Why is my ROAS dropping? Check outside the ad account first

ROAS often falls for a reason outside the ad account: tracking, attribution, the website, the offer, stock, the season or your margin. If CPM and click-through rate moved, suspect the ads; if they held and returns fell anyway, the cause sits after the click or in the counting.

If Ads Manager says ROAS fell and Shopify says sales didn’t, believe Shopify. One bad day is noise and three in a row are a problem, as our Framework puts it, so change nothing while you diagnose. Steps 1 to 3 come first, and step 3 tells you which of the rest you need.

Why the cause is often outside the ad account

Per click, ROAS = conversion rate × average order value ÷ cost per click. Cost per click is set in the ad account, while conversion rate and order value are set on your website. Split cost per click into cost per impression ÷ click-through rate and you have four inputs, two on each side.

Say 2% of the people who click buy, the average order is $80 and a click costs $0.80. Each click is worth $1.60, so ROAS is 2.0x. If a theme update takes conversion to 1.5% and nothing changes in the ad account, ROAS falls to 1.5x while every ad metric looks the same.

Reading the symptoms

What you see	Likely cause	Go to
Platform ROAS and Shopify disagree	Tracking, attribution or non-ad revenue	Steps 1 and 2
Both down, ad metrics steady	Site, offer or stock	Steps 4 to 6
CPM up, click-through steady	Auction or season	Step 7
Click-through down on older ads	Creative fatigue	Step 9
Slow slide, spend up	The cost of buying more	Steps 8 and 10
Slow slide, same ads for months	Creative fatigue	Step 9

1. Check that sales are still being counted

Compare the purchases Meta and Google recorded with Shopify orders for the same days. If tracked purchases fell sharply while orders held, it’s tracking. Fix it before anything else, because every later number depends on it.

The usual causes are on the website: a theme change or app uninstall that removed code, a checkout change, a new cookie banner, or the Conversions API (Meta’s server-side connection) disconnecting. Shopify also upgraded the Thank you page of every store that missed its 26 August 2026 deadline, so a purchase tag pasted into the old page needs replacing.

On Shopify, set the Facebook & Instagram channel’s data sharing to Maximum, and check in Events Manager that the pixel and Conversions API both send Purchase and are deduplicated. We hold Purchase event match quality, Meta’s score for how well your customer details match orders to people, to 8 or better, the floor 700 Miles Publishing cleared at 9.1 out of 10. On Google, turn on the Google & YouTube app’s option that sends hashed customer details with conversions, and count Purchase as every conversion with no duplicate tags. Our tracking work covers this.

On Glass Flooring Systems, a lead-gen account, moving a form into a cross-origin iframe (a form hosted on another site, which the page’s tag can’t see into) took Meta’s CompleteRegistration from 8.7 a day to 0.0 a day for nineteen days. Once the counting was rebuilt, attributed leads went from 29 in July to 74 in August while total leads from every source held at 126 then 130. Demand hadn’t moved. The counting had.

2. Check whether the business dropped or only the dashboard

Divide total Shopify revenue by total ad spend across every platform, for the drop and for an equal window before it. That’s blended ROAS, or MER, and no platform can inflate it. If blended held while platform ROAS fell, it’s a reporting problem. If both fell, the drop is real. If the platforms held and Shopify fell, it isn’t the ads.

Meta’s default attribution setting is 7-day click, 1-day view and 1-day engagement, and Google’s data-driven model spreads credit across the ads a buyer touched, so both can claim one order. On Loose Leaf Tea Market, Meta reported 2.23 times what the store’s own order records credited it with.

Then split it three ways. In Ads Manager’s Compare attribution settings, put 1-day click beside 7-day click: if only 7-day fell, recent sales haven’t been credited yet. In Shopify, split new from returning customers, because falling repeat revenue drags blended down while the platforms look fine. In Google, split brand search from the rest: on Aaron Basha a 9.16x account drew 87.0% of its value from brand-name searches, and everything else returned 1.62x.

3. Split the drop: which input moved?

Pull CPM and click-through rate from the ad accounts, and conversion rate and order value from Shopify, against the window before the drop. Then follow what moved:

• Conversion rate or order value down, ad metrics steady: steps 4 to 6, and hold off on new creative.
• CPM up, click-through rate steady: step 7.
• Click-through rate down, CPM steady: step 9, now.
• Both moved: step 9, then step 7, the order our Framework uses when each new customer costs more.

If ROAS slid over weeks

Compare against a window from before the slide began and the same weeks last year, not last month, which is already part of the slide. The slow causes are spend creeping up (steps 8 and 10), the same ads running for months (step 9), discounts dragging order value (step 5) and tracking decaying bit by bit (step 1).

4. Check site conversion, by device and by funnel step

In Shopify’s analytics, compare conversion rate now with the weeks before the drop began, by device and landing page, filtered by UTM source to your paid traffic. Read the funnel: sessions, added to cart, reached checkout, purchased. Then go through checkout on your own phone. Mobile falling alone means something broke on mobile: a pop-up, load time, a layout break, a missing express checkout.

On Southern Slumber, site conversion had fallen from above 2% to below 1%, with a Shopify theme change the suspected trigger. Meta’s funnel events later showed only 20.9% of add-to-carts reached checkout, against a benchmark near 50%, while checkout itself completed at 43.9%, so the loss sat in the step before checkout. Within a month we told them to cut ad spend to fund the site fixes.

On Flavortown, the site relaunched in the same weeks we rebuilt the Meta account, pixel and catalog from zero and took the learning reset, and account ROAS fell from 2.3x before that relaunch to about 1.3x by late October. Two causes stacked, and the ad account couldn’t split them. The page data could: site-wide conversion sat below 1%, against 2% to 4% on product pages. Compare your top product pages’ conversion rate with the site-wide rate.

5. Check the offer and order value

Compare this period’s offer with the last one’s. Did a sale end, a code change, the free-shipping threshold move, a bundle disappear, a price go up, or the mix shift toward cheaper products? If order value fell while conversion held, more spend won’t fix it.

On Southern Slumber the minicart quoted free shipping at $49 in one place and $29 in another, and three separate 15% discount codes were live at once. On Flavortown the same 10-piece stainless set sold for $139 on Amazon against $199 on the brand’s own site. And check that your ads promise what checkout charges.

6. Check stock and the product feed

In Shopify, pull sales by product variant. A best-selling variant whose sales drop to zero while its page still gets sessions is a stockout, and every ad pointing there is wasted. Meta catalog ads need the same check in Commerce Manager.

On Google, open Merchant Center: disapprovals, out-of-stock products, a stale feed, prices that differ from the site. If Shopping fell while Search held, suspect the feed before the bids. On Aaron Basha, 40.5% of the items in Merchant Center could not enter an auction and 531 pointed at dead pages. The dead links came from the feed’s primary source, a hand-maintained Google Sheet that couldn’t see a Shopify deletion.

7. Check the calendar and the auction

Choose the comparison window before you read any number: the same weeks last year rather than last month, never a window with a sale in it, and long enough for your order volume. On Aaron Basha, two single days were 62.8% of a month’s value, and without them 9.16x reads 3.41x, so that account is judged on 60 to 90 day windows.

If everything fell on one date, check the calendar for holidays, events and competitor sales. Rising CPM with steady click-through rate points to the auction, and if Meta’s CPM spikes while Google’s cost per purchase holds, it’s Meta’s auction, not your site. If it’s seasonal, hold steady: Google says Smart Bidding already manages seasonal events.

The Fitzroy, a Toronto formalwear rental business that runs on an event calendar, saw Google spend fall from $3,000 in June to $2,150 in July while cost per click rose about 30% and clicks fell from roughly 4,200 to about 2,000. Headlines and descriptions were ruled out, which left auction pressure: more advertisers bidding on the same keywords.

8. Check what changed inside the ad account

Open the change history in Meta and Google and look first for changes nobody on your team made. On Aaron Basha, Google’s auto-applied recommendations had pulled 11 keywords on 4 August, including the exact-match brand term, and the brand campaign had served $0 on a stale disapproval while a reseller ran 10% off on brand searches. On Loose Leaf Tea Market, Meta’s Health and Wellness enforcement in March 2026 collapsed spend 82% and purchases 93%, and purchases falling faster than spend is a ROAS drop. So check Meta’s Account Quality page, disapprovals on your top spenders and Google’s auto-applied recommendations.

Then read your team’s edits. If spend went up, part of the drop is the cost of buying more, so go to step 10. On Meta, an ad set leaves learning after about 50 results in the week after its last significant edit, and editing during learning resets it. Bundle your edits and change one thing at a time, so you know what moved the number.

9. Check for creative fatigue

Look ad by ad. It’s fatigue when specific older ads show rising frequency and CPM and a click-through rate sliding for weeks rather than days, while fresh creative holds up. If new ads fall as hard as old ones, go back to step 7. Read hook rate, the share of viewers who watch past the opening seconds, by concept too: ad-level frequency can look normal while the same people have seen one idea in many ads.

Before you decide fresh creative isn’t holding up, check that it delivered. On CHUUG, 286 of 824 active ads had spent under £5 in 30 days, held back by a Page cap of about 1,000 ads that quietly stops new ads delivering. If it is fatigue, brief new hooks (the opening line or shot), angles (the reason to buy you lead with) and formats.

10. Judge the new ROAS against break-even, not last month

ROAS tends to rise when spend falls and fall when spend rises, so the question is whether the new number still makes money at the volume you’re buying.

With made-up numbers: a $100 product with $35 of product cost and $10 of shipping keeps $55, a 55% margin before ads and payment fees. Break-even ROAS is 1 ÷ 0.55, about 1.82x, and each ad dollar earns margin × ROAS − 1: $0.65 at 3.0x and $0.32 at 2.4x, so going from 3.0x to 2.4x pays only if spend more than doubles. A steady 2.0x earns $0.10 and turns into a loss once shipping passes $15.

On CHUUG, the 1.74x break-even in the system had never been measured: order-level records put it at 1.61x, and re-costed shipping in August 2026 took it to 1.83x blended. On Natural Earth Paint, Meta was returning 1.6 in August 2024 and nobody knew the break-even, later modelled at 1.2 on product cost alone and 1.8 with operating costs in, a stricter bar than the formula above.

When this checklist doesn’t fit

If you sell leads, your CRM replaces Shopify as the reference. If every drop gets answered with “new creative” and nobody shows you the store numbers, read the questions I’d put to any agency. A falling ROAS reported as another good week is an accountability problem, not a market one.

Still can’t name the cause? Take your notes to whoever runs your ads, or to us. Our strategy call is for brands doing at least $50,000 a month in revenue and spending at least $5,000 a month on ads. You get a straight answer on whether we can help.

How this was made: drafted with AI help from Growth Mate’s working playbooks and case records. Every figure links to the page it comes from, except the made-up worked examples, which say so.