Managing Fashion Catalog Campaigns at Scale: 5,000 SKUs Across Meta & Google

Key Takeaways

• At thousands of SKUs the catalogue is the campaign — you steer products through feed attributes, custom labels and value signals, not by managing them one at a time.
• Feed hygiene is the foundation and the hidden ceiling: titles, GTINs, images, product types, price and availability decide whether anything above them can work, and at 5,000 SKUs feed problems are constant unless actively managed.
• Custom labels are the control layer — tag every SKU by margin, velocity, price band, season, lifecycle and return rate so each tier gets the budget and bidding it deserves.
• Segment by profit and velocity, not category alone: hero products carry spend, steady sellers run efficiently, new lines get controlled learning, and dead stock is starved or liquidated — never funded at the hero bar.
• Structure Meta Advantage+ product sets and Google Shopping/PMax around those tiers, and use exclusions and priority so PMax and Advantage+ don't quietly spend on low-value products.
• Feed value signals and account for returns — a fashion catalogue's blended ROAS hides which products actually make money after markdowns and returns, so optimise to contribution, not platform ROAS.
• Run it as a weekly and seasonal operating rhythm: re-rank velocity, refresh labels, monitor the feed, promote and prune products, and rebuild segmentation each season, because the catalogue is alive.

At 5,000 SKUs, the Catalogue Is the Campaign

The instinct that serves you well at fifty products actively harms you at five thousand, and recognising that is the first and most important shift. At a small catalogue you can look at individual products, write specific ads, set specific bids, and manage the account by hand. At 5,000 SKUs that approach is not just impractical, it is counterproductive: you will spend all your time on the loudest few products, the automation will be under-fed and mis-directed, and the long tail — which in fashion is most of your catalogue and often most of your opportunity — will be ignored. The single most common reason large fashion catalog accounts underperform is that they are being managed as if they were small ones, product by product, by a team drowning in the volume.

At this scale the unit you manage is no longer the individual product; it is the catalogue as a data asset and the segments you carve from it. The platforms' automation — Meta Advantage+ Catalog (formerly dynamic product ads), Google Shopping and Performance Max — becomes the engine that does the per-product serving, bidding and creative assembly at a scale no human team could match. Your job shifts from operating the machine by hand to engineering the inputs that tell the machine what to do: the feed that lets it understand every product, the labels that let you group products by what matters, and the value signals that tell it which outcomes are worth pursuing. Get those inputs right and the automation is astonishingly effective across thousands of products; get them wrong and no amount of manual bid-tweaking rescues it, because you are trying to hand-steer a system built to be steered by data.

This reframing organises everything that follows into three disciplines, in strict order of dependency. First, feed hygiene: make the product data clean enough that the platforms can understand, approve and serve every SKU — nothing works above a broken feed. Second, segmentation: tag every SKU with custom labels so you can group them by the dimensions that actually determine how they should be advertised — margin and velocity above all. Third, structure and signals: build campaigns around those segments rather than your category tree, and feed the automation value signals so it optimises toward contribution rather than the cheapest conversions. Master those three, in that order, and 5,000 SKUs stops being a burden and becomes a moat — a large, well-organised catalogue that the platforms can monetise far more profitably than a competitor's unmanaged one. The rest of this guide is each discipline in depth.

Feed Hygiene: The Foundation and the Hidden Ceiling

Nothing downstream works if the feed is broken, and at 5,000 SKUs feed problems are not a risk but a certainty unless actively managed — so this is where every serious catalog operation starts and returns, constantly. The feed is the structured product data the platforms ingest, and its quality is the invisible ceiling on everything above it. A product with a weak title, a missing or invalid GTIN, a low-quality or non-compliant image, a wrong or stale price, or an incorrect availability status will be under-served, disapproved, or shown for the wrong queries no matter how much budget, clever structure, or bidding sophistication you layer on top. You cannot bid your way out of a bad feed, and you cannot creative your way out of it either. Fix the feed first, or everything else is built on sand.

Consider the load-bearing attributes one by one, because in fashion each carries specific weight. Titles are the single biggest lever for Google Shopping relevance and matter enormously on Meta too — they should front-load the terms shoppers actually search, in a sensible order for the category (for apparel, typically brand, then product type, then key attributes like colour, material and gender, then a distinguishing detail), written for the algorithm's matching and the human's scan at once, not stuffed. Product identifiers — GTINs, MPNs, brand — determine eligibility and how well products are matched to queries; missing or invalid GTINs quietly suppress otherwise good products. Images must be high quality and policy-compliant, and in fashion the primary image choice (flat-lay versus on-model, for instance) materially affects click-through, so it is a performance decision, not just a compliance one.

Then the structural attributes: product_type (your own taxonomy) and google_product_category (Google's) must be correct, because they drive both relevance and — critically for this playbook — your ability to segment. Price and availability must be accurate and near-real-time, or you will pay to advertise out-of-stock or mispriced items, which in a fast-moving fashion catalogue with constant restocks and sell-throughs is a continuous, silent budget leak. And fashion-specific richness matters more here than almost anywhere: variant handling (size and colour as proper variants, not duplicate products), and rich attributes like colour, material, pattern, occasion, gender and age group, all drive both eligibility and the filtered, high-intent placements where fashion actually converts. A shopper filtering for 'women's black leather ankle boots size 6' is a high-intent buyer, and you only reach them if your attributes are complete and accurate.

The operational truth is that at 5,000 SKUs feed hygiene is not a one-time setup but a continuous process, which is exactly why feed-management platforms exist and why disciplined teams monitor the feed daily. Products get disapproved, images get flagged, prices drift out of sync, items sell out and come back — a feed that was clean last week is broken this week, and a silent feed problem is a budget leak spread across thousands of products at once, invisible until you go looking. Treat feed hygiene as a monitored, alerted, owned operation with eyes on disapprovals, stockouts, price mismatches and attribute coverage, whether you run it in a platform or in-house. The teams that win at catalog scale are, more than anything else, the teams that never let the feed rot.

Custom Labels: The Control Layer That Runs 5,000 Products

With a clean feed in place, custom labels are how you actually control thousands of products without touching them individually — and they are, in my experience, the single most under-used lever in large catalog accounts. Google gives you five custom label slots (custom_label_0 through custom_label_4); Meta lets you build product sets from any feed attribute. Both let you tag every SKU with the dimensions that determine how it should be advertised, and then group, budget and bid by those tags. The mistake almost everyone makes is to segment by category — 'dresses', 'shoes', 'accessories' — which is how your website is organised but has almost nothing to do with how products should be advertised. Category tells you what a product is; it does not tell you whether it makes money or whether it is selling. Those are the two things that should drive budget, and custom labels are how you encode them.

The labels that matter most in fashion, in rough order of importance. Margin band (high / medium / low) is first, because it lets you fund by profit rather than revenue — a rule that will change your results more than any bid tweak. Velocity or sales rank (bestseller / steady / slow / dead) is second, because momentum is the other axis that should drive spend; a fast-selling product deserves fuel, a dead one deserves starvation. Price band matters because it affects both conversion behaviour and the margin maths. Lifecycle (new arrival / core / end-of-season / clearance) matters because a product's stage dictates its strategy — new arrivals need learning budget and patience, core products need steady efficient spend, clearance needs liquidation logic. Season and collection let you pace budget into and out of ranges. And where you have the data, return rate deserves its own label, because in fashion a high-return SKU has a far lower true contribution than its order value suggests, and it should be advertised more cautiously than its ROAS alone would imply.

With these labels in place, you can express your entire commercial strategy as a small set of rules applied across thousands of products at once: fund high-margin bestsellers aggressively; run core steady sellers to a contribution-aware target; give new arrivals a capped learning budget and a defined ramp; pace seasonal lines in and out; and starve, exclude or liquidate low-margin dead stock. This is the difference between managing a catalogue and being managed by it. Without labels, every product competes for budget on blended, last-click ROAS, which quietly over-funds cheap-but-unprofitable fast movers and starves your real profit drivers — the automation optimises for the metric you gave it, and blended ROAS is the wrong metric. With labels and value signals, budget follows margin and momentum automatically, which is exactly what you want a powerful automation to be optimising toward. Building and maintaining these labels — deriving them from your own margin, sales and returns data and pushing them into the feed — is the highest-leverage ongoing work in a large catalog account, and it sits at the heart of a disciplined D2C performance practice.

Structuring Google: Shopping, Performance Max and Priority

Now structure the campaigns around your segments rather than around your website's category tree, starting with Google. The core principle is that products with different commercial roles deserve different campaigns, budgets and targets, and custom labels are how you separate them. Your hero and high-margin products belong in their own campaign — whether standard Shopping or Performance Max — with the budget and target that reflect their value, so they are never starved by being pooled with the long tail. Your steady mid-tier runs in efficient catalogue campaigns to a contribution-aware target. Your new and seasonal lines get a controlled learning campaign, deliberately separated so they are not smothered by established bestsellers that will always out-compete them for budget in a shared campaign. And clearance and dead stock are either excluded entirely or capped at a liquidation target where the goal is to clear inventory, not to hit your normal efficiency bar.

Performance Max deserves specific attention, because it is powerful and, left unconstrained, dangerous for a large catalogue. PMax will happily spend your budget on whatever converts most cheaply, which frequently means it drifts toward low-value, low-margin products and branded search it would have won anyway. The controls that matter: feed your segmentation through custom labels and organise asset groups by product tier so PMax's optimisation has structure; use listing-group and product exclusions to keep it off dead stock and low-margin items; and, where your account allows, separate your highest-value products into their own PMax or standard Shopping campaign so their budget is protected from the pull toward cheap conversions. The recurring lesson is that PMax's automation is only as good as the constraints and signals you give it — an unconstrained PMax on a 5,000-SKU fashion feed will quietly optimise toward the products you least want to scale, because those are the cheapest to sell, and cheap-to-sell and profitable-to-sell are very different things in fashion.

Standard Shopping still has a role even in a PMax world, precisely because it gives you more granular control — campaign priority settings, negative keywords, and clearer visibility into which products and queries are driving spend. Many sophisticated fashion accounts run a considered mix: PMax for broad, automated reach across the catalogue with tight exclusions, and standard Shopping (or the newer controls Google exposes) for the hero products and the queries where control and visibility matter most. Whatever the exact structure, the principle is invariant: the automation should be pointed at your profit drivers and fenced off from your dead stock, and that pointing-and-fencing is done through feed segmentation, campaign structure and exclusions, not through manual per-product bidding, which does not scale to thousands of products.

Structuring Meta: Advantage+ Catalog and Product Sets

On Meta, the same segmentation logic expresses itself through product sets. Advantage+ Catalog campaigns (the dynamic catalogue campaigns formerly known as DPA) serve products from your catalogue automatically to the people most likely to convert, and the lever you control is which products are in play and what you optimise them toward — which is defined by the product sets you build. Build product sets from the same attributes and custom labels you use on Google: a hero/bestseller set, a high-margin set, a new-arrivals set, a seasonal-collection set, a clearance set. Each becomes something you can target, budget, cap and creative-test independently, so your best products get their own well-funded campaigns and your clearance runs on its own liquidation logic rather than competing with everything else on blended performance.

Meta's automation, like Google's, is powerful and will push toward volume and cheap conversions unless you shape it. The two most important shaping levers are the product-set definitions (which control which products are eligible in each campaign) and the optimisation and value signals (which control what 'good' means to the algorithm). If you optimise a broad Advantage+ Catalog campaign to raw conversions or purchase value with no margin awareness, it will find the cheapest conversions — often your lowest-margin, highest-return products — and scale them, producing a great-looking ROAS and a shrinking contribution. If instead you segment into margin- and velocity-aware product sets and feed value signals that reflect contribution rather than order value, the same automation scales the products that actually make money. The catalogue campaign is only as smart as the sets and signals you give it.

Creative matters more on Meta than on Google, even in catalogue campaigns, because Meta is a discovery-and-demand-creation environment where the ad's stopping power drives performance. At catalogue scale you cannot make bespoke creative for 5,000 products, but you can and should invest in the catalogue creative layers Meta provides — dynamic formats, overlays (price, discount badges, 'new'), and collection and video treatments — and test them by product set. A well-chosen dynamic overlay on your bestseller and clearance sets, tested properly, can lift performance across thousands of products at once, which is exactly the kind of leverage catalogue scale rewards: one creative decision applied across a whole segment, rather than one ad per product. The unifying principle across Meta and Google is the same — you are not building thousands of ads, you are building a handful of well-defined, well-signalled product tiers and letting dynamic, catalogue-driven automation serve the right product to the right person, pointed at contribution.

Value Signals, Returns and the ROAS Trap

The most important and most neglected discipline in scaled fashion catalog advertising is what you optimise the automation toward, because both Meta and Google will faithfully optimise toward whatever signal you feed them — and if that signal is raw conversions or unadjusted order value, they will scale the wrong products. The fix is value-based bidding fed with the right values: instead of telling the platforms that every conversion is worth the same, or that a product's value is its order price, feed them a value signal that reflects true contribution — order value adjusted for margin and, crucially in fashion, for expected returns. When the automation knows that a high-return, low-margin product is worth far less than its order value suggests, it stops over-scaling it, and when it knows a high-margin, low-return product is worth more, it leans in. This single change — moving from conversion-count or unadjusted-value optimisation to contribution-aware value signals — is often the largest profitability improvement available in a mature catalog account, and it is invisible to anyone still staring at ROAS.

Returns are the specific fashion trap that makes the ROAS metric actively misleading, so they deserve their own emphasis. A blended ROAS treats a sale as a sale, but in fashion a meaningful share of sales come back, and the return carries real cost — reverse shipping, processing, often a markdown or write-off on the returned item. A product with a stellar ROAS and a high return rate can have a genuinely poor true contribution, while a product with a modest ROAS and near-zero returns can be your most profitable. If you optimise to blended ROAS, the automation will scale the high-ROAS, high-return product and starve the modest-ROAS, low-return one — exactly backwards. The only defence is to bring return data into your value signals and your reporting, so that the number the automation optimises toward, and the number you make decisions on, is contribution after returns, not ROAS. Wherever it is technically possible, feed returns-adjusted values; where it is not yet, at minimum monitor return rates by product tier and manually adjust budgets so you are not unknowingly funding your least profitable inventory.

The broader point is that ROAS is not a profitability metric and should never be the objective in a fashion catalogue, however comfortable and universal it is. ROAS is revenue over spend, and revenue is not profit — after COGS, shipping, fees, discounts and returns, two products with identical ROAS can have wildly different contribution. Every decision in a scaled fashion account should be made on contribution, and every signal fed to the automation should reflect contribution as closely as your data allows. This is harder than reporting ROAS, which is why most accounts do not do it, which is precisely why doing it is a durable advantage. The account that optimises to contribution while its competitors optimise to ROAS will, over a season, scale the profitable products while its competitors scale the unprofitable ones — and in a thin-margin, high-return category, that difference compounds into the gap between a brand that grows profitably and one that grows itself into trouble.

The Weekly and Seasonal Operating Rhythm

Running this well is an operating rhythm, not a launch, because a fashion catalogue is alive — products sell through, restock, go on sale, get discontinued, and change their commercial role constantly. Weekly, the disciplined operator re-ranks products by velocity and refreshes the custom labels, because yesterday's bestseller is next month's clearance and the labels must track that reality or your budget will keep flowing to products that no longer deserve it. They monitor the feed for disapprovals, stockouts and price errors, because a week's drift is a week's silent leak. They promote products between tiers (a new arrival that is now a proven bestseller moves into the hero set; a former hero that has slowed moves down) and prune the dead. And they read performance by segment and by contribution — 'how is my high-margin bestseller tier performing on contribution?' — rather than staring at 5,000 individual products, because the segment is the unit of decision, not the SKU.

Seasonally, the rhythm is larger: you rebuild the segmentation around the new collection, pace budget into the launch of new ranges and out of the end-of-season lines, adjust for the return-rate and discount patterns of the season, and reset your value signals for the new margin structure. Fashion's seasonality means a structure that was optimal in autumn is wrong for the spring sale, and the operators who win are the ones who rebuild deliberately each season rather than letting last season's structure ossify. This is also when you plan the clearance and liquidation logic for the outgoing season — deciding which end-of-season products to run at a liquidation target to clear inventory versus which to simply exclude — because clearing dead stock efficiently is itself a meaningful profit lever that an unmanaged catalogue leaves on the table.

The mistakes that sink scaled fashion catalogs are consistent and worth naming so you can watch for them: managing products individually instead of by segment; leaving feed hygiene to chance and discovering the leak only after weeks of wasted spend; optimising to blended ROAS and unknowingly funding dead stock and high-return items; letting Performance Max or Advantage+ run unconstrained so budget drifts to the cheapest conversions rather than the most profitable products; and letting the segmentation ossify while the catalogue moves on beneath it. Avoid those, get the feed and the labels and the value signals right, and 5,000 SKUs becomes exactly what it should be — not a burden to be survived but an advantage to be exploited. A large, clean, well-segmented, contribution-optimised catalogue is something the platforms' automation can monetise far more profitably than a competitor's unmanaged pile of products, and at scale that operational discipline is a genuine moat: the hardest thing for a competitor to copy is not your media buying but the disciplined data and structure underneath it.

A Practical Build Sequence for a 5,000-SKU Account

If you are inheriting or building a large fashion catalog account, the sequence matters as much as the components, because doing them out of order wastes time and money. Start with an audit of feed health — coverage, disapprovals, attribute completeness, image quality, price and availability accuracy — and fix it before you touch campaign structure, because optimising campaigns on a broken feed is optimising toward the wrong destination faster. In parallel, get your commercial data in order: pull margin by product, sales velocity and rank, and return rates where available, because these are the inputs to your labels and value signals, and without them you are segmenting blind.

Next, build the custom-label taxonomy from that commercial data and push it into the feed — margin band, velocity, price band, lifecycle, season, return-rate flag. Then restructure campaigns around the tiers those labels define, separating heroes and high-margin products from the long tail, isolating new and seasonal lines, and fencing off or capping clearance and dead stock. Then implement value-based bidding fed with contribution-aware, returns-adjusted values wherever technically possible, so the automation is optimising toward profit rather than conversions or unadjusted revenue. Only once all of that is in place do you scale spend, and you scale it tier by tier, watching contribution rather than ROAS, expanding the segments that prove profitable and holding back the ones that do not.

Finally, stand up the operating rhythm — the weekly label refresh, feed monitoring, tier promotion and pruning, and segment-level contribution reporting, plus the seasonal rebuild — because without it the whole structure decays within a couple of months as the catalogue moves and the labels go stale. The build is front-loaded and unglamorous: feed, data, labels, structure, signals, then scale. But it is the build that turns a chaotic 5,000-SKU account that leaks budget into dead stock and high-return items into a disciplined machine that scales your profit drivers and starves your losers automatically. In catalog advertising at scale, the account that wins is not the one with the cleverest bidding tactics; it is the one with the cleanest feed, the sharpest segmentation, the most honest value signals, and the discipline to keep all three current as the catalogue lives and breathes.

Sources, Scope and a Note on Fairness

A note on how to read this. The platform capabilities described here — Meta Advantage+ Catalog and product sets, Google Shopping, Performance Max, custom labels, value-based bidding — are current at the time of writing, but the ad platforms change their products, names and controls frequently, so verify the specific mechanics against Meta's and Google's own current documentation before you build. Custom-label counts, feed specifications and PMax controls in particular evolve, and a playbook is only as current as its last check against primary sources. The strategy in this guide — segment by margin and velocity, structure around tiers, optimise to contribution after returns — is durable; the exact buttons are not, so treat the tactics as illustrative of the principle rather than as a permanent instruction set.

This guide is published by Fluxsy and reflects our own operating point of view, not an independent standard. We have avoided inventing benchmark numbers, results or 'typical' figures, because a fabricated number in a technical playbook is worse than none — it misleads exactly the practitioners who would act on it. Where we reference how the platforms behave (for example that unconstrained automation tends toward cheap conversions), that reflects widely-observed practitioner experience and the platforms' own stated optimisation logic, but your account's behaviour depends on your data, your margins and your configuration, so measure your own before you conclude. The durable value here is the framework — feed, labels, structure, signals, rhythm — which holds regardless of which platform features exist next quarter.