The Dashboard Lied.The Bank Statement Didn't.
A Turkish founder selling herbal wellness into the US handed us a Google Ads account reporting 25× ROAS against a Shopify screen showing one order a week. In the 90 days after handover we fixed what the account counted, then rebuilt the store the clicks were landing on — purchase ROAS 0.7× to 19.6×.
Name withheld. Numbers real.
The Whole Business Is a Screen
The founder is Turkish and runs the entire thing from Türkiye. His customers are American, his prices are in dollars, and his products — herbal teas, aloe drinks, shakes, supplements, skincare — come from a US manufacturer whose warehouse ships each order straight to the buyer. He holds no stock and never touches a package. What he sells are consumables, so people finish them and buy them again, which is the part that pays off by month 5.
So the business arrives on a single screen: the Shopify orders page. That one tab is his warehouse, his shipping desk and his scoreboard. Which makes the next eighteen weeks easy to follow — they're a story about two screens disagreeing, and it goes wrong first, then expensively wrong, then right.
Five Orders in Eight Weeks
He switched Google Ads on himself, in good faith, with no idea what was wired underneath it. Within a week the dashboard was reporting 25× ROAS, and it kept saying so. The orders screen, one browser tab away, showed five sales in eight weeks.
The gap was mechanical, and it began at the conversion tag. That tag fired on add-to-carts and checkout starts, so every shopper who hesitated at the last step got filed as a sale. Google then did what Google does and optimized toward the thing it had been told to count — it went out and found more hesitators. Broad match and an empty negatives list gave it plenty of room. It hunted in India, Mexico and Canada, for a store that only sells inside the United States.
Google Shopping never got off the ground either. He submitted 364 products and Google rejected 364 of them: 22 for health-claim wording, 89 for promotional badges burned into the images. A rejected feed doesn't show a weaker version of your listings, it shows nothing at all, so the Shopping half of the plan was never actually running. It just looked like it was, on the same dashboard.
Two months in, the bill came to about $375 and the orders screen said five. He concluded that ads don't work for products like his — which is exactly what that account would tell anyone who read it honestly.
| Week | Orders |
|---|---|
| 1 | 0 |
| 2 | 1 |
| 3 | 0 |
| 4 | 1 |
| 5 | 1 |
| 6 | 0 |
| 7 | 1 |
| 8 | 1 |
| 9 | 13 |
| 10 | 14 |
| 11 | 6 |
| 12 | 14 |
| 13 | 7 |
| 14 | 2 |
| 15 | 1 |
| 16 | 2 |
| 17 | 21 |
| 18 | 29 |
The Agency's One Idea
In week 9 a performance agency took over, and its answer to every question was budget. Spend went to ~$2,700 in a single month, seven times what the founder had spent across his entire two. Orders moved with it: 47 in four weeks. After eight weeks of roughly one a week, 47 felt like the thing finally working.
Counted honestly, though, the month lost money. Purchase-only ROAS came in at 0.71 — every dollar of ads returning 71 cents of product. Conversion rate 0.34%, CPC $1.02. The traffic got more expensive without getting any better.
The agency's report said it was working, and by the agency's arithmetic it was: the same inflated conversion column as before, now quoted as a blended 9.4×. Nobody had touched the tag. They'd simply pushed seven times more money through it. The founder held the report next to his bank statement, found that the two described different businesses, and started looking for someone else.
One efficiency number stayed honest the whole way through: MER, all store revenue divided by all ad spend. A broken conversion tag can't reach it, because the numerator comes from the till and not from the ad panel. That's why months 1 and 2 have a MER at all, where purchase ROAS is simply blank: 0.98, then 1.23. The best month before the reset got to 2.08×. The dashboard was claiming 25× over the same stretch.
| Month | Spend ($) | ROAS | MER |
|---|---|---|---|
| M1 | $183 | purchases weren't tracked yet | 0.98× |
| M2 | $194 | purchases weren't tracked yet | 1.23× |
| M3 | $2,709 | 0.71× | 2.08× |
| M4 | $369 | 6.57× | 7.5× |
| M5 | $228 | 19.6× | 25.9× |
1,407 Conversions, 74 Sales
The account came to us in week 13. The audit took days rather than weeks, which is usually a bad sign — a mis-tracked account isn't hard to read once you know which column to distrust.
We pulled the full conversion record to see what the counter had been counting. Of 1,407 recorded "conversions," 74 were purchases. About 5%. Against those, the account claimed ~$59K in conversion value where real sales came to ~$8.9K, a 6.6× inflation sitting inside every report anyone had read. Both the agency and the founder had been making decisions off that number for three months, in opposite directions.
The rest of the audit filled in the smaller leaks. Around 12% of the spend had gone outside the United States entirely. Tablets had taken about $30 and sent back nothing. And 94% of clicks were arriving from phones onto a store built desktop-first, which meant the ads were paying to deliver people to a page that fought them. With the feed still rejected on health claims, Shopping wasn't coming back any time soon, so search would have to carry the whole budget.
First, make the numbers tell the truth.
Only ~5% were sales.
| Event | Count |
|---|---|
| Add to cart | 588 |
| Cart views | 554 |
| Checkout starts | 149 |
| Payment info | 42 |
| Real purchases | 74 |
The Boring Part
The first move was subtraction. We cut monthly spend from ~$2,700 to ~$370, about 86%, because there was no sense paying to push traffic through a broken tag onto a page that couldn't sell to it. Then two workstreams, in parallel:
- Conversion tracking rebuilt from zero — only a paid order counts
- Targeting narrowed to the US alone; tablets excluded
- Broad match out, exact and phrase in, with a negatives list that actually exists
- Campaigns consolidated so the budget stopped bidding against itself
- A weekly loop: search-term review, negative sweep, budget guardrails
- Product pages rewritten around what each product is and does, inside health-claim rules
- Mobile-first rebuild — 94% of the clicks were already there
- Brand voice and trust signals across the store
- Creative refresh for the ads and the product imagery
The health-claim rules that had killed the feed shaped the rewrite too. The pages couldn't promise outcomes, so instead they said what's in each product, what it's for and who it's for. That turned out to sell harder than the claims would have, because it gave a first-time buyer something specific to believe.
Meanwhile the orders screen read 2, then 1, then 2. Twelve orders across three weeks, and nobody hit the panic button — we'd told him in week 13 that the account would go quiet before it got better, because the bad spend has to leave before good spend can go in. Three boring weeks.
21 Orders, Then 29
We turned the ads back on in week 17 and the screen answered with 21 orders. Week 18 came in at 29 — more than his first two months put together, with 8 of them landing on the best single day. That's the green bar in Fig. 01.
Month 4, the reset month, closed at 6.6× purchase ROAS and 0.94% conversion, with CPC down 47% to $0.53 now that the campaigns weren't bidding on most of the planet. Month 5 closed at 19.6× ROAS and 6.18% conversion on ~$230 of spend.
Average order value went from $108 in the agency month to $137, up 27%, and that's the product pages rather than the ads. Better traffic was finally landing somewhere that could sell to it. One customer in nine came back for a second order before the 90 days were up, which is what consumables do once the first purchase goes well.
CPC went back up in month 5, to $0.74, because the budget had concentrated onto the handful of queries with real buying intent behind them. ROAS tripled anyway. The orders map reads like a US store now too: Texas first, then Florida, Ohio and New York, and nothing at all outside the market.
CPC ticked up in month 5 as budget concentrated on high-intent queries. ROAS tripled anyway.
| Month | Conversion rate (%) | CPC ($) |
|---|---|---|
| M3 | 0.34% | $1.02 |
| M4 | 0.94% | $0.53 |
| M5 | 6.18% | $0.74 |
Out-of-market spend, before: ~12%. After: 0%.
| State | Orders |
|---|---|
| TX | 17 |
| FL | 10 |
| OH | 9 |
| NY | 9 |
| GA | 9 |
| CA | 9 |
| PA | 7 |
| DE | 4 |
What Actually Did the Work
Only Paid Orders Count
Google Ads optimizes toward whatever you tell it to count. Told to count add-to-carts, it went looking for people who love adding things to carts, and it found them. That isn't the algorithm failing; it's the algorithm succeeding at the wrong task. Rebuilding the conversion action so a paid order is the only thing that registers reads like a chore on a checklist — every decision after week 13 stood on it.
The Pages Did More Than the Bids
The rewrite, the mobile-first rebuild and the trust work carried conversion from 0.34% to 6.18%. No bidding strategy moves a number that far, and none of them explain a basket going from $108 to $137 either, because that's a visitor deciding to add a second item — a page's job, never an ad's. Work that doesn't just look good, it sells.
Weeks 14, 15 and 16
A 2-1-2 stretch is where most resets die. The client panics, the agency reaches for a hero move, and the account goes back to being whatever it was before anyone touched it. We spent those three weeks running the same unglamorous loop — search terms, negatives, budget guardrails, creative iteration — and nothing on that list is worth writing down on its own. Together they're the reason week 17 happened.
The Expensive Month Lost Money
Month 3 outspent month 5 by nearly twelve to one and got 0.7× for it, where ~$230 got 19.6×. Budget doesn't repair an account; it multiplies whatever the account already is. Which is why the order matters: fix first, spend second. Doing it the other way around is how ~$2,700 bought 47 orders' worth of proof that the tag was broken.
There was a stretch where my whole job was refreshing the orders screen, and most days nothing moved. Now I wake up to orders that came in while America was awake and I wasn't. Best part: what the screen says is what the bank says. I've stopped asking anyone whether the numbers are real.
Questions We Get
Why does Google Ads report conversions I can't find in Shopify?
Because the tag is almost certainly firing on something short of a sale: add-to-carts, checkout starts, sometimes just pageviews. In this account, 1,407 counted "conversions" contained 74 actual purchases, about 5%, which is why the reports and the bank statement described two different businesses. Pull both numbers for the same date range and put them side by side. If they diverge, fix the tag before you change a single bid.
Is the ROAS in my dashboard real?
Only if the conversion value behind it is real. This account reported ~$59K in conversion value against ~$8.9K in actual sales, a 6.6× inflation, so its 25× ROAS wasn't optimistic — it was fiction. Build a purchase-only conversion action and read ROAS from that column and nothing else. The number will get much smaller, and it'll start being worth something.
Will a bigger Google Ads budget grow my store?
Not on its own. Here the agency spent ~$2,700 in one month, seven times the founder's entire two-month bill, and bought 47 orders at 0.71 purchase ROAS. After the rebuild, ~$230 a month returned 19.6×. Budget scales an account's existing economics in both directions, so fix what it counts and who it targets first, then turn it up.
Why did Google Shopping reject all of my products?
Health and wellness catalogs hit Google's healthcare-claims and image policies fast. This store had all 364 products rejected: 22 for claim wording, 89 for promotional badges sitting on the product photos. Rewrite the listings around what the product is instead of what it promises, take the overlays off the images, and let search carry the budget while you work through the queue. A rejected feed doesn't serve reduced listings, it serves none, so pausing it costs you nothing.
How long does it take to fix a broken ad account?
The audit took days, the rebuild about four weeks, and the turn showed in week 17 with 21 orders, then 29. Plan for the quiet part in the middle: orders here fell to 2, 1, 2 a week while the bad spend cleared out. That stretch is the work, not a sign the work isn't landing.
How much does branding affect conversion rate?
In this case, more than bidding did. Rewritten product pages, a mobile-first rebuild and clearer trust signals moved conversion from 0.34% to 6.18% and average order value from $108 to $137. Ads buy the visit; the store makes the sale.
What's your account actually counting?
Send a brief — we'll come back within 48 hours with a prioritized action list, not vague advice.