Google Ads ROAS dropped: a seven-point diagnosis

Check seven things in this order before you touch a bid: what Google counts, which campaigns serve, how much is modelled, the calendar, auction prices, the feed and the tag. Google's own troubleshooting page for Shopping campaigns lists 12 common reasons performance fluctuates, and the first three are a settings change, a feed change and a conversion-tracking change. Each check below says what to look at. A pass means that check is clean and isn't your cause; a fail means you've found a cause.

Did what Google counts change?

1. Lag and settings. Recent days look weak: Google says your recent performance "might not look as strong, because some of the people who clicked your ad haven't converted yet", and tells you to end your date range at least 30 days ago, or longer with a longer conversion window, for complete conversion data. Then open Change history in Google Ads and look for a new attribution model, a new conversion window, or an action switched between primary and secondary. Conversions created from Google Analytics events are "secondary by default", so check which actions feed the Conversions column. A window change also applies only going forward: Google's example is a window cut from 30 days to 10, which "will only apply to conversions recorded from that day forward". Pass: nothing changed on the date ROAS fell. Fail: a dated change matches the drop, so compare both periods under one definition.
2. Campaign mix. Performance Max and Search can compete for the same queries. Google says "Search campaigns containing an exact match keyword are prioritized to serve over Performance Max when a query hits an exact match keyword", and that brand exclusions limit what Performance Max can serve: "Brand settings inherently limit traffic." Compare ROAS by campaign, and by branded against non-branded queries. Pass: the mix didn't move, so look elsewhere. Fail: the mix moved while each segment's ROAS held, which makes the blended fall a mix shift. For illustration: if brand search returns a ROAS of 8.0 on 25% of spend and everything else returns 2.0 on the other 75%, blended ROAS is 0.25 x 8.0 + 0.75 x 2.0 = 3.5; at a brand share of 10% it is 0.10 x 8.0 + 0.90 x 2.0 = 2.6, with no change in either return.
3. Modelled conversions. Google's Conversions column reports observed and modelled conversions together, and consent-mode modelling needs "a daily ad click threshold of 700 ad clicks over a 7 day period, per country and domain grouping". Google also says "Consented users are typically 2-5x more likely to convert than unconsented users" (published by Google, with no sample or date, and not independently audited). So a change in your consent rate can change reported ROAS without a change in sales. Modelled data also "can take up to 5 days to fully process and stabilize", so the newest days move. Check whether your consent banner or its settings changed on the date of the drop. Pass: nothing changed, and Shopify revenue fell with Google's count. Fail: a banner change dates the drop while Shopify revenue held, so judge ROAS on Shopify revenue until the modelling settles.

Did the market change?

1. The calendar. Google notes that "Smart Bidding already manages seasonal events", but bidding doesn't create demand that isn't there. Compare the same weeks last year in Shopify: sessions, conversion rate and order value. Pass: last year's same weeks didn't dip, so the calendar isn't the cause. Fail: they dipped in a similar way, so some of the drop is seasonal. If you ran a sale, also compare the week after it ended with the week before it began.
2. Auction prices. Open Auction insights, which "lets you compare your performance with other advertisers who are participating in the same auctions that you are", and read impression share and overlap rate. ROAS is revenue per click divided by cost per click, so split it. Pass: cost per click is flat, so price isn't the cause and revenue per click (demand or the site) is where to look. Fail: cost per click rose while revenue per click held, so you're paying more for the same traffic. For illustration: a 10% rise in cost per click with nothing else changed turns a ROAS of 3.00 into 3.00 ÷ 1.10 = 2.73.

Did something you control break?

1. Feed and stock. Google lists "Changes to product attributes in the feed" and changes to "the review status of your ads or products" among the reasons performance shifts. Check Merchant Center for new disapprovals, and check that your best sellers are in stock: an ad for a sold-out product has nothing to sell (see Stockout or Meta reset? How to read a revenue drop). Pass: top products are approved and in stock. Fail: a best seller dropped out on the date ROAS fell.
2. The tag. Google's data exclusions page gives the example: "If a tag was accidentally removed from your site, Google Ads may not have tracked your conversions." Chart conversions per day in Google Ads beside Shopify orders and GA4 purchase events. Pass: the three lines move together. Fail: Google's line drops on one date while orders don't, which points to the tag, so check that day for a theme update, a checkout change or a new consent banner. A data exclusion then tells Smart Bidding to ignore the broken stretch, but it doesn't change reporting.

What doesn't Google Ads tell you?

Why ROAS moved, and whether the conversions it counts were caused by the ads. Google's data-driven attribution is built and reported by Google, and Google describes modelled conversions as attributed predictions that don't determine "whether or not a conversion happened". A ROAS drop says spend bought less reported revenue. A comparison group, such as a holdout, says whether it bought less actual revenue: see Cut a channel? Run a holdout first.

Sources, 30 September 2026: Shopping performance fluctuations (Google, 2026); Conversion windows (Google, 2026); Conversion delay (Google, 2026); Primary and secondary conversion actions (Google, 2026); Performance Max (Google, 2026); Brand settings (Google, 2026); Consent mode modeling (Google, 2026); Modeled online conversions (Google, 2026); Seasonality adjustments (Google, 2026); Auction insights (Google, 2026); Data exclusions (Google, 2026)

Related answers

• Google Ads vs Shopify revenue: why the numbers differ
• Why ROAS falls when you scale spend
• Stockout or Meta reset? How to read a revenue drop
• Does falling consumer confidence lower your ROAS?
• Incrementality testing