High-SKU Google Ads Management
You cannot manually manage 50,000 products. You can manage the rules deciding what they deserve.
As ecommerce catalogues grow, Google Ads stops being primarily a campaign-management problem. It becomes a classification problem. Nobody is sensibly making a manual decision about 10,000 products. Or 100,000. Or several million SKUs.
The question becomes: how do you decide which products deserve investment, at scale? That is where JudeLuxe starts.
High-SKU PPC breaks traditional account management.
With a small catalogue, humans can still inspect a lot manually. With a large catalogue, that falls apart. You cannot realistically review every SKU individually, set bespoke targets for every product, audit every stock position by hand, adjust every product as margin changes, reclassify every lifecycle stage manually, and keep doing it every week.
The account needs rules. Good rules. Commercial rules. Rules that reflect what the business actually needs from the catalogue.
The problem is not too many products. The problem is treating too many products the same.
A catalogue with 100,000 SKUs does not represent one advertising opportunity. It represents thousands of different combinations of margin, demand, stock, sales velocity, customer value, returns, seasonality, lifecycle, competitive position, and strategic importance.
Yet large ecommerce accounts are often compressed into a handful of campaigns, a few targets, and one blended ROAS. That is where commercial information disappears.
Averages become more dangerous as catalogues grow.
Imagine an account with 100,000 SKUs. A small percentage generates most of the revenue. Another group consumes meaningful spend but contributes very little. Thousands barely receive impressions. Some are high-margin, some low-margin. Some are overstocked, some nearly out of stock. Some acquire exceptional customers. Some should probably never have been advertised.
Then all of that gets reduced to: 4.8 ROAS.
Useful? A little. Enough to decide where the next £1 should go? Not remotely.
High-SKU Google Ads is a capital-allocation problem.
The catalogue is competing for a finite pool of advertising capital. Every product is effectively asking: do I deserve investment? The answer depends on what the product can create commercially. That means understanding what it costs, what it earns, how much stock exists, how quickly it sells, what type of customer it acquires, what stage of its lifecycle it is in, and what the business needs from it right now.
The campaign structure is downstream of those decisions.
Every SKU has a job.
This is the BOI® principle. Each product gets one primary commercial role at a time.
Scale
There is profitable demand available and room to grow.
Profit
The product should maximise contribution.
Gateway
The product is particularly valuable for acquiring the right customers.
Protect
The product or demand has strategic importance.
Recovery
Inventory needs to return to cash.
None
The product does not currently justify advertising investment.
One SKU. One job. At 100 products, you could classify manually. At 100,000: you need systems.
The classification needs data.
A SKU job should not be assigned because someone had a hunch in a spreadsheet. We use combinations of available commercial signals.
Margin
What does the product actually contribute? A high-revenue product with weak margin may be less commercially valuable than a smaller product with strong contribution. Revenue concentration and profit concentration are not always the same thing.
Stock
How much inventory exists? 3 units? 300? 30,000? Google largely sees: in stock. The business sees completely different levels of commercial urgency.
Stock age
How long has capital been tied up? A product that arrived last week and a product sitting in the warehouse for nine months should not necessarily have the same objective. Inventory gets older even when the Google Ads structure doesn't.
Sales velocity
Stock quantity without velocity can mislead. 1,000 units selling 250 per week: healthy. 1,000 units selling 12 per week: different problem. Velocity helps distinguish scale opportunity, healthy core stock, and potential recovery problems.
Demand
Is there enough demand to justify aggressive investment? A product can have fantastic economics and still have very little search opportunity. Commercial attractiveness and market opportunity both matter.
Customer value
Some products disproportionately acquire customers who repeat, subscribe, cross-sell, and spend more later. That can justify weaker first-order economics. Those products may be Gateway SKUs. The first transaction does not always contain the whole value of the product.
Returns
A product generating strong gross sales but high returns may look considerably less attractive after the return window. This matters especially in fashion, footwear, luxury, and sizing-sensitive products. Revenue before returns can flatter product performance.
Lifecycle
Where is the product commercially? Launch. Growth. Core. Mature. End of line. Clearance. A SKU's job should change as it moves through those stages.
Seasonality
Some products have narrow windows of commercial opportunity. Christmas. Back to school. Football seasons. Summer. Events. Product launches. The correct advertising objective can change quickly as the window closes.
Strategic priority
Not every important product is important because of short-term margin. A business may need to prioritise a new range, a supplier relationship, a hero product, a strategic category, or a major launch. Not every useful signal fits neatly in Google Analytics. Commercial judgement still exists.
How we classify large catalogues
The precise model changes by business. But conceptually, the process is:
Ingest the product data
Advertising. Product. Commercial. Inventory. Customer.
Normalise the identifiers
This is less glamorous than strategy and absolutely critical. Product IDs need to reconcile across the ecommerce platform, Merchant Center, Google Ads, commercial datasets, and inventory systems. If the IDs do not match, the rest becomes expensive fiction.
Create commercially useful features
For example: margin band, stock cover, velocity band, customer value, lifecycle, demand tier, return-rate tier, commercial priority.
Apply classification logic
Products move into appropriate BOI® roles.
Translate roles into execution
Campaigns. PMax. Shopping. Feeds. Custom labels. Targets. Budgets. Exclusions.
Measure
Did the product perform against the commercial objective?
Reclassify
Because the inputs change. The system needs to move faster than the catalogue changes.
This is not static bucketing.
A lot of high-SKU strategies are essentially: high performers, medium performers, low performers. Then products remain there until somebody remembers to update the spreadsheet. That is useful segmentation. It is not enough.
A product can be a "low performer" for entirely rational reasons. Maybe it's a new launch. Maybe it acquires exceptional customers. Maybe it is being used to recover cash. Maybe it has low current volume but strong future potential. Likewise, a "high performer" can deserve less investment if stock is disappearing, margin collapsed, returns increased, or incremental demand is exhausted.
Historic performance is an input. Not the job description.
High-SKU management should be dynamic.
Imagine a product starts as Scale. Demand is strong, margin is healthy, stock is plentiful. Then stock cover falls sharply, and it moves to Profit. A few months later, a replacement model is announced, and remaining inventory becomes Recovery.
Same SKU. Different commercial stage. Different advertising objective. Static classifications eventually become wrong.
Spend concentration matters.
Large ecommerce accounts often show extreme concentration: a relatively small percentage of products consumes most of the spend, most of the revenue, and most of the conversions. That isn't automatically a problem. The best products should often receive more capital.
The question is: is the concentration commercially justified? Maybe Google's allocation is excellent. Maybe established products are crowding out high-potential products. Maybe high-revenue low-margin products dominate. Maybe profitable long-tail demand is being ignored.
The distribution needs explaining.
The long tail needs a strategy too.
A large catalogue can contain thousands of products with almost no data. You cannot optimise each one individually. But you also shouldn't automatically exclude them all, or throw them all into PMax and hope.
We may group products according to category, margin, demand, lifecycle, stock, price point, product type, or other commercially meaningful characteristics. Then create sensible learning environments.
Low data does not mean no strategy.
New products are particularly vulnerable.
A mature automated system naturally has more confidence in products with history. New products have no conversion history, limited click data, and weak model confidence, so they can struggle to receive meaningful exposure. This creates a self-reinforcing loop: no spend, no data. No data, no spend.
High-SKU accounts need deliberate new-product discovery. Not blind faith that every new SKU will automatically receive a fair trial.
We separate testing from scaling.
A new product needs to earn more investment. That usually means answering: is there demand, does it convert, what does it contribute, does it acquire valuable customers, is the offer competitive? Then scale the products that prove the case. Not every launch deserves equal budget indefinitely.
Product feeds become critical at scale.
The larger the catalogue, the less manually manageable the product data becomes. Feed quality affects eligibility, query matching, product understanding, segmentation, and commercial labelling. We use product types, attributes, GTINs, custom labels, supplemental data, rules, and automated classifications to create structure at scale.
High-SKU PPC without feed strategy is essentially campaign management with one eye closed.
Explore feed optimisationPerformance Max can be extremely useful at scale.
This is exactly the kind of problem automation is built for. Huge catalogue, large auction volume, constantly changing demand, far more signals than humans can process. But PMax still needs good product data, commercial classifications, clear objectives, strong conversion signals, and appropriate boundaries.
Automation solves scale. It does not solve commercial intent.
Explore Performance MaxStandard Shopping still has a role.
Especially where we need cleaner query evidence, product testing, greater control, transparent segmentation, or a reliable baseline. We don't decide that one campaign type "wins" for all high-SKU accounts. The commercial question decides the architecture.
High-SKU Search
The complexity isn't confined to Shopping. Large retailers may also have thousands of categories, brands, product searches, models, part numbers, competitor opportunities, and long-tail generic queries. Search architecture needs to avoid becoming thousands of campaigns nobody can maintain, or one giant campaign with no useful commercial separation. We use automation and structure where it creates better decisions. Not complexity for its own sake.
High-SKU fashion
Fashion is one of the clearest examples of why product count alone tells only half the story. A catalogue may contain 10,000 parent products and 100,000 variants. Then sizes sell out, colours behave differently, returns vary, seasonality moves, markdown approaches, new collections arrive. The product available to Google can be commercially very different from the product available to the customer.
Broken size runs are the classic case. Only XS remains. The product is still technically in stock, Google can keep advertising it, but most customers cannot buy their size. That reduces conversion, wastes clicks, increases returns to search, and damages efficiency.
Variant-level commercial context matters. Especially when catalogue scale makes manual checks impossible.
Multi-brand retail
High-SKU retailers often sell many brands with different economics. One brand may offer 55% margin, another 30%. One may have supplier funding, another may be price-controlled. One may acquire new customers, another may mainly serve existing demand.
Putting everything under one blended target can create enormous cross-subsidisation.
Brand is another commercial dimension. Not just a campaign naming convention.
Marketplace-style catalogues
Some ecommerce businesses have enormous catalogues but shallow stock. Others have huge depth in a smaller product group. The correct architecture differs. Catalogue size alone is not enough.
We want to understand SKU count, available SKU count, stock depth, revenue concentration, spend concentration, demand concentration, and margin variation. Complexity has several dimensions.
High-SKU international accounts
Now multiply the catalogue by countries. 100,000 SKUs, five markets, different prices, margins, demand, shipping, stock, and customer value. The same product can have different jobs in different countries. That means the true classification unit can become Product × Market. At scale. Humans are not manually managing that. Systems have to.
Explore international Google AdsReporting high-SKU accounts
A 100,000-SKU account should not produce a report containing 100,000 rows and a cheerful note saying "please see attached." We report at decision-relevant levels: BOI® job, category, margin band, stock state, brand, market, lifecycle, customer role. Then drill down where commercial consequence, anomaly, opportunity, or risk justifies it.
Reporting should compress complexity without hiding it.
What we monitor
Depending on the business. Not every metric deserves equal attention. We focus on the ones that change allocation.
Rules need exceptions.
Automated classification is powerful. It is also capable of being confidently wrong. A rule might classify a product as weak. But buying knows it launches on television next week. Or a major influencer is about to promote it. Or the supplier is funding the campaign.
Rules create scale. Judgement handles context the data does not contain.
Human overrides need governance too.
The opposite problem also exists. If every merchant can manually override the system because "we really like this product", the model becomes decorative. Overrides should have a reason, an owner, a timeframe, and a review point.
Human judgement should add information. Not nostalgia.
We don't optimise every SKU. We optimise the system deciding how SKUs are treated.
At high scale, the unit of strategic work becomes the classification logic, the data quality, the commercial rules, the exceptions, and the feedback loop. Rather than manually clicking through product after product. Better rules create better allocation at scale.
An example
Imagine a retailer with 100,000 active SKUs and £500,000 monthly Google Ads spend. At account level, ROAS is healthy. But analysis shows:
£310,000
Established products with proven economics
Reasonable.
£80,000
High-value acquisition products
Also reasonable.
£45,000
Recovering ageing stock
Intentional.
£35,000
Strategic product launches
Useful.
£30,000
Weak margin, low demand, little stock, no strategic role
The problem.
The account does not necessarily need a new bid strategy. It needs £30,000 reallocated. That is the difference.
Now scale the consequence. A 1% improvement in allocation on £50,000 monthly spend is £500. On £500,000 it is £5,000. Every month. Before considering the commercial output that improved allocation might create.
The cost of classification errors grows with the account. That is why high-SKU complexity matters.
What we don't do
Scale should improve decision-making, not obscure it.
Who high-SKU Google Ads management is for
This approach becomes especially valuable when:
Frequently asked questions
What counts as a high-SKU Google Ads account?
There's no hard cutoff, but the work materially changes once manual review becomes impossible. Below a few hundred SKUs, humans can still classify products by hand. Above that, classification, feeds, and rules become the actual work. The principles on this page apply from thousands of SKUs upwards.
Why not just put the whole catalogue into Performance Max?
PMax is extremely useful at scale, and huge catalogues are exactly the kind of problem automation is built for. But PMax still needs good product data, commercial classifications, clear objectives, strong conversion signals, and appropriate boundaries. Automation solves scale. It does not solve commercial intent.
How often do product jobs change?
There is no universal schedule. The cadence depends on stock volatility, catalogue size, seasonality, margin changes, promotions, lifecycle, and demand. What matters is that the classification is refreshed quickly enough to remain commercially useful. A dynamic framework does not mean randomly moving products every Tuesday. It means changing decisions when the evidence changes.
Do you optimise every SKU individually?
No. We optimise the system deciding how SKUs are treated. At high scale, the unit of strategic work is the classification logic, the data quality, the commercial rules, the exceptions, and the feedback loop. Better rules create better allocation at scale.
How do new products get a fair chance?
Deliberately. A mature automated system has more confidence in products with history, so new products can get trapped in a loop of no spend, no data. We separate testing from scaling: a new product answers whether there is demand, whether it converts, what it contributes, and whether it acquires valuable customers. Then we scale the products that prove the case.
What's the minimum spend for high-SKU PPC management?
£15k+/month on Google Ads is typical for catalogues at this scale. The economics of classification infrastructure only pay back when the allocation decisions are large enough to matter.
The high-SKU question.
Not "how do we optimise 100,000 products?" You don't. Not individually. The better question is: what rules should decide what 100,000 products deserve?
High-SKU PPC is a classification problem before it is a campaign problem. Every SKU has a job. Every £1 needs a reason.