Most advertisers arrive at Performance Max carrying a rule they learned somewhere else, and it is the wrong rule.
The rule is consolidate. Fewer campaigns, more data in each one, let Smart Bidding learn. For Search campaigns that advice is correct and we have argued for it ourselves, with data: all 24 of our Search campaigns were below the learning threshold because the accounts were split too many ways. Split a Search account into eight campaigns and none of them will produce enough conversions to bid intelligently.
Then people carry the same instinct into Performance Max and Shopping, put everything into one campaign, and wonder why half their budget disappears into products that never sell.
For Shopping and Performance Max, consolidation is not the holy grail. The default should be the opposite. Unless you have a compelling reason to consolidate, segmentation will usually get you better results.
That single reversal is the thing most PMax guides never explain, and it has a specific technical cause that is worth understanding before you touch anything.
This is a long article because the topic genuinely is complicated. If you want the short version: segment when you have enough conversion volume to feed each container and a real business reason to separate them, consolidate when you do not, and never segment because it looks tidier in a report.
Part 1: Why Performance Max is not Search
The reason you cannot fix this with asset groups
Here is the constraint that determines everything else.
In a Standard Shopping campaign you can set bids at the ad group level. Manual CPC, or a Target ROAS on the ad group itself. That means one campaign can genuinely contain several different bidding intentions.
In Performance Max, bids cannot be adjusted at the asset group level. There is no target field. There is no bid field. The bid strategy and the efficiency target live at the campaign level and nowhere else.
The same is true of budget. One campaign, one daily budget, no matter how many asset groups you build inside it.
So when somebody tells you to control your best products separately, or protect your margins on a low-margin category, or push harder on new arrivals, the honest answer is that there is exactly one lever available: a separate campaign.
Asset groups do a different job. They exist so your creative stays relevant to the products it is shown alongside, and so you can read performance for a chunk of your catalogue. They cannot carry a different target, a different budget, or a different bidding intention. Once you internalise that, the whole segmentation question becomes clearer, because every “how do I control X” question in Performance Max resolves to “do I need a campaign for X.”
What the campaign structure is actually for
It is worth being explicit about the goals, because a lot of bad structure comes from never asking.
A campaign structure exists to align your Google Ads setup with your business goals, to take back a degree of control by pointing Google in a direction, to feed the algorithm information it can act on, and to let you make bidding decisions based on your real efficiency and growth targets.
An asset group structure exists to give you an overview you can report on or act upon, and to keep your creative relevant to what it is selling.
Those are different jobs. Confusing them is where oversegmentation starts.
The question is not what the perfect structure is
There is no perfect Performance Max structure and anyone selling you one is selling you a template.
The useful question is: how do I work out what the best structure is for this specific account, this catalogue, this conversion volume, this business goal, this level of available time?
Everything below is a way of answering that question rather than a structure to copy.
Part 2: The golden rule, and the eight reasons to break it
The rule
Unless you have a compelling reason to consolidate your Shopping and Performance Max campaigns, use advanced segmentation to drive better results.
Note the direction. Segmentation is the default. Consolidation is the exception that needs justifying. This is the exact inverse of how you should think about Search.
The eight compelling reasons to consolidate anyway
1. You do not have enough conversion data to segment. The working threshold is 50 conversions per month. Our Campaign Structure Recommender will tell you whether your volume supports the structure you are considering. Below that, splitting a campaign in two gives you two campaigns that cannot learn instead of one that can.
2. Segmentation by your chosen criteria would leave individual buckets starved. You might have 200 conversions a month overall and still fail this test if you are trying to split into six buckets.
3. Segmentation is not immediately necessary. New market, new product line, no history. There is nothing to segment on yet. Consolidate, gather data, revisit in ninety days.
4. The campaign runs for a short duration. A three-week seasonal promotion will end before a performance-based segmentation has finished learning.
5. You have resource constraints. More campaigns means more monitoring, more budget management, more chances for something to drift unnoticed. If nobody has time to manage six campaigns properly, six campaigns is the wrong answer.
6. You have limited expertise managing complex structures. This is an honest reason, not a failure. A well-run simple structure beats a badly-run complex one every time.
7. Your return on effort is higher elsewhere. If the feed is a mess, the landing page converts at 0.4%, or conversion tracking is measuring the wrong thing, restructuring campaigns is not where the next win lives. Fix the foundation first.
8. You want to minimise management cost. Legitimate, particularly on smaller accounts where the fee has to make sense.
If none of those eight apply, segment.
The eight reasons to segment
The mirror image. Only create a separate campaign if at least one of these is true:
You want to isolate an experiment
You have a different business objective for that segment, meaning a different bid strategy or a different target
You want to allocate a specific budget to it
You want to use campaign-specific conversion goals
You want campaign-specific settings, such as New Customer Acquisition or different location targeting
You have business insights the algorithm cannot possibly have
You need separate reporting for a category or segment
You are seeing attribution discrepancies for different segments in your third-party tools
“It would look neater” is not on that list. Neither is “the client likes seeing it split out,” which is usually solved with a report rather than a campaign.
Reactive versus proactive segmentation
This distinction decides whether a tactic will work in your account at all.
Reactive segmentations are based on what has already happened. Performance-based bucketing is the obvious example: you look at how products performed and sort them accordingly. These require a lot of data, need 30 or more days before they start working properly, and do not work at all for short campaigns.
Proactive segmentations are based on what you already know before any data arrives. Margin, seasonality, inventory levels, price competitiveness, new versus established products. You do not need history to know which of your products have a 12% margin and which have 60%.
If you are data-poor, proactive segmentation is available to you when reactive segmentation is not. That is the practical route for smaller accounts that keep being told they lack the volume to segment.
Part 3: The ten segmentation tactics
These are the tactics worth knowing, roughly in order of how often they earn their keep.
1. Segment by category
Difficulty: easy. Type: proactive.
Why you would: to allocate budgets by category performance or business priority, to set different targets per category based on margin, and to analyse category performance properly.
How: populate your product_type attribute so it reflects your real category structure. Create separate campaigns for each mid to high volume category. Use ad groups or listing groups for subcategories and brands underneath. Adjust budgets and targets from each category’s own data and goals.
This is the simplest effective segmentation and it is usually the right first move on a new account. The warning that goes with it is the same one that goes with all of these: do not oversegment. Watch your conversion volume per campaign.
2. Segment by performance
Difficulty: hard. Type: reactive.
The Heroes, Sidekicks, Villains and Zombies approach. This gets its own full section below because it is both the most effective and the most commonly botched.
3. Bestsellers versus regular products
Difficulty: easy to medium. Type: proactive or reactive.
Why: your bestsellers deserve budget certainty and usually justify a more aggressive target. Mixing them with the long tail means competing internally for the same budget.
How: custom label your bestsellers, split them into their own campaign, and set the target that reflects what they are actually worth rather than a blended account average.
4. Products on sale versus regular price
Difficulty: easy. Type: proactive.
Why: discounted products convert at a different rate and carry a different margin. Both facts argue for a different target.
How: use sale_price or a custom label driven by your promotion schedule. This is one of the few segmentations that genuinely justifies itself on short campaigns, because the products move in and out of the bucket automatically if your feed is set up properly.
5. Segment by inventory level
Difficulty: medium. Type: proactive.
Why: advertising products that are about to go out of stock wastes money and irritates customers. Advertising overstocked products harder clears inventory.
How: pipe stock levels into a custom label from your ecommerce platform. Bid down or exclude products below a stock threshold. Bid up on overstock.
This one pays for itself fastest in accounts with volatile inventory.
6. Segment by value
Difficulty: medium. Type: proactive.
Why: a $15 product and a $900 product do not belong under the same target. A blended ROAS target will systematically overspend on one and underspend on the other.
How: split by price tier, then set targets that reflect the actual margin and average order value in each tier. The Ecommerce Unit Economics Calculator is the quickest way to get those numbers straight.
7. Segment by price competitiveness
Difficulty: medium to hard. Type: proactive.
Why: if you are 20% more expensive than the market on a product, no amount of bidding fixes that. If you are 15% cheaper, you should be bidding much harder than your average.
How: use Merchant Center price competitiveness data or a third-party feed tool to label products, then treat the competitive and uncompetitive groups differently.
8. New products versus established products
Difficulty: easy to medium. Type: proactive.
Why: new products have no history, so in a consolidated campaign the algorithm will ignore them in favour of proven sellers. Separating them forces exposure and gets you data faster.
How: label by launch date, run a dedicated campaign with a deliberately loose target for a fixed period, then graduate products into the main structure.
9. Segment by seasonality
Difficulty: medium. Type: proactive.
Why: to shift budget toward products entering their season and away from products leaving it, before the performance data catches up.
How: label products by season in the feed, then move budget on a schedule rather than reacting after the fact.
10. Query-level sculpting
Difficulty: hard. Type: proactive.
Why: to influence which campaign catches which type of query rather than letting internal matching decide for you.
This is the most advanced tactic on the list and it depends on understanding how Google resolves internal competition, which is the next section.
Part 4: How internal matching decides which campaign wins
You cannot sculpt what you do not understand, and a lot of segmentation failures are really matching failures.
When a search query could be served by more than one of your campaigns, the priority runs roughly like this:
First, an exact match keyword identical to the search term. If somebody searches “smart wallet” and you have [smart wallet] as an exact keyword in a Search campaign, that wins over broad and phrase keywords and over Performance Max.
Second, phrase or broad keywords, or Performance Max search themes, that are identical to the search term.
Third, relevance. If nothing is identical, Google decides which keyword or search theme is most relevant based on meaning, and on the combination of keyword and landing page in the ad group.
The practical consequence: your search themes function like keywords in this hierarchy. If you add a search theme that duplicates a keyword you already have in a Search campaign, you have created internal competition rather than incremental reach. This is the single most common way advertisers cannibalise themselves without noticing.
It is also why segmentation and cannibalisation have to be thought about together. Splitting into more PMax campaigns without controlling what each one is eligible for just multiplies the number of ways your campaigns can compete with each other.
Part 5: Performance-based segmentation in full
This is the most popular advanced segmentation and the one with the most ways to go wrong.
The four labels
ProductHero introduced the naming that the industry now uses, so credit where it is due. Their published analysis across thousands of Shopping and Performance Max campaigns found four consistent product behaviours:
Heroes. Fewer than 10% of your products generate 80% or more of your revenue. These are the ones worth protecting and pushing.
Sidekicks. Products that convert well but do not get enough visibility. They are performing near target and have room to improve if given budget.
Villains. Around 50% of advertising spend goes to products that do not convert. Half your budget, on average, funding products that are not earning it.
Zombies. More than 60% of products barely receive impressions or clicks. They are not failing, they are simply never being shown.
Read those two percentages together, because they are the argument for this entire approach. Roughly half your money goes to products that do not sell, while roughly two thirds of your catalogue never gets a chance to try.
The goal is to activate the Zombies and level up the Sidekicks, while shifting spend away from Villains and toward Heroes.
Why a consolidated campaign creates this problem
Left alone, Performance Max goes after the easiest conversions available. That is not a flaw, it is the objective it was given. It finds the products that already sell and pushes them, because that is the fastest route to the target you set.
The consequence is that your catalogue’s long tail never gets tested. Kirk Williams of ZATO Marketing makes the point that Google will not have enough budget or time to test every product in a PMax campaign, with an estimated 20% of the algorithm’s work going to testing, and asks reasonably why it should bother testing the tail when it is already hitting target on the core products.
That is exactly right, and it is why the fix has to be structural. You cannot ask a campaign to do something its objective actively discourages. You have to put the neglected products somewhere they are the only option.
The three strategies
Which one you use depends on your business goal, and on how much conversion volume you have to spread around.
Strategy 1: focused on revenue.
Structure: one campaign containing Heroes, Sidekicks and Villains. A second campaign containing only Zombies.
You are isolating the products that never get shown and forcing the algorithm to show them. Suited to smaller accounts with lower conversion volume, because it only requires two buckets.
Strategy 2: focused on profitability.
Structure: one campaign containing Heroes, Sidekicks and Zombies. A second campaign containing only Villains.
You are isolating the budget drain and restricting it. Also suited to smaller accounts, same reason.
Strategy 3: focused on both, full separation.
Structure: four campaigns, one per label.
This gives you maximum control and maximum insight. It is only viable if you have 30 or more conversions per bucket per month, excluding Zombies, which by definition will not have them yet. If your total is 60 conversions a month, this structure is not available to you no matter how appealing it looks.
Strategy 3, alternative: combined volume.
Structure: Heroes and Sidekicks together in one campaign, Zombies in a second, Villains in a third.
The compromise. You merge the two buckets most likely to be volume-starved individually so that the combined campaign clears the threshold, while keeping the two buckets you actually want to control separately.
The things that quietly break this
Set the same target ROAS on each bucket. This is counterintuitive and it catches people out. The instinct is to set a lower target on Villains to restrict them and a higher target on Heroes. Do not. The segmentation itself is doing the work of separating them. Setting different targets per bucket on top of that double-counts the effect and produces behaviour you did not intend.
Set a lower target in your labelling tool than in Google Ads. The labelling tool’s threshold and your Google Ads target are not the same number and should not be. If they match, products oscillate between buckets constantly.
Review your thresholds regularly. The definition of a Hero changes as the account changes. A threshold set in January against a 4x ROAS goal is wrong by June if the goal moved to 6x. Stale thresholds silently mislabel your catalogue.
Understand that products move. Labels update daily in most tools, which means a product can be a Villain today and a Hero next week, and it will move between campaigns automatically. That is the point of the system, but it also means your campaign budgets need to be able to cope with the composition changing underneath them.
Do not segment while your brand terms are still in the mix. Covered in the next section, and it invalidates everything above if you get it wrong.
The tools
You cannot do this manually at any real catalogue size. The main options:
ProductHero Labelizer. The original, and the reason everyone uses this vocabulary. Straightforward, accessible for smaller accounts, works through a supplemental feed in Merchant Center. The fair criticism, made by competitors, is that it works off historical performance and lacks predictive depth, so it can be slow to react.
smec Campaign Orchestrator. More sophisticated multi-dimensional scoring that incorporates first-party data like margin and inventory, aimed at larger accounts that want to move beyond what native automation offers.
FlowBoost Labelizer Script. A script-based alternative for advertisers who want control over the logic without a subscription.
Profitmetrics Shopping Booster. Margin-led, useful if profit rather than revenue is the actual goal.
Whichever you pick, the labels have to reach Google Ads through a supplemental feed in Merchant Center, and there is a lag of a few hours to a day before a new custom label becomes usable for targeting in Google Ads. Plan launches around that rather than being surprised by it.
One practical constraint worth knowing before you design anything: you only get five custom labels. Every segmentation you want to run simultaneously has to fit within those five fields. This is the real ceiling on how clever your structure can get.
Part 6: Brand exclusion, the segmentation nobody calls segmentation
If you do only one thing from this article, do this one.
Why it matters more than any other split
Brand terms inflate Performance Max data severely. People searching your brand name were going to convert anyway. They are cheap conversions and there are usually a lot of them.
Two things follow. First, your data becomes untrustworthy. The campaign looks far better than it is, and you have no real idea how it is performing on the traffic that actually needs winning. Second, and worse, Smart Bidding uses those cheap brand conversions as headroom to overspend on non-brand terms. We measured a version of this problem directly: one account’s inflated conversions moved our PMax cost per lead benchmark by 5x. The brand traffic subsidises waste elsewhere and hides it at the same time.
The goal is to scale the account, not the campaign
This is the mindset shift.
If you do not exclude brand, Performance Max looks excellent. Its numbers are wonderful. Everyone is pleased.
But the goal was never to make one campaign look good. It was to grow the business. Performance Max is one piece of the account, not the account itself, and a PMax campaign that looks brilliant because it is eating your own brand traffic has not added anything. It has just moved conversions from a cheaper campaign into a more expensive one and taken credit for them.
How to do it
There are two mechanisms and they are not equivalent.
Campaign-level brand exclusions. The native feature. Uses your registered brand list. Easier to maintain but less precise, and it will not catch every misspelling and variant.
Campaign-level negative keywords. More precise and more work. Catches the variants that brand exclusions miss.
Use both. They cover different gaps.
Then build the fallback
Excluding brand from PMax without catching it elsewhere means you have simply stopped bidding on your own name, which is worse than the problem you were solving.
Build dedicated Branded Search and Branded Standard Shopping campaigns to catch that traffic. Performance Max catches your non-branded queries. Search and Standard Shopping catch your branded ones.
One caution: since auction dynamics changed, non-branded terms increasingly leak into Branded Shopping campaigns. Exclude them frequently and aggressively. If your negative keyword lists fill up quickly, switch your exclusion thresholds from impressions to clicks, and automate the whole thing with a script rather than doing it by hand every week.
Part 7: Lead generation is a different game
Everything above assumes a product feed, which puts it in Shopping and ecommerce territory. Most lead generation accounts do not have one, and the segmentation logic changes accordingly.
Start consolidated
For lead gen, the default flips back toward consolidation. Our full lead generation playbook covers the rest of that setup. Without a good reason to segment, work with multiple asset groups inside a single campaign, keep your data points together, and give the algorithm as many examples as possible.
The reasons to break that are the familiar three: you need different daily budgets per service, you need different targets or bid strategies, or you are optimising toward different conversion goals.
The Fire and Ice structure
Once you do have a reason, the most useful advanced structure for lead gen splits by channel temperature rather than by service.
The problem it solves: a single Performance Max campaign pushes the same assets across every network, optimises for the same conversion goal everywhere, and gives you no way to control bids or budgets per network. That is fine when every network is doing a similar job. It falls apart when they are not.
Ask yourself the question that exposes it: who is realistically going to book a demo or a consultation call with an unknown B2B company from a Display or Discover ad?
Nobody. Which means a single campaign is asking cold traffic to complete a hot-traffic action, failing, and then optimising away from those placements entirely.
The structure separates them:
Fire, focused on warm and hot channels, running lead generation. Higher target CPA, specific audience signals, optimising toward your high-value conversion actions, with assets written for people close to a decision.
Ice, focused on cold channels, running prospecting. Lower target CPA, broader signals, optimising toward lower-value conversion actions, with assets written for discovery. This is also where New Customer Acquisition belongs, bidding only for new customers.
Asset groups inside each sit per service.
Match the offer to the temperature
The structure only works if the offer changes too, which is the part people skip.
Cold, top of funnel: infographic, whitepaper, cheatsheet, checklist, coupon. Warm, middle of funnel: case study, live demo, webinar, ebook, trial. Hot, bottom of funnel: consultation, quotation, contact, trial, demo.
Putting a “book a consultation” call to action into your cold campaign and then blaming Performance Max for poor lead quality is not a structure problem. It is an offer problem wearing a structure problem’s clothes.
The bid strategy floor still applies
Performance Max has hard preconditions before target-based bidding is valid at all. Conversion tracking implemented and verified, one primary conversion action defined, a clear campaign objective of volume or value, and asset groups consolidated and correctly mapped.
Then the volume thresholds:
We found seven of our ten Target CPA campaigns sitting below this line, so it is worth checking rather than assuming.
- Under 25 conversions a month: Target CPA and Target ROAS are not viable
- 25 to 49: allowed, but unstable
- 50 or more: eligible
Every campaign you create by segmenting has to clear those on its own. This is the arithmetic that kills most ambitious structures, and it is worth doing before you build rather than after.
Part 8: What goes wrong when you do not segment
Half your budget funds products that do not convert. Around 50% on ProductHero’s numbers. In a consolidated campaign you cannot see it and you cannot stop it.
Most of your catalogue never gets shown. Over 60% of products barely receive impressions. Those are not bad products, they are untested ones, and some of them are Heroes you have never met.
You cannot set different targets for things that genuinely have different economics. A 60% margin product and a 12% margin product sitting under one blended ROAS target means you are systematically underbidding on the profitable one and overbidding on the thin one.
New products get ignored. They have no history, so the algorithm has no reason to test them when it can hit target on proven sellers instead.
Your reporting tells you nothing actionable. Account-level ROAS is an average of things you cannot separate. You know the number. You do not know what to do about it.
Brand conversions hide everything. If brand is still in the campaign, all of the above is happening behind a good-looking headline figure.
Part 9: What goes wrong when you oversegment
The opposite failure is just as expensive and much more common among advertisers who have read one article about segmentation.
Every campaign drops below the learning threshold. This is the big one. Split 60 conversions a month across four campaigns and you have four campaigns at fifteen conversions each, all of them below the point where target-based bidding functions. You have taken one campaign that could learn and made four that cannot.
Budgets get stranded. Six campaigns with six daily budgets means money sitting idle in campaigns with no demand today while the campaign that has demand hits its cap by lunchtime.
Internal competition multiplies. More campaigns eligible for overlapping queries means more auctions where you are bidding against yourself.
Every change resets learning everywhere. With more campaigns you make more changes, and each one restarts a learning period in a campaign that was already short of data.
Management overhead exceeds the benefit. At some point the time cost of monitoring the structure is larger than the performance gained from it, and nobody notices because the time cost does not appear in the account.
It stops matching reality. Structures built around last year’s catalogue, last year’s margins and last year’s goals quietly become wrong, and nobody rebuilds them because rebuilding is a big job.
Part 10: The wrong practices that are everywhere right now
These are the ones we see most often in accounts we take over.
Copying a structure from a case study. Somebody read that a retailer went from 450 campaigns to three, or that four-bucket performance segmentation transformed an account, and implemented it on an account with 40 conversions a month. The structure was right for that account. Structure is not transferable.
Segmenting before excluding brand. Everything you measure to decide your segments is contaminated by brand conversions. You will label products Heroes because brand traffic happened to land on them. Exclude brand first, gather 30 days of clean data, then segment.
Segmenting to make reporting look nicer. The client wants to see categories separately, so somebody builds a campaign per category. Ad group reporting and segment views already do this without splitting your conversion data. Build the report, not the campaign.
Believing asset groups give you control they do not. People build ten asset groups expecting to steer budget and targets per group. There is no bid or budget control at asset group level. The structure looks sophisticated and does nothing.
Setting different targets per performance bucket. Feels logical, breaks the model. The segmentation is already doing that job. Set the same target across buckets.
Adding search themes that duplicate existing keywords. Search themes compete in the same priority hierarchy as keywords. Duplicating your Search keywords as PMax search themes creates cannibalisation and calls it reach.
Running target-based bidding on segmented campaigns that are below threshold. Splitting a campaign does not reduce the volume requirement for each piece. Under 25 conversions a month, targets are not viable, and no amount of structural elegance changes that.
Leaving Final URL Expansion on without thinking about it. In a segmented structure this quietly undermines the segmentation, because traffic gets routed to pages you did not intend for that campaign.
Never revisiting thresholds. A Hero threshold set against an old ROAS goal keeps labelling products against a target the business abandoned months ago.
Assuming more structure equals more skill. It usually signals the opposite. The best structures are the simplest ones that still respect real differences in the business.
Treating New Customer Acquisition value as real revenue. If you assign extra value to new customers, your reported value exceeds your actual revenue, and you cannot deduct that assigned lifetime value when the customer returns and buys again. The reports look great and mean less than they appear to.
Fixing structure when the problem is somewhere else. The required fix order is conversion tracking, then goals and KPIs, then campaign structure, then unit economics, then bid strategy, which is the same sequence as the 9 pillars of a profitable Google Ads account. Restructuring an account with broken conversion tracking just gives you a better-organised version of the same wrong data.
Part 11: A case study in knowing when to stop
Worth including because the conclusion is not the one most segmentation articles reach.
The account: eight product categories, two of them major, around 100 SKUs total. The two major categories had been split into separate Performance Max campaigns, alongside Search, Display, Video and Discovery campaigns.
The problem: full-asset Performance Max performance began declining on one of the major categories.
First intervention: switch that category to a Feed-only Performance Max campaign. Result: spend down 43%, ROAS up 10%. Better, but still performing below target.
Diagnosis: a lot of mismatching and waste in the search terms. Our Search Term Negative Keyword Analyzer is built to surface this kind of leakage quickly. The campaign was being served on queries it had no business appearing for, and Performance Max offered no adequate way to exclude them.
Second intervention: revert that category to Standard Shopping entirely, to get proper control over search term exclusions.
Result: ROAS up 66%, back above target, with real control restored.
The lesson is that segmentation is not always the answer, and Performance Max is not always the answer either. Sometimes the correct structural decision is to move a category out of Performance Max completely. That is worth remembering in a market where the assumption is that PMax is where everything ends up.
It is also worth noting the broader trend: a growing number of ecommerce advertisers have stopped running Performance Max alone and now pair it with Standard Shopping specifically to regain the query-level control PMax does not offer.
Part 12: The diagnostic sequence when Performance Max underperforms
Before restructuring anything, work through these in order. Restructuring is expensive and often the wrong lever.
- Is the offer good enough? High clicks and CTR with low conversion rate points at the offer or the landing page, not the campaign. The Vuut Offer Grader is a quick way to pressure-test it.
- Where was the budget actually spent? Look at channels, placements, products and search terms.
- Which networks are consuming the spend? Use a script for this, since Google’s native reporting will not tell you properly. Mike Rhodes’ PMax script is the standard tool.
- Are you pushing the right products? If not, do you need product ID exclusions or a genuine structural change?
- Are your targets set correctly? A target below what the campaign can achieve chokes delivery rather than improving efficiency.
- Are your budgets set correctly?
- Are your assets causing the underperformance?
- Does the product feed need work? Titles, images and price carry roughly 90% of the weight.
- Is there significant waste in search term insights?
- Are the settings right? Location targeting, Final URL Expansion, brand exclusions.
Structure is question four. Nine other things come first or alongside.
Part 13: The decision framework
Run through this before you build anything.
Step 1. Is the foundation sound? Conversion tracking verified, one primary conversion action, clear objective. If not, stop. Fix that first, and read why Google Ads generates bad leads if the tracking is measuring the wrong thing.
Step 2. Have you excluded brand and built the fallback campaigns? If not, do that and wait 30 days for clean data before making segmentation decisions.
Step 3. What is your total monthly conversion volume? Under 50, consolidate and revisit later. Over 50, continue.
Step 4. Divide that volume by the number of campaigns you are proposing. Does each one clear 30 a month, excluding any Zombie bucket? If not, reduce the number of buckets until it does, or use a combined structure.
Step 5. Which of the eight reasons to segment applies? If you cannot name one, do not segment.
Step 6. Reactive or proactive? If you lack the data or the time horizon for reactive segmentation, use a proactive one. Margin, inventory, price tier and product age need no history.
Step 7. Does it fit in five custom labels? If not, prioritise.
For a wider health check across tracking, structure, bidding and unit economics, the 9-Pillar Google Ads Scorecard covers all of it in one pass.
Step 8. Who is monitoring it, and when is the review? Put a date on the calendar to check thresholds and volumes. Structures decay.
FAQ
Should I segment my Performance Max campaigns or consolidate them? For Shopping and Performance Max, segmentation should be your default and consolidation the exception you justify. This is the opposite of Search, where consolidation is usually right. The reversal exists because PMax gives no bid or budget control below campaign level, so a separate campaign is the only way to treat part of your catalogue differently.
How many conversions do I need before I can segment? Around 50 a month total as an entry point, and 30 or more per resulting campaign per month for target-based bidding to work in each one. Under 25 per campaign, Target CPA and Target ROAS are not viable at all.
Can I control bids at the asset group level? No. Bids and budgets exist only at campaign level in Performance Max. Asset groups control creative relevance and give you a reporting view. This is the single most important constraint to understand.
What are Heroes, Sidekicks, Villains and Zombies? A performance labelling framework popularised by ProductHero. Heroes are the under 10% of products driving 80% or more of revenue. Sidekicks convert well but lack visibility. Villains absorb roughly half your budget without converting. Zombies are the 60% plus of products that barely get impressions.
Do I need a tool to do performance segmentation? At any real catalogue size, yes. ProductHero Labelizer, smec Campaign Orchestrator, FlowBoost Labelizer Script and Profitmetrics Shopping Booster are the main options. Manual labelling does not survive contact with a moving catalogue.
Should I set different ROAS targets for each performance bucket? No. Use the same target across buckets and let the segmentation do the work. Setting different targets on top of the split double-counts the effect.
Do I need to exclude brand before segmenting? Yes, and it is not optional. Brand conversions inflate your data, so any segmentation decision made on unfiltered data is being made on numbers that are wrong. Exclude brand, build Branded Search and Branded Shopping fallback campaigns, then wait for clean data.
How does segmentation work for lead generation without a product feed? Start with multiple asset groups in one campaign. Segment only for different budgets, different targets or different conversion goals. The main advanced structure splits by channel temperature, running one campaign for warm and hot channels and another for cold prospecting, with the offer matched to each.
How often should I review my structure? Thresholds and bucket definitions monthly. The structure itself quarterly, or whenever margins, goals or the catalogue change materially.
Is it possible that Performance Max is the wrong campaign type entirely? Yes, and it is worth considering. In the case study above, the eventual fix was moving a product category out of Performance Max and back into Standard Shopping for search term control, which produced a 66% ROAS improvement. Pairing PMax with Standard Shopping rather than replacing Shopping with PMax is an increasingly common approach.
About The Author
