Case study · Jewelry · DTC
CPA Down 38% Without Touching Bids
Title rewrites on the top 80 SKUs and custom labels did the work that bid tuning could not.
Monthly revenue
From €216.0K
ROAS
From 1.8x
Monthly spend
From €120.0K
Time in the account
Written by Christopher Krassnig, Founder & CEO of ZenoX Media. Runs the accounts in these case studies. Every number below comes out of one of them.
How Do You Cut Jewelry Google Ads CPA by 38% Without Changing Bids?
Published Last updated
- Niche
- Jewelry DTC
- Market
- Netherlands
- Spend on day one
- €120.0K/month
- Where they started
- 1.8x ROAS on €216.0K/month
- How long we ran it
- 90 days
What we changed
The Spec Sheet, Not the Sales Pitch
Every lever pulled on this jewelry dtc account, in the order the story tells it. No move listed here that the story does not show.
- 01
Title rewrites on the top 80 SKUs, driven by real search-term volume
- 02
Auto-rollback armed under every title change
- 03
Custom labels split Champions, Sleepers, and Wasters
- 04
Champions got their own PMax; Wasters got cut
- 05
Zero bid changes for the whole 90 days
The story
How This Jewelry DTC Account Actually Scaled
- 01
ROAS 1.8 was not a bidding problem
Previous agency had been tuning Smart Bidding for months. CPA kept climbing. €120K a month going out at 1.8x. The leak was upstream: titles that read like SKU codes, no custom labels, Champion products buried in the same campaign as Wasters.
- 02
What each of those three things costs you
Titles first. Shopping matches a shopper's words to the words in your title. A title that reads like a part number matches almost nothing a human types, so those products never get into the auctions worth being in. Then labels, or the lack of them: with no custom labels there is nothing to split a budget by, so every product sits in one pool. And a single pool means the bidding sees one blended average of your best and worst products, and optimises to the middle of it.
- 03
Why more bid tuning could never fix it
Smart Bidding only sorts what you hand it. It decides how hard to compete for an auction. It cannot decide which auctions your catalogue is eligible for, because your titles and attributes already did that. Months of target changes were re-arranging the same limited set of impressions. The bid was never going to reach the actual problem.
- 04
Title rewrites on the top 80
Adaptive Feed Optimizer reads search-term volume and rewrites titles around the queries actually driving sales. Not guesses about what shoppers might type, the account's own search terms. 80 SKUs ship in one batch, because a handful at a time takes a quarter to show anything. Auto-rollback armed if any drop below baseline.
- 05
The rollback is what makes it safe
Every rewritten title carries its old numbers as a baseline. If a product falls under that line, the old title goes back automatically. That is the only reason you can ship 80 changes on a €120K account in one go without holding your breath. The downside is capped before the change goes out, not argued about after.
- 06
Custom labels split the budget
Champions got their own PMax. Wasters got cut. Sleepers got a fair shot in a low-budget test campaign, because a product with no sales is not always a bad product. Sometimes it simply never got shown, and a small ring-fenced budget is the cheapest way to find out which it is. Smart Bidding suddenly had a clean signal to optimise against.
- 07
The CPA curve, month by month
€78, then €62, then €53, then €48. The first drop is the new titles walking into better auctions. The drops after that are the bidding compounding on cleaner data, because every week the model is learning from a catalogue that finally makes sense. Nothing about that curve needed a bigger budget.
- 08
CPA down 38%, ROAS up 60%
€120K in at the start. €120K in at the end. Same spend, whole way through, and not one bid changed in ninety days. Better data did all of it. The bid was never the problem. The story Google was reading off the catalogue was.
- 09
What to take from it
Before you touch a target, go and read your own product titles. If they do not contain the words your buyers type, no bidding strategy on earth will save the account. Fix what Google reads first. Then let the machine do its job.
The curve, knowing what happened
Show your working
How We Measured This
- The window
- 90 days, read as monthly snapshots at month 0, 1, 2 and 3. Month 0 is the account before we started.
- Where the numbers come from
- Revenue, spend, ROAS and cost per sale all come from the client's own Google Ads account. Spend was held at €120K a month the whole way, so the CPA drop is not a budget effect.
- Why the brand is not named
- The brand is not named. These are live accounts, and the edge they built stays theirs, so we publish the numbers and keep the name out of it.
- What you can verify
- The source account is private. ZenoX has not published a redacted export or an independent check for this case.
Drop the URL on WhatsApp. We pull your account up live on a thirty-minute call and show you which leaks would get fixed first.
If we cannot scale you, we say so on the call. No retainer trap. No pressure.