How do I tell that low ROAS comes from a lower average order value, not higher spend?

The first, basic checkpoint is whether dynamic values are used at all. When setting up a conversion, you need to select "Use different values" for each conversion. Without that selection, the system uses one fixed value for every conversion regardless of the real order amount, and correctness around discounts or shipping doesn't matter at that point, since the value isn't tied to the transaction to begin with.

Before reacting, it's worth confirming the cause really is competitive, not internal. Auction insights lets you compare your performance against other advertisers in the same auctions and see whether new players have shown up, or whether existing competitors have grown their impression share. The report is available for Search, Shopping, and PMax campaigns.

It helps to know upfront that Smart Bidding already manages typical seasonality automatically. The dedicated tool, seasonality adjustments, is recommended only when you expect a major conversion rate change, since the system already handles ordinary seasonal events on its own. So if the current dip falls within a historically typical seasonal cycle, Smart Bidding is likely already adapting to it, and no additional action is needed.

The key thing to understand: CTR and ROAS measure fundamentally different things, and strength in one doesn't automatically carry over to the other. The official Quality Score component Expected CTR measures how likely a click when the ad shows for a given query, relative to competitors. That's a purely predictive measure of copy appeal and query relevance, and it doesn't factor in what happens after the click or the economics of the conversion at all.

The official tool for separating products by type (promotional versus core) is custom labels in the product feed. They let you group products by any business logic within listing groups, seasonality, margin, promo participation, and target or analyze each label independently, instead of relying only on standard attributes like category or brand.

Official documentation directly describes the mechanic behind this temporary drag on the account average. The length of the learning period depends on three factors: how many conversions the campaign, ad group, keywords, or products get, how long the conversion cycle is, and the bid strategy itself (Manual CPC isn't affected by this period at all). A strategy calibrates to objective over up to 3 weeks or 1 to 2 conversion cycles, meaning performance is genuinely unstable during that window not because anything is broken, but because the system is still collecting data.

Official Target CPA documentation directly names the site as a separate, non-advertising factor in results: actual CPA depends on factors outside our control, such as changes to your website or ads, or rising competition in auctions. Google itself officially separates causes into "advertising" ones (ads, bids, competition) and "site" ones, rather than attributing everything to campaign settings.

This effect comes from the same Smart Bidding adaptation logic that applies to any significant change in conditions. A sharp rise in conversion during a promotion, followed by a sharp drop once it ends, counts as a significant change for the bid strategy too, requiring calibration time of up to 3 weeks or 1 to 2 conversion cycles, depending on data volume and conversion cycle length. The problem is that during the promotion itself, the strategy adapted to unusually high conversion and more aggressive bids, and once the promotion ends, it has to "unwind" that adaptation, a process that also takes time, during which ROAS can look worse than the business's real underlying performance.

The mechanics of accounting for returns were already covered as a technical setup: conversion adjustments let you restate or retract after the fact. For estimating the actual scale of returns' impact on final ROAS, a separate capability matters: since uploading an adjustment requires an adjustment time, you can segment your conversion data and see how much time typically passes between the original conversion and its later correction. That's a direct way to gauge how "delayed" the effect of returns is on your reporting.

The official mechanism built specifically for this scenario is the conversion window. The setting is ideal for businesses selling products with a a longer buying cycle the standard 30-day window. Google is directly acknowledging that the default setting doesn't suit every business model, and provides a dedicated official tool for long cycles instead of forcing everyone into one standard.