeven ROAS is simply the inverse of contribution margin: at 50% margin you break even at 200%, at 20% margin you break even at 500%. Take a catalog with three bands and a single 400% account target:

Band  Contribution  Breakeven  Under one 400% target,
      margin        ROAS       per $100 of revenue:

A     55%           182%       spend $25, contribution $55  -> +$30
B     35%           286%       spend $25, contribution $35  -> +$10
C     18%           556%       spend $25, contribution $18  ->  -$7

Band A is being held at 2.2x its breakeven.
Band C loses money on every sale the target permits.

Read Band C first: at a 400% target the algorithm is rewarded for buying revenue that destroys $7 of contribution per $100 sold, and it will buy a great deal of it, because low-priced low-margin items are usually the cheapest things in the catalog to convert. The target is not merely too loose for that band; it is actively pointing the bidder at your worst inventory.

Then read Band A: it makes $30 per $100 of revenue at a 400% target and would still make roughly $16 at a 260% target — while buying substantially more volume, because a looser target wins auctions a tighter one sits out. The blended target is leaving that growth unbought. So the single number is simultaneously funding your losses and capping your profits, which is why segmenting by margin is not a refinement but a correction.

The practical version is to derive a target per band as a consistent multiple of that band's breakeven — roughly 1.4x is a reasonable starting cushion — rather than picking numbers that feel right. Going further and reporting margin-adjusted conversion values so the bidder optimizes contribution directly is better still, and we cover that mechanic in the profit-on-ad-spend piece. On Shopping it matters more than anywhere else, because catalogs mix margins more widely than any other channel's inventory does.

Structure: priority and segmentation

Standard Shopping gives you two structural tools worth mastering. The first is product group segmentation — splitting bids by category, brand, margin band, or custom label so a bestseller and a slow-mover are not priced identically. Custom labels are the underrated one: tag products by margin tier or seasonality in the feed, then structure campaigns around those labels. Done properly, your feed becomes the strategy document and the campaign structure is just its reflection.

The second is campaign priority combined with negative keywords — the classic query-sculpting play. A high-priority campaign with low bids and brand negatives catches generic queries cheaply; a low-priority campaign with high bids catches the brand and high-intent queries the first campaign passed on. It takes discipline to maintain, but it converts one blunt channel into a tiered one where bid matches intent.

Four thresholds that separate a leak from noise

Diagnosis needs numbers you can fail, otherwise every review becomes a matter of taste. These are the four we check first, with the floors that keep them honest.

• →Spend concentration. Shopping spend is naturally Pareto-shaped, so a small share of SKUs carrying most of the spend is normal and not a finding. What is a finding: a single product group carrying the overwhelming majority of spend, which means the segmentation is decorative — you have the appearance of structure with the economics of one blended bid.
• →The unsegmented remainder. Any "everything else" group carrying meaningful spend is a blind spot by construction, because it contains both your undiscovered winners and your quiet losers at the same price. The correct amount of spend in an "everything else" group is close to none.
• →Zero-conversion spend share. Sum the spend on products with no conversions in a 90-day window. Some of that is legitimate — long consideration cycles, expensive items — but once it becomes a large fraction of the total you are funding a catalog-wide experiment nobody designed. Fix it by exclusion or by feed work, not by lowering the account target.
• →Individual product waste. A product that has spent well past your target cost per acquisition with nothing to show for it is a candidate for exclusion, but only above a significance floor — roughly 50 clicks, or a clear multiple of target CPA in spend. Below that, the honest label is worth watching rather than confirmed, and acting on it is how operators talk themselves into pruning products that were merely unlucky.

The floor matters more than the threshold. Any one of these checks run on a thin window will generate an impressive-looking list of problems that are actually variance, and a Shopping account has thousands of SKUs to generate false positives from. When in doubt, widen the window before acting.

The weekly operating rhythm

• →Search terms review: add negatives for irrelevant queries, and mine converting queries for title improvements — the loop that compounds, because today's winning query becomes tomorrow's title.
• →Feed diagnostics: disapprovals, missing identifiers, price and availability mismatches between feed and site. Each one silently shrinks the catalog you are bidding on.
• →Price competitiveness: Shopping is comparison shopping. If you are consistently priced above rivals on identical items, no bid fixes it and no feed work hides it.
• →Margin-weighted performance: revenue by product group is vanity, contribution by product group is sanity. If you can only build one report, build this one.

The honest limits

• →Every threshold above is a starting point calibrated on mid-market catalogs, not a law. A 40-SKU brand and a 40,000-SKU marketplace have completely different natural concentration and zero-conversion profiles, and applying the same floors to both will mislead you in opposite directions.
• →Margin-band targeting requires accurate per-product margins including shipping, payment fees and returns. Segmenting on wrong margin data is worse than not segmenting, because it hard-codes the error into the bidding.
• →Returns are the missing term in most Shopping margin math. A band with a 40% return rate has a very different real contribution than its product margin suggests, and the categories with the best-looking ROAS are frequently the worst offenders.
• →The arithmetic here assumes contribution margin is stable across the price points inside a band. Where it is not, the band is drawn wrong, and no target will be right for it.

Shopping rewards operators who work it in order — eligibility, feed, margin, structure, then bids — and punishes set-and-forget harder than any other channel, because the catalog underneath it never stops changing. If you want the comparison with the automated alternative, we wrote up Shopping vs Performance Max separately, and our Google Shopping service page covers how we run the channel end to end.