Product titles written for a website, not for search.
Titles read like showroom labels. They were missing the words people actually type — the metal, the purity, the stone, the occasion, the style.
UNIBRIGHT SOLUTIONS · AI PPC · CASE STUDY
From 2.1x to 5.8x ROAS with Google Shopping
We optimized Google Shopping campaigns for a high-value jewellery retailer by improving product feeds, refining campaign structure, and fixing revenue tracking to drive more sales, lower acquisition costs, and maximize ROAS.
5.8xROAS
+164%NEW REVENUE
−38%COST PER SALE
840PRODUCTS OPTIMIZED
01 / THE PROBLEM
The Problem
Gold is a hard thing to sell through advertising. The basket value is high, the decision is slow, and the buyer needs to trust an online store with a purchase most people still prefer to make in person.
The account had the usual shape of a struggling ecommerce campaign. Money going out, some sales coming in, and no clear read on which of those sales the advertising had actually caused. Most of what looked like performance was the store buying back people who were already searching for it by name.
Underneath that was a more basic issue. In Shopping and Performance Max, Google does not match your keywords — it matches your product data. The campaign was being managed through bids and budgets while the thing that actually decides what gets shown, the product feed, had barely been touched.
REVIEW
What Our Review Found
Missing product attributes.
Gaps in the identifiers and product details Google uses to decide which searches a product can appear for. Every gap narrows the range of searches a product can win.
Brand and non-brand mixed together.
One campaign covering both, so the account looked profitable overall while it was mostly buying its own name back cheaply. The real performance was hidden inside an average.
Every product treated the same.
Best sellers and slow-moving stock competing for the same budget at the same bids, regardless of margin.
Out-of-stock products still being advertised.
Paying for clicks on items nobody could buy, and sending those shoppers to a dead end.
No revenue values in conversion tracking.
Without a value on each sale, the platform cannot bid toward a return. It optimises for the number of sales rather than the money they bring in — a serious problem when basket values vary this much.
Nothing catching people who did not buy first time.
For a slow, high-value purchase, almost nobody buys on the first visit. There was nothing bringing them back.
BUILT
What We Built
Rewrote every product title to a consistent structure.
Metal and purity, product type, stone, style and occasion, in a fixed order. This single change widens the range of searches each product can appear for more than any bid adjustment.
Filled in the missing product data.
Identifiers, materials, weights and product categories completed, so Google knows exactly what each item is.
Kept pricing in sync.
Gold prices move. A feed price that disagrees with the site price gets products disapproved, so pricing updates automatically rather than manually.
Excluded out-of-stock items automatically.
No spend on products nobody can buy.
Split brand from non-brand.
So the account reports two separate truths: what it costs to defend the brand, and what it costs to find a new customer. Those are different jobs with different economics.
Grouped products by price band and margin.
High-value pieces get different budgets and targets from entry-level items, because a sale is worth a different amount.
Fixed revenue tracking.
Actual order values passed back, so bidding can target a return rather than a volume of orders.
Built remarketing for a slow decision.
People researching a high-value purchase over days or weeks now get brought back, rather than being paid for once and forgotten.
Put trust signals in the ads and on the landing pages.
Certification, returns and guarantees placed where a hesitant buyer looks for them. For a purchase of this size, reassurance converts better than a discount.
IMPACT
Where AI Helped — and Where It Did Not
AI made the feed work possible. Rewriting titles and filling attributes across a full catalogue is exactly the kind of job that gets postponed indefinitely because it is enormous and dull. AI drafts it in days.
Our specialist set the title structure, decided which attributes mattered for jewellery searches, and worked out the price bands. Then the output was checked, because a wrong material or purity in a feed is not a typo — it is a misdescribed product and a disapproval risk.
05 RESULT
The Results
RESULTDETAIL
5.8x return on ad spendUp from 2.1x, over six months
Non-brand revenue up 164%Sales from people who were not already searching for the brand — the number that shows the advertising is finding new customers
Cost per sale down 38%Higher sales volume on a lower budget
840 products optimisedTitles and attributes rebuilt across the catalogue
98% of products approvedUp from 71% — disapproved products cannot be advertised at all
Non-brand impression share 61%Up from 22% — the feed work is what widened the reach
Revenue tracking fixedOrder values now recorded, so bidding targets return rather than order count
MEANS
What This Means
The feed is where the leverage is.
Bids adjust what you pay. Product data decides whether you appear at all. One is a dial, the other is the door. Taking product approvals from 71% to 98% alone put a quarter of the catalogue back into the auction.
Brand and non-brand have to be separated.
Mixed together, they flatter an account. Split apart, you can see what new customers actually cost — and that is the only number that tells you whether to spend more.
Tracking revenue, not orders, changes the bidding.
When basket values range widely, optimising for order count quietly pushes budget toward the cheapest sales instead of the most valuable ones.
High-value purchases need a second visit.
Nobody buys gold on impulse from a first click. Without remarketing, most of the money spent on the first visit is wasted.
Reassurance beats discounting at this price point.
Certification, returns and guarantees do more for a hesitant high-value buyer than money off.
FIX
What This Does Not Fix
Worth being straight about:
Prices that are not competitive.
Shopping results put you next to your competitors with the price showing. There is nowhere to hide.
Weak product photography.
In jewellery the image is the product. A poor photo loses the click before price is even considered.
Thin margins.
Paid advertising needs enough margin to pay for the click and still profit. Where the margin is too thin, the honest answer is that this channel is not the right one.
Trust you have not earned.
A new store selling high-value items faces genuine hesitation. Advertising can address it; it cannot replace reviews and a track record.
Stock gaps.
Advertising works best on products you can reliably supply.
WORKS
Why This Works Elsewhere
Any online store running Shopping or Performance Max has this problem, and most have it badly. The signs are easy to check: product titles that read like showroom labels, one campaign covering brand and non-brand together, and no revenue values in the conversion data.
Fix the feed before touching the bids. It is unglamorous work and it is where the return is.
NEXT STEP
The Next Step
Give us access to your Google Ads account and your product feed. You get back how many of your products are actually eligible to appear, what your non-brand advertising really costs you, and the feed changes that would widen your reach without raising your budget.
Services Provided:
Google Ads management, Shopping and Performance Max campaigns, product feed optimisation, campaign structure and segmentation, conversion and revenue tracking, remarketing, landing page conversion support, reporting