Case study · Google Ads · Smart Bidding · Diagnosis
An Ads account
bidding in the dark.
Small budget, three products, two languages. The campaigns were set to “maximize conversions”. The question nobody had asked: which conversions?
- 50%
- of conversion actions dead but still steering bids
- ~17%
- of a month’s spend on searches no campaign was for
- 3
- campaigns, one per product, each in two languages
Summary
Problem
Half of the account's conversion actions were dead e-commerce leftovers still feeding Smart Bidding, and the main campaign had no conversion goal selected at all.
Solution
Real lead conversions imported from GA4, one goal per campaign, language-split campaigns merged so each product learns from all its data, and wasted searches excluded.
What the account was optimising for
- Not one conversion action came from GA4. Checked over a real 90-day window rather than trusting the status labels: half of them, mostly “Add to cart” variants from a template, had zero conversions yet were marked primary and included in the account goal. They were diluting the signal Smart Bidding learns from.
- A campaign bidding with no goal. Set to “maximize conversions”, with no conversion selected for it. It came back once more after being fixed, so I logged the exact campaign names to stop the next check from being ambiguous.
- Every campaign scored on one catch-all event, including walk-in sign-ups at the front desk. I gave each campaign its own goal (see how those events were made reliable).
“The ad isn't showing”: not the budget
The usual reflex is more budget, and Google's own suggestion was about ten times the current one. The impression-share data said otherwise: 54% of lost impressions were lost to Ad Rank, only 31% to budget. Quality Score for the two main keywords hadn't even been computed yet, so it wasn't a relevance problem either. The diagnosis: a cold start, Smart Bidding with no conversion signal bidding cautiously against much bigger competitors.
So the work went into the levers that fix Ad Rank: ad copy (one ad went from “Good” to “Excellent”), sitelinks, negative keywords, and a real conversion signal to bid on. Budget only went up later, on the one campaign that was converting, once the data showed it losing impressions to budget.
One campaign per product, not per language
Separate English and Spanish campaigns split an already thin conversion signal in two. I merged each product into one bilingual campaign with a Spanish ad group, keeping keyword history. Then I checked what the merge had produced and caught a Spanish ad sending people to the English page.
Spanish keywords came from data, not guesses: real search volumes and the site's own Search Console queries. One neighbourhood showed real impressions but is several kilometres from both sites, and no lead had ever mentioned it, so I left it out.
Wasted searches
A month of search terms showed about 17% of spend going to searches no campaign was meant for: the brand name on campaigns that didn't exclude it, private-party venue searches in Spanish, “free” coworking, and competitor names. 33 negatives added through the API after a dry run. “Fiesta” deliberately kept, because company parties are business the terrace wants.
When the private offices filled up, that campaign was paused, and the budget increase went to the one campaign that was converting. An ad for a product you can't deliver is spend with no possible return.
““Raise the budget” is an answer. It's rarely the answer to the question that was asked.”