A Nine-Month ROAS Plateau Requires an Operating-System Audit

Nine months is long enough to require a structured investigation. It is not long enough, by itself, to assign blame.

A catalog campaign is a chain of connected decisions. Product data determines whether items are eligible to serve. Inventory determines whether a click can plausibly convert. SKU selection determines which products receive exposure. Creative affects attention and product discovery. Audience and channel allocation determine where the budget is placed. Measurement determines whether the resulting activity is being interpreted correctly.

A weakness anywhere in that chain can produce the same headline: stagnant ROAS.

For example, Google Merchant Center documents product-data quality requirements and item-level approval issues that affect catalog eligibility. If high-intent products have approval or attribute issues, an agency may respond to falling revenue by changing bids while the stronger products remain unable to serve. The work looks active; the underlying constraint remains untouched.

ROAS should remain distinct from CAC, ACOS, revenue contribution, and net profit. A campaign can have an acceptable platform ROAS while acquisition cost rises elsewhere. A marketplace ACOS can improve while revenue contribution declines. None of those metrics proves audited net profitability after returns, COGS, shipping, fees, and taxes.

Recovery Starts by Establishing a Trustworthy Baseline and Failure Hypotheses

A recovery plan should begin with a baseline that both the brand and agency accept as the reference point. Without one, each weekly report becomes an argument over whether a movement in ROAS represents progress, seasonality, attribution drift, or a temporary change in mix.

This is a buyer-facing diagnostic request, not proof that the incumbent caused the plateau. A brand may find that the agency has been flagging inventory or feed issues that were never resolved internally. It may find the opposite: repeated bidding changes with no record of the conditions that prompted them.

The important shift is from observations to hypotheses. “Catalog campaigns are underperforming” is an observation. “High-spend products are becoming unavailable before exclusions take effect” is a hypothesis that can be tested against stock, product status, spend, and conversion data. “Creative fatigue is reducing response among a defined audience” should also lead to a test with a stated comparison, decision date, and owner.

Granular reporting alone does not make the baseline trustworthy. A SKU report can be precise and still mislead if inventory status is stale, discounts are omitted, or the conversion definition differs across channels. Google Merchant Center provides item-level diagnostics for identifying product-data issues and affected inventory, which is useful evidence for investigating catalog eligibility; it does not settle an attribution or commercial-value question.

An agency that cannot state what would change its mind is reporting activity rather than managing a recovery.

Catalog Recovery Requires Controls Across Feed Health, Inventory, SKU Selection, Creative, Audiences, and Channels

The reason catalog accounts become difficult to repair is that each domain can mask another.

A feed problem can resemble weak demand when products stop serving or lose required attributes. Inventory can resemble a landing-page problem when shoppers arrive on products with missing core sizes. A weak SKU mix can resemble a bidding problem when budget follows products that attract clicks but do not convert. Creative fatigue can look like an audience issue when the same assets have exhausted their ability to earn attention.

The agency should be able to trace the path from product state to media decision.

Consider a product set where top-selling SKUs face stockouts or missing variant sizes. Without dynamic catalog filtering, performance marketing spend continues to flow towards these low-converting landing pages, systematically depressing account ROAS regardless of bid optimizations.