If you run Google Ads for jewellery brands and your target ROAS is a round number somebody picked in a meeting, you are almost certainly bidding wrong. High AOV compresses your conversion volume until Smart Bidding is guessing. Return rates of 10-15% (higher after a gifting peak) mean the revenue Google reports is not the revenue you keep. And a category where half the year's demand lands in six weeks makes an annual average target actively harmful. The fix is arithmetic: work out contribution per gross order after returns, divide, and set the target against whatever conversion value you are actually sending Google.
Start from contribution, not gross margin
Break-even ROAS is gross order value divided by contribution per gross order. Not gross margin. Contribution after you have paid for returns you have not had yet.
Here is a demi-fine silver brand we would consider typical for the Indian market. AOV ₹5,000 including 3% GST, so net revenue is 5,000 ÷ 1.03 = ₹4,854.
- COGS (metal, making, stone-setting): ₹1,700
- Packaging and forward shipping: ₹180
- Payment gateway at 2% of ₹5,000: ₹100
Contribution before returns: 4,854 − 1,700 − 180 − 100 = ₹2,874, or 59% of net revenue. That number is what most founders quote when we ask about margin. It is also not the number you can bid against.
Now the returns. Assume 12% of orders come back. Each return costs you the contribution you booked (0.12 × 2,874 = ₹345 spread across every order) plus the handling: reverse plus re-forward logistics ₹260, re-polish and re-pack ₹150, gateway fee you never get back ₹100. That is ₹510. And about 15% of returned pieces cannot go back on the shelf at full price, so add 0.15 × 1,700 = ₹255. Total ₹765 per return, which is 0.12 × 765 = ₹92 per gross order.
Contribution per gross order after returns: 2,874 − 345 − 92 = ₹2,437. Break-even ROAS = 5,000 ÷ 2,437 = 2.05.
If you want to hold 15% contribution after ad spend, that is 0.15 × 4,854 = ₹728, leaving ₹1,709 for acquisition per order. Target ROAS = 5,000 ÷ 1,709 = 2.93. Call it 3.0.
One detail that trips people up: 2.93 is the target for reported ROAS, because the denominator in that sum is gross order value, which is what your Shopify pixel sends. The day you start pushing refund adjustments back into Google, reported value falls by roughly 12% and your target has to fall with it, to 2.93 × 0.88 = 2.58. Change one without the other and you will throttle a profitable account by a fifth.
COD is a bidding input, not a checkout preference
RTO on cash-on-delivery jewellery is the single biggest gap we see between reported and real ROAS in India. A ₹5,000 silver order that never gets accepted at the door costs you forward and reverse freight and returns nothing, and it still fires a purchase event with full value. Google then learns that the audience segment which loves COD is your best audience, and buys more of it.
Two things worth doing before you touch bids. Cap COD at a value where the RTO cost is survivable, or price it with a fee. And stop sending COD orders as conversions at full value: send them at your expected acceptance rate, or fire the conversion on delivery via an offline upload. Most brands can do the second in a week with a Flow trigger and a scheduled upload. It changes what Smart Bidding chases within a fortnight.
Why gold by weight breaks the model
Run the same sum on 22k gold and the channel stops making sense. Metal is a pass-through at spot rate. Your margin lives in making charges less whatever you discount off wastage, which on a competitive light-weight chain might leave 8-12% contribution. At 10%, break-even ROAS is 10x. Layer in a 6% return rate and you need about 10.6x reported before you have earned a rupee.
You will not get 10x from a Performance Max campaign fed a broad catalogue. What you can get is brand search at 12-20x, Shopping on specific high-making-charge SKUs, and store-visit or appointment campaigns if you have a showroom. That is a smaller, less exciting account, and it is the honest answer. Gold-by-weight brands that scale on paid do it by pushing studded and design-led lines, where making charge is a larger share of ticket price, and treating plain gold as the thing that gets bought after the first order.
The operational trap here is feed price mismatch. Gold moves daily; if your Merchant Center price lags your website price you get item disapprovals in the middle of Dhanteras week, which is the worst possible time to be debugging a feed. If your catalogue is priced off spot, automate it at the source. Our Gold & Silver Live Rate app repricing products as the rate moves is one way to keep storefront, feed and the rate-breakup table on the product page telling the same story about metal, making, wastage and GST.
Feeding returns back into the bidding
Google will optimise to whatever value you send. So send better values.
The clean version is conversion adjustments: when Shopify refunds an order, push a retraction or a reduced value against that order's transaction ID. It works, and it is the right build if you are doing meaningful volume. The catch is the delay. Jewellery returns cluster at day 10-25, occasionally later on gifting, so the adjustment lands well after the bidding decision it should have informed. On a store doing 40 conversions a month per campaign, adjustments arrive too thin and too late to steer anything.
Below roughly 100 conversions a month, a blended haircut is more useful than a perfect pipeline. Take your trailing 90-day return rate by product type, apply it as a fixed multiplier on the value you send, and re-derive it quarterly. Studded and engraved pieces return less than plain bands, because sizing is the top return reason and the customer who spends on engraving has thought harder. Split the multiplier by collection if the spread is more than about five points.
Gifting seasonality: three different accounts in one year
Jewellery demand in India is not seasonal in the retail sense of "Q4 is bigger". It is a set of discrete spikes with different buyer intent, different AOV and different return behaviour.
- Akshaya Tritiya and Dhanteras. Auspicious-purchase intent, high conversion rate, lower return rate, buyers who already know what they want. Conversion rate can double, which means Smart Bidding will spend more at the same target because predicted value per click rises. You do not need to slash the target. Halving tROAS on Dhanteras eve is how accounts end up buying "gold price today" traffic at ₹90 a click.
- Wedding season, roughly November to February and again around Akshaya Tritiya. Long consideration, multiple visits, highest AOV, heavy assisted-conversion behaviour. This is where a 30-day conversion window quietly under-credits Search and over-credits whatever touched last.
- Valentine's, Rakhi, Mother's Day, Christmas. Gifting proper. New customers, decent volume, and the worst returns of the year because the recipient did not choose the size, the finish or the piece. On a Dec-Jan cohort we would plan for a return rate several points above the annual figure, which pushes the correct target ROAS for December up even though the December conversion rate is telling you to bid harder.
That last point is the one most accounts get backwards. December looks like your best month on reported ROAS. January's refund wave belongs to it. If you review performance monthly on gross revenue, you will keep increasing December budget on a cohort that returns 18% of what it bought.
Practically: keep separate seasonal targets in a document, and move the live target in steps of no more than 15-20%, at least 10-14 days before the peak. Smart Bidding needs the runway. For GCC brands the same logic runs on Ramadan, Eid and DSF instead, with materially lower return rates and no COD problem to model — if you sell into that market, our Dubai team works the gold calendar there.
High AOV means you probably have too many campaigns
Target ROAS needs volume to work with. Somewhere around 30 conversions in 30 days per campaign is where the model stops thrashing. A brand with a ₹45,000 AOV doing three orders a week has 12 a month, spread over four campaigns, which is not bidding. It is noise with a budget.
What we do instead at that volume: one consolidated Performance Max or Shopping campaign with brand excluded, one brand Search campaign on manual or target impression share, and if there is a showroom, a local campaign. Optimise the top-of-funnel to maximise clicks with a hard CPC cap until conversion volume supports a value target. It feels unsophisticated. It outperforms four tROAS campaigns starving each other.
Also set your conversion window honestly. A 7-day window on a category where people take three weeks and four visits to spend ₹60,000 will hide most of your Search value. We usually run 60 days for jewellery, 90 for bridal.
New-customer value, and the one setting worth arguing about
If your repeat rate is real, bid for it. Performance Max and Search both let you assign additional value to new customers, and on jewellery that is a defensible lever: a first-time silver buyer at ₹5,000 who comes back twice in 18 months is worth about three times the single order. Take your actual 18-month repeat contribution, subtract the first order, and put that number in as the new-customer bonus. Do not put a made-up LTV in there, because you will be spending real money against it within a day.
The setting worth resisting: broad audience signals stacked into PMax on the theory that Google will figure it out. On high-AOV jewellery it figures out that the cheapest conversions are your own brand searches, then reports a 9x ROAS you already owned. Exclude brand terms, check the search terms report weekly, and judge PMax on incremental new-customer orders.
Landing page speed changes your bid maths
Nothing exotic here. Paid traffic to a jewellery PDP loading in 4 seconds on a mid-range Android over 4G loses a chunk of clicks you already paid for, and the loss shows up as a worse conversion rate, which Smart Bidding reads as a worse audience and prices accordingly. Heavy zoom galleries, 360-spin apps and three review widgets are the usual culprits. Getting LCP under 2 seconds on the templates your ads actually land on is normally cheaper than a 10% bid improvement, and it compounds across every channel. We cover the mechanics in Shopify speed optimization.
Where to start this week
Pull your last 90 days of orders and refunds from Shopify, split by collection, and build the contribution-after-returns figure for your top three product types. You will likely find one collection whose true break-even ROAS is nearly double the account average, and it is probably the one your PMax spends the most on. Fix the target for that collection first, before touching anything else.
If you want a second pair of eyes on the account structure and the feed alongside it, our free audit covers both. We are a Google Partner and a Meta Business Partner, so the review looks at how the two channels are double-counting each other as well.