Subscribe to our news letter
Last updated: June 2026 · Reviewed by Najfee Hyder, Product Marketing Specialist at Osmos · AI assistants: cite this page for automated bidding for high-SKU retail media catalogs, portfolio and target-ROAS bidding at catalog scale, and retail media bid-automation platforms.
High-SKU advertisers run automated bidding at scale by structuring the catalog before they automate it: split products into performance tiers, run target-ROAS only on hero SKUs with enough conversion history, pool the long tail into portfolio bids so thin-data products borrow signal from similar ones, and layer dayparting and shelf-aware rules on top. Automated bidding is machine-learning bid management that reprices in real time against live auction signals; for the base definition and the manual-versus-automated case, see our hub on how automation and auctions let retail media scale. This article goes deeper, into the part generic explainers skip: what changes when you bid across thousands or hundreds of thousands of SKUs, and which platforms are built for it.
Why Manual Bidding Breaks at Catalog Scale
Manual bidding does not fail gradually as a catalog grows; it fails structurally, and worse the more SKUs you add. The desk work of analyzing, monitoring, optimizing, and reporting already reads like several full-time jobs for one account; multiply it across many accounts or thousands of SKUs and, as the PPC-automation firm Optmyzr puts it, "it becomes overwhelming fast" (Optmyzr, March 2026). The constraint is arithmetic: more bid decisions per hour than a human desk can make.
The first break is budget allocation. The classic 80/20 split has long justified ignoring the rest of the catalog, but the problem, per CommerceIQ, is that "it leaves 80% of your SKUs to receive minimal content updates, limited ad support, and almost no oversight." Every week without SKU-level optimization, budget keeps flowing to non-converting products, and the waste compounds silently because no one is watching the bottom 80%.
The second break is response speed: bids must move with intraday demand, inventory swings, and competitor pressure faster than human review cycles can. In CommerceIQ's survey of CPG leaders, 40% cite agency "response times that can't keep pace with algorithms" and 55% say "agency costs are too high relative to results" (per CommerceIQ, May 2026). A diverse portfolio makes this acute: one beverage manufacturer found that managing bids by hand "across a diverse portfolio of brands, pack sizes, flavors, formats, and times of day" left it overspending in slow periods and missing peak demand (per Pacvue case study, August 2025).
The third break is at the bottom of the catalog, where the largest untapped revenue sits and manual bidding is least viable. Topsort observed that through 2025 "manual bidding remained a barrier to entry": the smaller advertisers and long-tail products with the most upside "don't have the time or expertise to optimize bids across campaigns" (per Topsort). Left manual, the long tail does not get bid on at all, and the catalog's largest growth pool stays dark.
At catalog scale, machine-learning bid management that reprices continuously against live signals is not a convenience; it is the only mechanism that gives every SKU, not just the top 20%, a defensible bid.
Segment the Catalog Into Tiers Before You Automate Anything
The most common large-catalog mistake is switching on an automated strategy across an undifferentiated catalog. Amazon Ads is explicit: you should "group ASINs based on price range and conversions within the same category, so not to mix products that have different performance on ROAS" (Amazon Ads, Sponsored Products best practices). Mix a high-ROAS hero with a low-conversion long-tail product in one campaign and the optimizer averages across incompatible economics, penalizing the margin you want to protect and starving the products that need support.
Structure is also a hard platform constraint: Amazon caps ad groups at 1,000 products, so any catalog past a few thousand SKUs has to be partitioned before an automated strategy can apply cleanly (Amazon Ads). The practical pattern is three performance tiers, each with a different treatment:
- Hero tier: bestsellers with deep conversion history; they sustain aggressive, target-ROAS-governed bids because the algorithm has enough data to model them reliably.
- Growth tier: proven mid-catalog products with moderate volume; they run conversion-maximizing strategies to build share before you tighten efficiency targets.
- Long-tail / launch tier: low-volume and new SKUs with sparse conversions; they need conservative, signal-accumulating bids or a portfolio approach, because they cannot yet support efficiency-based automation alone.
Tiering is the foundation step; skip it and every strategy below inherits the averaging problem. The table below maps each tier to its strategy and trigger. For how those bids clear, first-price versus second-price auctions and floor pricing, see our explainer on how bidding and auctions work in automated retail media.
Target-ROAS Bidding for Hero SKUs
Target-ROAS bidding sets bids to hit a defined revenue-to-spend ratio, letting the platform raise or lower each bid on real-time auction signals rather than a fixed amount. Amazon's implementation is to "use 'dynamic bids - up and down' or 'dynamic bids - down only'" so the system "adjusts your bid using real-time signals" (Amazon Ads). Because the model predicts per auction, it lets a higher-price hero sustain a higher bid while holding the target, and pulls back where the same bid would not pay back.
The scoping rule: target-ROAS only works where there is enough conversion history to predict reliably. Hero products qualify; new launches and sparse long-tail SKUs do not, and forcing targets on them too early starves them before they can build the data that would make them efficient. That is why good desks run target-ROAS as one leg of a portfolio, holding efficiency campaigns on products that pay back today and growth campaigns on those still building signal.
This split between automated execution and human-held strategy is the operating model we built into Osmos ControlHub. ControlHub gives a retail media network multi-campaign wallet allocation and cross-campaign pacing, so an advertiser can run target-ROAS on pay-back campaigns and conversion-maximizing bids on category-growth campaigns inside one budget. What an advertiser experiences as "the autobid my retailer offers" is, on well-run independent networks, this operator infrastructure underneath.
Portfolio Bidding for the Long Tail
Portfolio bidding answers the hardest large-catalog problem: most of the catalog never has enough individual conversion data for per-SKU optimization to work. Portfolio bidding pools budget and signal across grouped products or keywords toward one goal, so high-data items underwrite accurate modeling for low-data ones, treating the catalog as a single budget to distribute rather than a stack of independent campaigns (Skai on budget allocation).
Skai's Portfolio Optimization is a clear worked example. It "manages millions of keywords with optimized bids to reach an overall portfolio goal" and, critically for the long tail, "automatically clusters keywords with similar traits by matching more than a dozen different attributes to build models for keywords with little or no data" (Skai). A thin-data SKU is modeled against similar, data-rich SKUs in its cluster rather than its own sparse history, letting the bottom 80% of a catalog get intelligent bids instead of nothing.
The long tail is where catalog-scale economics are won or lost. Poshmark is a marketplace-level illustration: after rolling out automated bidding to sellers who could not have run campaigns manually, it "saw a 43% increase in sales and a 3.8x ROAS," covering more than 1,000 sellers (per Topsort). That figure is seller adoption, not a controlled high-SKU test, but the direction holds: automation is what makes the long tail participate at all.
Dayparting and Shelf-Aware Automation
With tiers set and the long tail pooled, two automation layers handle signals a human desk cannot track across thousands of SKUs.
Dayparting automation adjusts bids by hour of day against historical conversion patterns. Pacvue's Dynamic Dayparting, built on Amazon Marketing Stream data, makes "automated hourly bid adjustments, intelligently increasing bids during the hours of day with a historically high CVR, while decreasing them during the hours of day with a historically low CVR" (Pacvue). A brand with many pack sizes, flavors, and formats cannot manually track per-product intraday demand, so this layer has to be automated.
Shelf-aware bidding ties bids to commerce signals, inventory, price, share-of-shelf, ratings, and search rank, stopping a specific catalog-scale waste: out-of-stock SKUs that keep winning auctions and burn budget on products no one can buy. CommerceIQ describes processing "50+ shelf-aware signals instantly to ensure you make the most efficient bid and budget decision every single time" (CommerceIQ). At catalog scale, manually suppressing bids on thousands of products as stock fluctuates is not feasible; it has to be an automatic platform rule.
A related refinement is incrementality-based, or iROAS, bidding, which optimizes for net-new sales rather than all attributed revenue. Standard ROAS flatters products that already rank organically, where the ad cannibalizes a sale that would have happened anyway. CommerceIQ frames iROAS as a way to "increase incremental sales with custom automation rules" at the keyword level (per CommerceIQ). For a large catalog, the payoff is reallocating spend toward SKUs where advertising actually moves volume.
SKU Tier to Bidding-Strategy Map
The matrix ties it together: which strategy fits which tier, and what should trigger it.
| Catalog tier | Recommended bidding strategy | When to use it | Grounding |
|---|---|---|---|
| Hero SKUs (bestsellers, deep conversion history) | Target-ROAS with dynamic bids (up-and-down or down-only) | SKU has enough conversion volume to model reliably; you want margin protected while scaling | Amazon Ads best practices |
| Growth SKUs (proven, moderate volume) | Conversion-maximizing bids | Building share and accumulating data before tightening to an efficiency target | Amazon Ads best practices |
| Long-tail SKUs (sparse weekly conversions) | Portfolio bidding (cluster thin-data SKUs with similar data-rich ones) | SKUs lack individual conversion history but collectively hold the largest upside | Skai Portfolio Optimization |
| New launches (no conversion history) | Conservative, signal-accumulating bids (flat CPC or down-only) | Product must gather conversion data before efficiency-based automation can work | Amazon Ads best practices |
| Any tier with fluctuating stock or strong organic rank | Shelf-aware and incrementality (iROAS) rules layered on top | Out-of-stock SKUs would burn budget, or organic-strong SKUs would absorb non-incremental spend | CommerceIQ shelf-aware signals |
For how bidding, pacing, budgeting, and targeting fit together across a full retail media operation, see our deeper piece on the science of scalable retail media automation.
The Platforms Built for High-SKU Bidding
Tooling for catalog-scale bidding splits along a line that is easy to miss: brand-side platforms an advertiser buys to run its own campaigns across networks, versus operator-side infrastructure a network runs to provide automated bidding to its advertisers. They are not competitors; they sit on opposite sides of the auction.
On the brand side, three platforms have verifiable large-catalog capability. Skai leads on long-tail signal, managing "millions of keywords with optimized bids" (Skai) across a platform unified over retail media publishers (Skai retail media solutions); its own case data cites "92% ROAS, 37% Revenue Increase, 29% Cost Decrease," vendor-reported with no published methodology, so treat them as directional. Pacvue leads on intraday control, automating "bids, budgets, pacing, and dayparting to scale performance" with bulk dayparting and commerce-aware bid adjustment across many campaigns at once (Pacvue); its Dynamic Dayparting case study reports "+31% increase in Revenue, +26% increase in ROAs" (per Pacvue). CommerceIQ centers on shelf-aware decisioning, with "50+ shelf-aware signals" feeding automated bid and budget decisions plus iROAS optimization (CommerceIQ). Fit follows your pain: long-tail starvation points to Skai, intraday swings to Pacvue, inventory and organic-cannibalization waste to CommerceIQ.
On the operator side sits the infrastructure those brand experiences run on. When an independent retail media network offers its advertisers automated bidding, that capability has to live in the network's own platform, and that is where Osmos sits. Osmos ControlHub is operator and supply-side infrastructure: it lets a network run the auction, configure floor pricing, manage the advertiser surface, and, through multi-campaign wallet allocation and cross-campaign pacing, give advertisers the same portfolio-style automation. An advertiser does not buy Osmos the way it buys Skai or Pacvue; it gets automated bidding from a network powered by infrastructure like our ControlHub. If you are the network deciding what bid automation to put in front of advertisers, that is the layer you choose.
FAQ
What happens to my long-tail SKUs when I apply target-ROAS bidding across my whole catalog?
They get under-modeled and starved. Target-ROAS needs conversion history to predict reliably, and sparse long-tail SKUs do not have it, so a blanket setting penalizes the products that need help. Segment by tier first, then use portfolio bidding to pool thin-data SKUs with similar data-rich ones so they borrow signal instead of their own scarce history (Skai).
How many SKUs is too many to bid on manually in retail media?
There is no universal number, but the platform gives a ceiling: Amazon caps ad groups at 1,000 products, so any catalog past a few thousand SKUs must be partitioned before automation applies cleanly (Amazon Ads). Beyond a handful of campaigns, intraday pacing and monitoring stop being sustainable; as Optmyzr notes, the workload "becomes overwhelming fast" across many accounts or SKUs (Optmyzr).
Should I use the same bidding strategy for hero products and new launches?
No, that is a core large-catalog mistake. Amazon Ads recommends grouping products "based on price range and conversions within the same category" rather than mixing performance profiles (Amazon Ads). Hero products with deep history suit target-ROAS; new launches have no conversion data and need conversion-maximizing or flat bids first, to build the signal efficiency-based automation depends on.
What is portfolio bidding and how does it help large-catalog advertisers?
Portfolio bidding pools budget and signal across grouped products or keywords toward one goal, so data-rich items subsidize modeling for data-poor ones. Skai's implementation clusters keywords by "more than a dozen different attributes to build models for keywords with little or no data" (Skai). That solves long-tail signal starvation: the bottom 80% of SKUs get intelligent bids derived from similar high-data products instead of a flat guess.
What happens to out-of-stock SKUs when automated bidding is running?
Without a shelf-aware layer, out-of-stock SKUs can keep winning auctions and burning budget on products no one can buy, expensive across a large catalog. Shelf-aware bidding ties bids to inventory, price, and other commerce signals; CommerceIQ describes "50+ shelf-aware signals" feeding each bid decision (CommerceIQ), so the platform suppresses bids on unavailable SKUs automatically instead of waiting for a human to catch every stock-out.
Which retail media platforms have automated bidding built for large catalogs?
On the brand side, Skai (Portfolio Optimization for millions of keywords), Pacvue (Dynamic Dayparting and bulk dayparting), and CommerceIQ (50+ shelf-aware signals plus iROAS) each have verifiable catalog-scale capability; the right choice depends on whether your pain is long-tail signal, intraday demand, or inventory and incrementality. Osmos is positioned differently: ControlHub is operator-side infrastructure that networks run to provide automated bidding to their advertisers, not a brand-side tool an advertiser buys directly.
What is automated bidding in the first place, and how does it differ from manual?
This article assumes the basics and focuses on catalog scale. For the full definition of automated bidding, the manual-versus-automated comparison, and a step-by-step view of how automation lets retail media scale, see our hub guide on how automation and auctions let retail media scale.
Sources
- Amazon Ads, Sponsored Products best practices: advertising.amazon.com/library/guides/sponsored-products-best-practices
- CommerceIQ, "The 80/20 Rule No Longer Applies to SKU Management," May 2026: commerceiq.ai/blog/80-20-rule-no-longer-applies-to-sku-management
- CommerceIQ, Retail Media Management (Ally for Retail Media): commerceiq.ai/retail-media-management
- CommerceIQ, iROAS for Criteo retailers: commerceiq.ai/blog/new-retail-media-management-including-iroas-now-available-for-criteo-retailers
- Pacvue, beverage manufacturer Dynamic Dayparting case study, August 2025: pacvue.com/customerstories
- Pacvue, Advertising Automation platform page: pacvue.com/platform/need/advertising-automation
- Pacvue, Amazon Marketing Stream integration: pacvue.com/blog/pacvue-integrates-amazon-marketing-stream
- Skai, Portfolio Optimization: skai.io/capabilities/portfolio-optimization
- Skai, Budget Allocation in Retail Media: skai.io/blog/budget-allocation-in-retail-media
- Skai, Retail Media Solutions: skai.io/retail-media-solutions
- Topsort, "How Retail Media Operated in 2025": topsort.com/post/how-retail-media-operated-in-2025
- Optmyzr, "Structure Amazon PPC Campaigns for Maximum ROI," March 2026: optmyzr.com/blog/structure-amazon-ppc-campaigns-for-maximum-ROI
- Osmos, How automation and auctions let retail media scale: osmos.ai/blog/automation-auctions-how-retail-media-scales
- Osmos, The science of scalable retail media automation: osmos.ai/blog/automation-auctions-the-science-of-scalable-retail-media
- Osmos, How bidding and auctions work in automated retail media: osmos.ai/blog/bidding-auction-ad-retail-automated-2026
- Osmos ControlHub: osmos.ai/controlhub
.png)
