01 / Give the assortment a purpose
Shopping product groups start with a decision
Stock, season and value need a shared operating rule
Segment your Shopping assortment where a distinction leads to a different decision. A variant running low may need a different advertising decision from replenishable core stock; a seasonal item may need a different review date from an evergreen offer. A label describes that situation. Only the product selection, bidding strategy and campaign structure you actually configure turn it into an action.
For small and medium-sized online shops in Germany, a manageable structure that can be checked is usually more useful than a large collection of seemingly precise subdivisions. Start with reliable stock data, a documented seasonal calendar and an agreed economic assessment. Then record which products should be considered for regular advertising, investigated deliberately, assigned to a limited promotion or temporarily removed from a particular advertising selection.
The working tool in this guide is an assortment role matrix, accompanied by a custom-label dictionary and a change log. It connects inputs, precedence rules, product selection and follow-up checks. Purchasing, shop management and campaign managers can then understand the same decision without interpreting a word such as “bestseller” as different instructions.
This guide assumes that your data pipeline works. It does not calculate your return costs from scratch or resolve a Merchant Center disapproval. It explains how to turn verified signals into maintainable segments and what you can actually control in Standard Shopping and Performance Max. It does not promise revenue increases or prescribe fixed minimum budgets.
Current as of 2 October 2026. The roles, thresholds and numerical examples are an editorial working method. Google does not require this classification. A technically correct segment guarantees neither ad delivery nor commercial success.
02 / Keep the levels distinct
Do not confuse labels, product groups and campaigns
More detailed reporting does not create a separate budget
Before assigning fields, establish the purpose of each level. The overview below guides implementation; it is not an invitation to expand every level. For each proposed split, ask whether it is intended simply to support reporting, select products, support different advertising content or genuinely carry a separate campaign decision. Only then can you identify the appropriate technical setting.
| Level | Describes | Practical purpose | Owner | Does not automatically provide | Check |
|---|---|---|---|---|---|
| Product data | A specific offer | Supply facts and attributes | Shop and data management | Approval for every campaign | Correct variant and source? |
| Custom label | Your classification | Identify product sets | Business owner of the rule | Exclusion or a higher bid | Which rule generates the value? |
| Product group | Selection in Standard Shopping | Subdivision and appropriate controls | Campaign manager | A separate campaign budget | Which strategy is active? |
| Listing group | Product selection in Performance Max | Assign offers to an asset group | Campaign manager | An individual manual product bid | Which offers are included? |
| Campaign | Shared control framework | Set objectives and budget | Marketing owner | Equal spending across products | Why is a separate campaign needed? |
A “high contribution” segment may initially be only a reporting group. Your review might show that these items are already receiving appropriate consideration. A new campaign would then add maintenance without a defined purpose. The situation differs when a limited promotion needs an expressly agreed budget envelope that cannot sensibly be managed in the existing structure.
Record the objective and implementation separately. “Give seasonal stock more visibility” is an intention; “include this set of offers in this active selection” is a configuration you can verify. Whether that produces more worthwhile demand requires suitable outcome data. Keeping these apart prevents an attractive account structure from being presented as a successful experiment before any evidence exists.
03 / Reliable inputs
Classify offers at the correct level
A product family can have several stock situations at once
Create a working file containing the stable offer ID, market, language, product type, available stock, stock timestamp, seasonal phase and economic class. Add the source and responsible person. Together, these fields must identify a specific offer unambiguously. Total stock for a product family is insufficient when the advertised size or colour is running low while other versions remain plentiful.
Your own product hierarchy can be represented through product_type. It differs from Google's predefined product category; when several product types are submitted, the first is used for Shopping bidding organisation and reporting. Choose a deliberate primary path. A changing season or economic role should not replace the actual product type in this hierarchy.
Follow one record from your inventory system to the processed offer. Do identity, market and matching agree? If this foundation is missing, use the guide to setting up Merchant Center and a Shopping feed. For segmentation, a shop field simply being called “tag” is not enough. What matters is whether the intended value reaches the correct attribute on the correct offer.
Mark unknown inputs explicitly. A missing cost record is not the same as a poor contribution; a delayed inventory file does not prove that a shelf is empty. Your classification must not convert uncertainty into an apparently certain category. Assign the record a review status and specify who will supply the missing information and which advertising decision applies in the meantime.
04 / The label dictionary
Give every custom label one clear meaning
Short value lists are easier to check than combined codes
Define the dictionary first and the rules afterwards. The custom_label_0–4 specification allows five optional attributes, one value per product in each, and up to 1,000 distinct values per attribute across the account. These are not visible ad texts. Do not confuse this technical ceiling with a sensible number of segments for your shop.
The following scheme is an editorial proposal. It reserves separate fields for stock, season, economic class, data approval and the resulting role. Use it only if those fields are not already serving another purpose. Silently replacing an established meaning could change other campaigns whose managers know nothing about your new project.
| Field | Meaning | Example values | Authoritative source | Update trigger | If information is missing |
|---|---|---|---|---|---|
| custom_label_0 | Stock position | stable, tight, excess | Stock and demand estimate | Relevant inventory update | unknown |
| custom_label_1 | Seasonal phase | evergreen, prepare, active, exit | Approved seasonal calendar | Phase change | unknown |
| custom_label_2 | Economic class | strong, standard, weak | Verified contribution assessment | Change in underlying assumptions | unknown |
| custom_label_3 | Data approval | ready, review | Freshness and plausibility | Every classification run | review |
| custom_label_4 | Advertising role | core, explore, clearance, hold | Documented precedence rule | After input checks | hold |
The English codes provide technical consistency and remain unchanged in every language version. Their business definitions belong in the dictionary. Do not translate “core” into a German export value while an active selection still looks for “core”. Treat renaming as a change to the interface between data management and the campaign.
Keep precise definitions outside the short codes. “strong” might represent an approved contribution band, but it must not mean something different to each department. Raw amounts, calculation period and approval remain in the working register. This makes values explainable without generating a separate label value for every amount, date and person responsible.
05 / Stock as a risk signal
Distinguish fulfilment capability from advertising headroom
Low stock does not automatically mean out_of_stock
The availability requirements concern whether the offer can actually be ordered and fulfilled. Do not mark an orderable offer out_of_stock to restrict advertising. Internally, its advertising classification may still be “tight”: available for delivery, but with little capacity for additional demand. The two fields answer different questions and must remain understandable alongside each other.
For a simple planning indicator of your own, divide available, unreserved units by a justified estimate of daily demand. The result is calculated stock cover in days. This assumes comparable units and a useful demand estimate. Without observed demand, or around a major seasonal change, the result is uncertain; do not enter an artificially precise stock-cover figure.
A hypothetical item has 24 saleable units and assumed demand of 3 units per day. That gives 8 days of cover. If confirmed replenishment is 12 days away, there may be a gap. This is a reason to review advertising headroom, not a universal threshold for a Google bid. Reservations, other sales channels and the reliability of the delivery date also matter.
Define “tight” in relation to your purchasing process. Stock that can be replenished locally may warrant a different assessment from a discontinued model with the same remaining quantity. Likewise, “excess” needs a documented stock or sell-through objective; a large quantity alone does not prove overstock. Record which target quantity or planned period led to the classification.
Prevent daily switching around a boundary. As your own rule, you might require confirmed goods receipt and an adequate buffer before moving from “tight” back to “stable”. The business determines that buffer. More important than a seemingly scientific threshold is that purchasing and marketing interpret the same state in the same way.
06 / Seasonality with an end date
Plan phases instead of a permanent seasonal tag
A calendar label is not a seasonality bid adjustment
An item may be preparing for an upcoming season, within its active demand window or moving into the exit phase. Record the start, end, market, assumption and owner in a calendar. The label contains only the current phase. Campaign managers then do not have to guess whether “winter” means preparation, peak season or clearance.
Google Ads seasonality adjustments address expected substantial conversion-rate changes during short events; Google particularly identifies periods of 1–7 days. A season label instead describes a product set. A multi-month autumn plan should therefore not automatically become a seasonality adjustment. Increased search demand alone is not a measured change in conversion rate either.
Anchor the phase in actual ordering and delivery requirements. For gifts, the last reliable delivery deadline may matter more than the occasion's calendar date. Define separately when the advertising selection should change and when the product can still be sold normally. An expired advertising promotion does not automatically invalidate the product information.
Agree the return rule before activation. Who ends “active” if the owner is unavailable? What happens if delivery is late? A named backup contact and a controlled end date are often more useful than extra subdivisions. Without maintenance, seasonal segments remain active beyond their occasion and may later be mistaken for a deliberate ongoing strategy.
07 / Economic classification
Use verified values instead of renaming ROAS as margin
A class supplies context, not an automatic conversion value
Define the economic assessment before creating the classification. For that earlier step, use the Shopping contribution register for margin, returns and stock. Segmentation imports the approved class together with its period and version. An apparently precise class built on unknown costs remains unreliable even if it appears in the feed without a technical error.
An expensive item is not necessarily more valuable to the business. High revenue may accompany high purchasing, shipping or expected return costs. Conversely, a low contribution class need not trigger immediate total exclusion: a deliberately limited test or a justified clearance objective may still make sense. Such an exception needs a documented purpose and an accountable owner.
With Target ROAS bidding, Google works with reported conversion values. A “strong” custom label transmits no additional profit amount and changes no conversion action. Check what the value in use actually measures. If you are still reporting revenue, a revenue-based ROAS remains different from the contribution assessed internally.
Avoid applying the same adjustment twice. If an agreed value model already accounts for a cost component, do not casually apply another blanket deduction for it in a new target. The role matrix initially needs only a defensible category; changes to measurement or the bidding model should be prepared and reviewed as separate decisions.
Specify when the economic class needs fresh approval. Promotions, changed purchasing terms or better returns data may trigger a reassessment. Keep the previous definition so that a change from “standard” to “strong” is not later mistaken for a sudden improvement in advertising performance. The classification can change even when customer behaviour has not.
08 / Assortment role matrix
Derive one role using an explicit order of precedence
Unresolved data takes priority over an ambition to grow
The following matrix is an editorial decision rule you can copy. Work from top to bottom: the first matching row determines the role. Keep the underlying figures and assessments in your working file; the label carries only the result. The advertising action described must then be configured separately. None of the codes in this table is a built-in Google switch.
| Precedence | Condition | Role | Intended action | Approved by | Recheck when |
|---|---|---|---|---|---|
| First | Data missing or contradictory | hold | Remove from the new selection; investigate | Data owner | Source has been corrected |
| Next | Stock tight | hold | Limit additional advertising pressure | Purchasing and marketing | Replenishment is confirmed |
| Next | Season exit and stock excess | clearance | Consider limited clearance activity | Assortment owner | End date or stock change occurs |
| Next | Season active or evergreen; class strong or standard | core | Consider for regular selection | Marketing owner | Season or value changes |
| Otherwise | Data ready, but no preceding condition applies | explore | Define a justified test or defer | Test owner | Agreed review date arrives |
An attractive contribution value therefore cannot override an unresolved stock situation. A product can be seasonally active and running low at the same time; under this rule it receives “hold”. A business may choose different priorities, but must document them expressly. Incompatible rules in separate spreadsheets otherwise give the same variant a different role depending on import order.
In this proposal, “hold” initially means no automatic inclusion in the new regular selection. Stopping, limiting or retaining another campaign that is already running requires an expressly assigned action. The register must therefore name the campaign, selection and owner. A general label alone does not demonstrate that the product has been excluded from advertising everywhere.
09 / Data path and ownership
Decide which source is allowed to write each label
The classification must survive the next import
Choose an authoritative source for every label. Classes are often calculated in an inventory system, a maintained spreadsheet or an upstream data-processing step. The tool matters less than clear ownership. If a shop app, spreadsheet import and manual edit independently write to the same field, a seemingly successful test can be overwritten at the next scheduled run.
A supplemental product data source can enrich or override existing attributes; it does not replace a standalone primary source and cannot add products itself. With conventional ID matching, both the matching and the data-source context must be correct. Check the account's advanced data-source management prerequisites before assuming this route is available to you.
Alongside the offer ID, retain the source file, validity timestamp, rule version and owner in your working file. These do not all need to be transmitted as custom labels. They establish why a product was “core” yesterday and “hold” today. Without saved inputs, it can be difficult to answer that question reliably after an automated overwrite.
Specify what happens when a data delivery fails. Retaining an old file indefinitely can be as risky as clearing every label. Agree an appropriate freshness limit and fallback decision for each signal. Stock information may need a tighter limit than an approved economic class; treating both indiscriminately as “current” would hide that difference.
Initially test the data path on a small, deliberately chosen set: a regular item, a low-stock variant, a seasonal transition and a missing input. Check each processed value and the resulting product selection. These cases test decision logic; they are not a statistical performance study and should not be sold as proof of improved campaign returns.
10 / Publish the rules
Check the draft first, then the processed result
The order of multiple rules can change the role
When using Merchant Center attribute rules, account for their sequential processing. Google provides a draft, rule testing and subsequent application. An input file that looks correct therefore does not establish the final value. Check which source is read and whether a later rule changes the result again. Your confirmation must refer to the processed product.
For the substantive accuracy of other attributes, use the product feed optimisation register. Here, focus on the classification: permitted value, correct variant, documented precedence rule and correct final result. A role change should not incidentally change titles, prices or identifiers. If another change is necessary, give it a documented reason of its own.
- Preserve the inputs
Save sample records with source, timestamp and expected role. Include missing and contradictory inputs.
- Compare the draft
Compare expected and calculated values. Investigate every discrepancy before applying the rule.
- Check processing
After applying the rule, inspect the same offer IDs and the actual label values in Merchant Center.
- Confirm the selection
Then check included and excluded product sets in Google Ads, together with the residual branch.
Keep a rollback path. It must restore the last verified rule and its corresponding campaign selection. An earlier file alone is insufficient if the selection criteria have changed in the meantime. Record changes to both the data and the advertising account together, including who made them and when.
11 / Standard Shopping
Subdivide product groups around the decision you need to make
Bid controls depend on the strategy actually in use
With Shopping campaign product groups, you select subdivisions using submitted attributes. Under manual CPC bidding, bids can be edited on leaf groups with no further subdivisions. Do not assume that an automated strategy uses the same controls. Check the active bidding setting before assigning a supposedly individual bid target to a group.
A simple implementation starts with the relevant role set and subdivides it by product type if necessary. Do not divide by every available attribute simply because the menu allows it. Ask whether another branch supports a specific decision or a necessary check. Otherwise, it makes maintenance harder without improving the underlying decision.
Check the remaining “Everything else” branch explicitly. A new or misspelled label can end up there. Decide deliberately whether the residual set stays included or is excluded, and test an unknown value. Creating only the desired visible groups and assuming that everything else has automatically disappeared is not a reliable safeguard.
A dictionary change does not rebuild the selection structure on its own. A changed product value can alter membership under existing criteria, but new values and renamed branches require a deliberate reconciliation. Compare item sets before and after processing and inspect individual examples. A branch without products may indicate an error, but it may also be genuinely empty.
12 / Performance Max
Use listing groups to select products
An asset group is not a separate budget account
Google's Performance Max listing group documentation describes their place within asset groups. Bids are determined automatically; another listing group does not create a manual bid for an individual product. Keep product assignment and the creative theme aligned without automatically turning every economic class into a separate asset group or campaign.
Create a new asset group only when it has a defensible purpose for products and creative assets. Different product needs may require different images and messages. A different internal contribution code alone does not produce a different customer proposition. If you expressly need a separate budget decision, examine the campaign level and explain why the added separation justifies its maintenance cost.
Check overlaps across all affected selections. If a variant appears in several asset groups, adding it to your new group does not remove it from the old one. An internal assignment list containing the product set and intended role makes these cases visible. It explains the structure; it should not claim that a particular group is guaranteed a fixed share of expenditure.
Also consider destination URLs. Listing groups restrict the included product offers, but are not a general exclusion of every website address. The final URL and, where applicable, Final URL expansion must also fit the intention. When you want a complete advertising stop, inspect the entire relevant configuration rather than relying only on an offer removed from the product list.
13 / Pausing and excluding
Specify the exact scope of every pause
Stopping advertising, stopping orders and ending a season are different states
For a short advertising pause, Google documents the pause attribute. The ads value concerns advertising destinations; the documented short-term limit is 14 days. Do not plan a longer removal as an indefinitely extended short pause. Record the return date and verify the processed state. A saved intention is not yet a confirmed stop.
The excluded_destination attribute excludes a particular destination. Shopping_ads has a broader scope than selection within one campaign and also affects availability in the product-group selector. Check whether you intend to change only one campaign or exclude the offer from that destination generally. Other destinations and website content need separate consideration.
Do not use this level as a substitute for unresolved assortment decisions. A deliverable offer can remain available in the shop while advertising for a particular promotion ends. If orders are genuinely no longer being accepted, however, actual availability must be correct. A marketing label can neither override that fact nor conceal it from customers.
“hold” becomes an effective advertising restriction only when the relevant selection or setting has been changed. Document the scope, start, owner and resumption. Do not promise that data changes take effect immediately without checking their status.
14 / A worked example
Test the matrix on three different offers
The same contribution class can lead to different roles
This example is entirely hypothetical. A shop classifies three offers using the dictionary above. Their economic classes have already been approved; calculating them is outside this example's scope. Every decision initially concerns a new, clearly named advertising selection. Other active campaigns are checked separately for the same products.
Offer A is seasonally active, economically strong and tight on stock. Although it looks attractive, the stock rule takes precedence: its role is hold. The owner checks replenishment and expressly decides which ongoing advertising should be limited in the meantime. Only a confirmed stock change justifies reclassification; a desire for additional revenue does not.
Offer B is in the exit season phase, has excess stock and a known economic class of standard. Its role is clearance. Before inclusion in clearance activity, price, timing, creative assets and ownership must be agreed. The role does not automatically authorise a discount or a particular budget. If approval is not obtained, the intended action remains unresolved.
Offer C is evergreen, has stable stock, a standard class and ready data. It receives core and may be included in the regular selection. If its stock source becomes unusable on the next run, the data rule takes precedence. The role changes to hold in a controlled way and creates a specific investigation task; the last good state is not silently presented as current.
The demonstrable success of this exercise is limited: expected roles agree with processed values and the confirmed selection. Whether clearance is profitable or the regular selection performs better requires later evidence. This keeps technical acceptance separate from an economic effectiveness assessment while still providing an immediately useful checkpoint.
15 / Comparable reporting
Retain a record of which products belonged to each segment
A different mix can explain a better average
Save the offer set and rule version whenever you make a material change. Removing weaker items from a segment can improve its average even when no remaining product performs better. Alongside the new groups, therefore, compare a documented common product set. Record arrivals, departures and changed classifications.
Allow for the time between the advertising interaction and purchase. Google explains this effect in its conversion lag reporting guidance; the report views it describes are not universally available for every campaign type. Recent periods may still be incomplete. Base decisions on sufficiently mature data and record the reporting cut-off date.
Also distinguish product-level reporting from campaign-level reporting. An ad containing several products may be counted differently at product and ad levels. A discrepancy is therefore not automatically a tracking error. Check the level and period each metric represents before combining tables or calculating apparently missing performance.
Record simultaneous changes to prices, promotions, budgets and delivery conditions. A before-and-after comparison alone does not isolate segmentation's effect. It can provide clues, but cannot establish incremental revenue. First determine whether the structure fulfils its agreed purpose and whether observed commercial outcomes justify continuing with it.
16 / Maintenance and acceptance
Connect every trigger to a follow-up check
The matrix is complete only when it has owners and a rollback path
Add the following maintenance protocol to your role matrix. For each change, record the date, affected IDs, expected state, actual finding and approval. The table shows example triggers. Frequency depends on your operations and data paths; it is not a daily work schedule imposed by Google.
| Trigger | Owner | Check | Approve when | Rollback if incorrect |
|---|---|---|---|---|
| New inventory update | Purchasing and data management | Freshness, reservations, variants | Label explains the actual state | Use the verified fallback decision |
| Seasonal transition | Assortment planning | Market, date and delivery window | Phase and selection agree | Restore the previous valid plan |
| New value basis | Economic assessment owner | Definition, period and assignment | Class has documented approval | Retain the previous verified version |
| Rule change | Feed owner | Draft, processing and residual set | Expected cases are mapped correctly | Revert the rule and selection together |
| Campaign change | Marketing owner | Included sets and overlaps | Intended scope is confirmed | Restore the documented selection |
Stop a new approval when product counts differ without explanation, values are unknown or ownership is missing. “The upload succeeded” answers none of these questions. A useful sign-off states which products are included where, which are deliberately absent and how exceptions are handled. A specific unresolved discrepancy remains a visible task instead of disappearing inside a blanket green status.
After introduction, check the next regular data run as well. Only then can you see whether the ongoing source preserves the intended classification. Agree a short, repeatable control routine afterwards. For collaboration, a smaller structure maintained reliably is more valuable than a detailed matrix with no accountable owner.
17 / Frequently asked questions
Questions about segmenting Shopping products
Does every shop need all five custom labels?
No. Use fields only where they have a clear purpose. The dictionary shown is a proposal. Check an existing meaning before reusing a field.
Does a strong label automatically raise bids?
No. It classifies a product. The active strategy and configured structure determine the controls. A label neither changes the reported conversion value nor guarantees extra spending.
Can low-stock but orderable goods be marked out_of_stock?
Not as an advertising switch. Actual ordering and fulfilment information must remain correct. Use an appropriate selection or setting for the intended advertising scope and verify its effect.
Does every season need its own campaign?
No. A different calendar status initially justifies a classification. A separate campaign needs an independent control purpose, such as a justified budget envelope, and a maintainable data foundation.
Can I bid manually per listing group in Performance Max?
No. Listing groups organise included products; bids are determined automatically. An additional group is neither an individual CPC bid nor a separate budget.
What happens to new, unknown label values?
Check the residual branch and actual assignment. New values do not automatically create your intended selection structure. The dictionary needs an explicit rule for unknown inputs.
How often should the role matrix be updated?
When the agreed relevant changes occur. Stock, season and economic inputs have different freshness requirements. Define the source, owner and fallback decision if data fails to arrive.
Does better segment ROAS prove that the change helped?
No. Product mix, data maturity, promotions and budget can alter the comparison. Check comparable product sets and values first; a simple before-and-after number does not isolate a causal effect.
18 / Begin implementation
Start with a dictionary and a selection you can verify
Useful segmentation connects data maintenance with campaign decisions
Choose a manageable part of the assortment. Confirm identity, sources and the economic definition, write down precedence rules and test expected roles on specific offers. Only then transmit the values and configure the appropriate product selection. Every technical change remains traceable to a business decision.
Before starting, document included sets, the residual branch, owners and the rollback path. After processing, inspect the affected products; after the next regular run, verify stability. This checked foundation can grow into a larger structure once another subdivision solves a real problem.
If you want to build or review product data, custom labels and the appropriate advertising selection together, Salestudia can help with setting up and optimising Google Merchant Center.