Product Segmentation in Google Ads: Splits That Move Budget

Segment products by what should change their budget. Start with how each product behaved in ads over a full year: steady sellers, the profitable long tail, products that spend without selling. Then split by assortment groups whose economics differ, such as product type or price band. A segment moves money only when it gets its own budget or target ROAS, which in Performance Max means its own campaign. You know it works when, after the change, each segment’s share of spend moves closer to its share of revenue.

How product segmentation works in Shopping and Performance Max

Product segmentation means splitting the catalogue into groups that Google Ads treats differently. In both Standard Shopping and Performance Max, you set budgets for campaigns, not for single products, so every split works through product attributes in your feed. The Performance Max guide for online stores shows how segmentation fits into the rest of the set-up, from conversion goals to asset groups.

In Standard Shopping, you split products into product groups inside ad groups. In Performance Max, you split them into listing groups inside each asset group. Both use largely the same attributes from Merchant Center: item ID, brand, Google’s product category, product type, condition, channel (online or local) and five fields of your own (Manage a Shopping campaign with product groups). In Performance Max, subdividing by Google’s product category works only when you target the US, the UK, Australia, Germany, France, Japan, Italy, the Netherlands, Brazil, Norway, Sweden or Turkey (Manage a Performance Max campaign with listing groups). A store selling in Czechia, Poland or Ukraine builds its groups on product type, brand, item ID or custom labels.

Those five fields are custom labels, numbered 0 to 4, and you decide what each one means. Google’s own examples are season, selling rate, clearance, margin, release year and price range. Shoppers never see the values (Custom label 0–4). For a step-by-step set-up, see custom labels in Merchant Center.

Where a segment changes money and where it only changes the report

The consequence for Performance Max is easy to miss. A label used only to split listing groups or asset groups inside one campaign gives you a separate report and, with asset groups, separate creative. The products still share one budget and one target. Google draws the same line in its Retailer best practices for AI-powered Performance Max campaigns. It advises consolidating where you can and running separate campaigns when products need a different budget or target ROAS, for example new products, top-selling products or seasonal moments.

Standard Shopping has one more lever. When the same product sits in several Standard Shopping campaigns, campaign priority decides which one bids. The higher-priority campaign’s bid wins even if a lower-priority campaign bids more. Once its budget runs out, the campaign next in priority bids (Use campaign priority for Standard Shopping campaigns).

A tenth of the catalogue takes 68.1% of spend in the median store

Performance Max runs in 141 of 146 online stores in our data and takes a median 95.7% of their ad budget (GetProfit data, 146 stores, June 2025 – June 2026). For most stores, then, segmentation is a question of how you cut Performance Max. Standard Shopping is the second tool.

Inside those campaigns, the budget spreads unevenly. In our study of 1.4 million products, across 130+ stores over 13 months:

• 55.0% of products received no ad budget at all;
• 37.1% received budget but never sold, and 7.9% sold;
• in the median store, the top 10% of products took 68.1% of spend, and the top 1% took 27.7%;
• the median store spent 40.5% of its budget on products that never sold in 13 months. In the top quarter of stores by ROAS, that share was 39.8%; in the bottom quarter, 71.8%.

These are observations, not an experiment. They show where money goes, not why. Still, they make clear what segmentation is for: deciding which groups of products compete for which part of the budget. For why more than half of a catalogue gets no budget at all, see why most products get no impressions.

What to segment products by

Each common scheme answers a different question about a product. Only some of those answers should change where money goes.

Google’s retailer best practices suggest labelling high-priority products as “bestseller”, “trending” or “holiday product” and moving them into separate campaigns or asset groups. For the holidays, its example structure has three campaigns: holiday merchandise, high-margin products and everything else.

We compare ten schemes in custom label ideas and show which of them change where budget goes. Two of them need their own rules: price-band labels have to survive price changes, and seasonal labels have to follow demand. If you already sort products into A, B and C by revenue, see how ABC analysis compares with performance labels for ads.

Our approach uses two axes: how a product behaves in ads first, its assortment group second.

Why a full year beats a 30-day window for performance labels

A common approach ranks products by revenue over the last 30 days and refreshes the labels every day. A feed-tool vendor’s guide, for example, splits products into Top 10, Top 25, Top 50 and Top 100 on that basis (How to Segment Your Products by Sales Data Automatically). Our study of 1.4 million products shows why a month is too short.

Top sellers change fast. Of the products that reached winner status, 64.7% were one-offs: one conversion in one month. Of bestsellers, 71% stayed in the top for 3 months or less out of 13. A stable winner, with two or more months of sales, held that status for a median of 3 months. A 30-day window gives a one-off sale the budget of a bestseller. And a structure built once around this quarter’s bestsellers describes a different catalogue three months later.

The bottom of the list is harder still. On 514,602 products with spend, we tested whether a spend threshold can tell when to switch off a product with no conversions. The share of correct “switch off” decisions fell as spend grew: from 88.1% for products with $5+ spent to 66.2% at $500+. In the study’s own classification, 35.4% of “losers” went on to convert, after a median of 2 months, and 10.6% of “zombies” recovered. A single rule like “no sale after N spent, switch it off” cuts off part of your future sales.

A product also needs traffic before you can judge it at all. The study puts the minimum honest test at 15–20 clicks, roughly $3–4 at its average cost per click of $0.20. Below that, a verdict rests on noise. A product that has spent too little for anyone to judge it belongs in the grey zone, not among the products to switch off.

We weigh the choice of window in 30 days or 12 months and set out where to draw each line in ROAS and spend thresholds for labels. To mark weak products without excluding the ones that would still sell, see performance-based custom labels.

The eight labels the portal assigns from 12 months of data

GetProfit’s portal puts every product into exactly one of eight labels, based on how it behaved in ads over the last 12 months. It is our own method, not a Google standard. A label describes the product’s role, not its quality: the same product can get different labels in different seasons.

Dormant is the portal’s name for what is often called zombie products: money goes out and no sale comes back. To remove differs from Dormant in the size of the damage. Taken together, the labels separate products that pay off from spend with nothing back: products that take noticeable money and never pay it back.

The 12-month window deals with the traps of a short one. A Bestseller needs sales in at least 6 months out of 12, so a one-off sale can’t make one. A product with too little history lands in Newcomers, not in Loss-making. And a product that spends without selling, but hasn’t yet spent enough to write off, goes to Dormant rather than To remove.

One limit to keep in mind: the labels judge products by ROAS, which counts revenue, not profit. The portal has no data on product costs, so margin stays your call.

In the portal’s Products and labels section, every product gets its label, and the portal recomputes the labels with every data refresh. Any group exports to CSV, and changes export in Google Ads Editor format, so you decide what to apply. A separate article compares this method with the popular heroes, sidekicks, villains and zombies split. We described an earlier form of the method in our 2024 segmentation story: five groups by sales volume, with budget moved between them daily.

The second axis: assortment groups with different economics

A performance label alone mixes products that shouldn’t share a target. A bestselling dinner plate and a bestselling wine glass can both earn their place and still need different targets, because they differ in margin, order value or season. So in the label-based structures we build, the second axis is the assortment group: products with shared economics.

The assortment group usually comes from the first level of product type, refined by the second level or by a price band where one category spans a wide range of prices. We then cut campaigns where the two axes cross: assortment group × label.

Google uses only the first product type value for bidding and reporting. It recommends the full path with ” > ” between levels, such as Home > Women > Dresses > Maxi Dresses (Product type [product_type]). In both product groups and listing groups, you can subdivide by product type up to five levels deep.

The grid grows fast, and this is where segmentation collides with campaign structure: every campaign needs enough conversions to learn on. GetProfit’s portal treats 30 conversions a month per campaign as the minimum and 50+ as comfortable. It suggests a split only if each new campaign would get at least 30 a month.

Owners of stores with many product lines ask a fair question: isn’t one campaign for the whole catalogue absurd? Not necessarily. Every split by category leaves each campaign less data to learn on, and one campaign versus one per category explains when a single campaign does better.

Example store, not client data.

A tableware shop with 3,000 products gets 300 orders a month. It has three assortment groups: plates and bowls; glassware; cutlery and kitchen tools. It plans four tiers: a core of Bestsellers, Strong and Long tail; a below-target tier of Marginal, Loss-making and Dormant; Newcomers; and To remove, which is excluded.

Three groups × three active tiers gives 9 campaigns, and at 30 orders each they would need 270 a month. On paper, 300 is enough. In practice, Newcomers bring almost no orders by definition, Dormant products bring none, and the rest split unevenly.

Say the core brings 240 orders: plates and bowls 130, glassware 50, cutlery and kitchen tools 60. Marginal and Loss-making products bring the other 60, spread across all three groups. A workable structure has five campaigns, not nine:

1. Core, plates and bowls: 130 orders a month.
2. Core, glassware: 50.
3. Core, cutlery and kitchen tools: 60.
4. Below target, from all three groups: 60, with its own budget so these products can’t take money from the core.
5. Newcomers from all three groups: a small fixed budget, judged on clicks and first sales rather than ROAS.

Every product sits in exactly one of these campaigns, except To remove products, which sit in none. If you run labels for many clients, one scheme for many stores sets out what to standardise and what to tune per store.

How to get labels into the feed and the campaigns

1. Pick a free column. Check which custom label columns your live campaigns already read. Your listing groups and product groups select products by label value, so overwriting a column they use moves products out of those groups at the next sync.
2. Keep the values short and stable. Each product gets one value per label, 1–100 characters long. Case doesn’t matter, and each label allows up to 1,000 unique values across the account. Google ignores values over that limit for both reporting and bidding (Custom label 0–4).
3. Choose how the values get in. Write them in your store’s feed export, assign them with Merchant Center rules, or send them in a supplemental feed matched by item ID. Google’s own rule example turns price into bands such as 0-5, 5-10, 10-20 and 20+. We compare rules, supplemental feeds and scripts in automating performance labels. For a ready sheet with the formulas, use the product segmentation template.
4. Allow for the sync. New or edited custom labels can take 24–48 hours to appear in Google Ads (Use custom labels for Shopping ads).
5. Build the groups. In Performance Max, subdivide each asset group’s listing groups by the label. In Standard Shopping, subdivide product groups and set campaign priority where campaigns share products. What to check after: each product sits in one place, and nothing you mean to sell is excluded.
6. Expect a short tail. In Standard Shopping, products you exclude from product groups may still get impressions and clicks for 48 hours, according to Google’s product groups guide.

How often to recompute labels without restarting learning every time is its own question: see how often custom labels should change. When labels sit in the feed but campaigns don’t split the way you set them up, start with custom labels not working.

How to check that your segments actually move budget

A segment that exists only in the feed is a tag. It moves money when three things are true, and you can check each one.

Each segment that needs its own budget or target has one. Map every label to a campaign. If several labels share one Performance Max campaign as asset groups, they share its budget and target: Google sets bids from the campaign’s objective, not per listing group. In Standard Shopping, check that product groups carry the bids you meant, and that campaign priority favours the right campaign where products overlap.

Every product sits in exactly one segment. Count products per label in the feed and per campaign in Google Ads. The totals should match, and the only products missing should be the ones you excluded on purpose. In Standard Shopping, Google states that a product must be in a product group to be eligible for ads at all.

Each segment’s share of spend moves closer to its share of revenue. For each segment, put its share of spend next to its share of revenue, before the change and after it, over windows of equal length. Leave the ramp-up out. Google advises letting a new Performance Max campaign run for at least 6 weeks before you evaluate it. It also advises leaving its first 1–2 weeks out of the analysis (Effectively use Smart Bidding with Shopping and Performance Max campaigns).

Example store, not client data.

A tableware shop spends 40,000 a month and gets 180,000 in revenue, so its ROAS is 4.5. Before the split, its segments look like this:

The below-target tier (Marginal, Loss-making and Dormant) takes 16,000, or 40% of spend. Suppose the new structure gives that campaign 8,000, gives Newcomers a test campaign of 2,000, excludes To remove and keeps the total at 40,000. If the segments move money, the below-target share falls to 20%, To remove falls to zero, Newcomers take 5%, and the core grows from 50% to 75%. If the shares look the same after the evaluation window, the labels didn’t move money: go back to the first two checks.

Watch for two traps when you read the shares. First, compare shares within one report. Performance Max’s Listing groups tab reports product-level data and counts an impression for each product in a multi-product ad, so its totals differ from the campaign tab.

Second, judge success by more than ROAS. Cutting spend on weak segments can raise ROAS while revenue falls, so look at revenue in the core segments as well.

What to do: a segmentation plan in seven steps

1. Measure the starting point. Take 12 months of product data and find the share of spend on products that never sold. What to check after: compare it with the 40.5% median in our study, measured over 13 months, keeping in mind that your niche may differ.
2. Label by behaviour over 12 full months. Judge only products with at least 15–20 clicks; the rest haven’t been tested yet.
3. Choose assortment groups. Start from the first level of product type, and split further only where margin, price or season differ enough to need their own target.
4. Count conversions for each planned campaign. Merge any planned campaign that falls short of 30 conversions a month with a neighbouring one. What to check after: every campaign you keep can reach that number on its own.
5. Write the labels into a free custom label column. Allow 24–48 hours before you build groups on new values.
6. Build the structure. One product, one place. Exclude To remove products, limit Dormant and Loss-making ones rather than switching them off, and give Newcomers a small test budget.
7. Check that the money moved. After 6 weeks, put each segment’s share of spend next to its share of revenue and compare both with the period before. Then re-sort products between campaigns on a fixed schedule. Do it more often than once a quarter, because a stable winner holds its place for a median of 3 months. But leave enough time between re-sorts, or campaigns keep learning on a moving set of products.

If you run ads for a client, agree each structural change with the account owner before you make it.

See which of your products sell and which only spend. Every product gets a label based on how it behaves in ads. The portal changes nothing without your consent.

Check your catalogue →

Sources

• Manage a Shopping campaign with product groups — product group attributes, 7 levels and 5,000 product groups per ad group, a bid per product group, eligibility only inside a product group, 48 hours after exclusion. Checked 2 October 2026.
• Manage a Performance Max campaign with listing groups — listing group attributes, category subdivision by country, 1,000 listing groups per asset group and the custom label recommendation, bids set from the campaign objective, product-level reporting in the Listing groups tab. Checked 2 October 2026.
• Custom label 0–4 [custom_label_0–4] — five labels, Google’s example definitions, limits on characters and unique values, values hidden from shoppers, rules for price bands. Checked 2 October 2026.
• Use custom labels for Shopping ads — one value per product, 24–48 hours for new or edited labels to appear in Google Ads. Checked 2 October 2026.
• Use campaign priority for Standard Shopping campaigns — how priority decides which campaign bids for a shared product. Checked 2 October 2026.
• Product type [product_type] — only the first value is used for bidding and reporting, the full path with ” > ”. Checked 2 October 2026.
• Retailer best practices for AI-powered Performance Max campaigns — consolidation, reasons for separate campaigns, custom labels for high-priority products, the holiday example. Checked 2 October 2026.
• Effectively use Smart Bidding with Shopping and Performance Max campaigns — at least 6 weeks before evaluating Performance Max, ramp-up left out. Checked 2 October 2026.
• How to Segment Your Products by Sales Data Automatically — a feed-tool vendor’s example of 30-day revenue ranking with daily label updates. Checked 2 October 2026.
• GetProfit data: 146 online stores, June 2025 – June 2026 — Performance Max share of ad budget.
• GetProfit study of 1,404,808 products in 130+ stores over 13 months — budget concentration, products with no budget, spend on products with no sales, switch-off accuracy, how long bestsellers last, the minimum test.
• GetProfit portal methodology — the eight labels and their rules, 12-month window, 30 and 50+ conversions a month per campaign, the split rule.