There is no magic number, because ad spend is an output of your economics, not a fixed budget. Two sane ways to set it: top-down, as a slice of the revenue you put into marketing; and bottom-up, from your unit economics, spending as much as you can while each new customer is still acquired profitably (while ROAS stays above your break-even ROAS). Start small, prove it is profitable, then scale in steps.
“How much should I spend on Google Ads?” is the most common budget question and the one with the least satisfying honest answer: as much as you can spend profitably, and not a euro more. That sounds like a dodge, but it is actually the whole method. Here is how to turn it into a real number for your store.
Why a fixed budget is the wrong starting point
Picking a round number first (“we'll do €3,000 a month”) and hoping it works is backwards. Two stores can spend the same and get wildly different results because their margins, prices and conversion rates differ. Spend should fall out of your economics, so the right question is not “what budget?” but “how much can I spend while still making money on the marginal sale?”
Two sane ways to size it
1. Top-down: a share of revenue
As a planning sanity-check, many eCommerce brands run total marketing at roughly 10 to 30 percent of revenue, with Google Ads being one slice of that alongside Meta, email and the rest. This keeps you in a reasonable range and stops you either starving growth or blowing the budget. It is a guardrail, not a target.
2. Bottom-up: your unit economics
This is the one that actually decides the number. Work out what you make on an order and how much you can pay to win a customer while staying profitable:
- Your break-even ROAS is 1 divided by your gross margin (a 40% margin breaks even at 2.5x). Below it you lose money; above it you profit.
- As long as new spend keeps producing sales above that break-even line (with a buffer for shipping and returns), spending more makes you more total profit, even if ROAS drifts down a little.
- The moment extra spend pushes ROAS toward break-even, you have found your profitable ceiling for now.
A sensible starting budget
When you are just beginning, the goal is not scale, it is proof. Start with enough budget to generate a meaningful number of conversions per week (so bidding has data), aimed at your clearest demand first: your brand terms and your best high-intent Shopping products. Prove that converts profitably, then widen into less certain territory. Starting narrow and profitable beats starting broad and hoping.
The minimum that actually works
There is a practical floor, and it is about conversions, not euros. Smart bidding learns from a steady flow of conversions; a budget so small it produces only a handful a week will look erratic and never really settle. So the minimum is less “X euros a month” and more “enough to buy enough conversions in my niche for the algorithm to learn.” In expensive, low-volume niches that floor is higher than in cheap, high-volume ones.
When and how fast to scale
Scale while ROAS stays comfortably above your break-even line, and do it in steps, not jumps. Big, sudden budget increases reset the learning and can spike your costs for a week or two. Raise budgets gradually, let performance settle, and keep checking that profitability holds. When ROAS starts sliding toward break-even as you add spend, the market is telling you where your profitable ceiling sits, at least until you improve margins, conversion rate or creative and lift the ceiling.
Do not forget the other costs
Ad spend is not the whole cost of the channel. Budget also has to cover management (whether that is an agency, a freelancer or your own time, see our guide on what management costs) and creative (images and video for Shopping and Performance Max). A realistic plan treats those as part of the marketing line, not an afterthought.
Not sure whether your current budget is under-fed, over-extended, or spent in the wrong places? A free audit looks at your margins, your ROAS and your spend, and tells you where your profitable ceiling actually is.
