Why a daily amount says nothing
An ad test for an online shop answers one question: does an order from this channel cost less than the shop earns on it. To answer, you have to measure the cost per order: spend divided by the number of purchases. And the precision of that measurement depends on the denominator.
Three purchases for UAH 900 give a cost per order of UAH 300. But under the same conditions the next UAH 900 may bring one purchase or six, and that is ordinary spread, not a change in the ads. The advice "test for a week at UAH 300 a day" takes into account neither your cost per order nor the number of purchases that amount will bring. It is the same for a shop with a UAH 300 average order and one with a UAH 15,000 average order, so it is useless for both.
So the test budget is calculated backwards: first the allowable cost per order, then the number of purchases needed for a conclusion, and only then hryvnias and days.
Step 1. Allowable cost per order
The allowable cost per order is what the shop can spend on advertising for one purchase without going into the red. It equals the profit from one order before advertising: average order value minus cost of goods, delivery, payment fees, losses on refusals and returns. How to calculate it for your shop, with a calculator, is in the article on online shop unit economics.
In the examples below the allowable cost is a hypothetical UAH 300. If an order from advertising costs less, the first purchase is already profitable; if more, the channel only makes sense if customers come back. This article calculates the test from the first purchase. If your business lives on repeat sales, the target can be raised, but then calculate it from actual profit per customer over a year, not from hope. Why cost per lead and cost per customer are different numbers is in the article on customer acquisition cost.
Step 2. How many purchases give a conclusion
The number of purchases fluctuates randomly under identical conditions. How much can be calculated. The table below shows the range in which the true cost per order lies if in the test it came out at exactly UAH 300.
| Purchases in the test | Spent, UAH | True cost per order, UAH | Spread |
|---|---|---|---|
| 5 | 1,500 | 129–924 | −57% … +208% |
| 10 | 3,000 | 163–626 | −46% … +109% |
| 20 | 6,000 | 194–491 | −35% … +64% |
| 30 | 9,000 | 210–445 | −30% … +48% |
| 50 | 15,000 | 228–404 | −24% … +35% |
| 100 | 30,000 | 247–369 | −18% … +23% |
Five purchases for UAH 1,500 mean the true cost per order could be 130 or 920. Such a test shows nothing except that purchases are possible at all. Thirty purchases narrow the spread to −30 … +48%, and that is enough to decide if the result is far from the target.
The second question is how many purchases you need to say with confidence that the channel beats the target, if the test cost per order came out below UAH 300.
| Cost per order in the test | Purchases for a confident conclusion | Test budget |
|---|---|---|
| UAH 150 (half the target) | 12 | UAH 1,800 |
| UAH 180 (40% below) | 19 | UAH 3,420 |
| UAH 210 (30% below) | 36 | UAH 7,560 |
| UAH 240 (20% below) | 87 | UAH 20,880 |
| UAH 270 (10% below) | 365 | UAH 98,550 |
The closer the result is to break-even, the more expensive it is to prove it is on the right side. A channel that delivers orders 10% cheaper than target cannot be told apart from one that breaks even without spending almost UAH 100,000. The practical conclusion: a test should look for channels that are clearly better than target, not prove that a channel is "slightly profitable".
Stopping rule: when a channel is definitely worse than target
The opposite conclusion, "this channel will not pay off", is cheaper. If there are no purchases or almost none, after a certain amount of spend you can stop with the same confidence.
In multiples of the allowable cost per order: 0 purchases — 3.69; 1 — 5.57; 2 — 7.22; 3 — 8.77; 5 — 11.67. If this spend brings this many purchases or fewer, the true cost per order is above target with a probability of at least 97.5%.
For your shop, multiply the coefficient by your allowable cost per order. Zero purchases after spending 3.7 times that cost is a signal to stop, not to "give it another week". One or two purchases after 6–7 times — the same.
One caveat: at the start of a campaign the platforms are learning, and orders in this period are more expensive than they will be later. So the stopping rule is applied to a campaign that has a chance to exit learning (see below), and purchases are counted allowing for delay: someone who clicked on Monday may buy on Thursday.
What Meta and Google need for a campaign to learn
Statistics is not the only constraint. Ad systems decide themselves whom to show an ad to, and they need data for that.
- Meta (Facebook, Instagram). According to Meta's help page on the learning phase, an ad set usually exits learning after about 50 results in the week after its last significant edit. While learning lasts, the cost per result is higher and unstable, and an edit restarts learning. If you optimise for purchases with an allowable cost of UAH 300, 50 purchases a week means about UAH 15,000 a week. For a small shop that is often more than the whole test budget.
- Google Ads. According to Google's help page on the learning period, an automated bid strategy typically needs one or two conversion cycles, that is, the time from click to purchase; manual CPC has no learning period. According to the help page on average daily budgets, a campaign can spend up to two daily budgets on a single day, but no more than 30.4 daily budgets in a month.
What this means for the budget. If 50 purchases a week is out of reach, Meta offers two ways out: combine ads into one ad set so that results are not split across several, or optimise for an earlier event than purchase (add to cart). But the test still has to be judged on purchases and cost per order: cheap add-to-carts say nothing about payback. In Google, set the test's daily budget from the total test amount, not from "what we can spare per day": a campaign with a UAH 300 daily budget can spend 600 on some days.
Formula and example
The calculation for a hypothetical shop with an allowable cost per order of UAH 300:
| Step | How to calculate | Example |
|---|---|---|
| 1. Allowable cost per order | Profit from one order before advertising | UAH 300 |
| 2. Purchases for a conclusion | 30, a spread of −30 … +48% | 30 |
| 3. Test budget | Step 1 × step 2, over two to four weeks | UAH 9,000, ≈ UAH 640 a day over 14 days |
| 4. Stopping rule | 3.7 × step 1 with no purchases at all | UAH 1,107 |
Thirty purchases is a compromise: that spread is enough to tell a clearly good channel from a clearly bad one. Results close to target need more, as the table above shows.
How to calculate it for your shop:
- Calculate the allowable cost per order: profit from one completed order before advertising.
- Multiply by 30. That is the test budget if the channel works roughly at target. If it is better, 30 purchases will come cheaper; if worse, the stopping rule kicks in.
- Divide by the number of test days. Two to four weeks, to account for differences between weekdays and the delay between click and purchase. UAH 9,000 over 14 days is about UAH 640 a day.
- Write down the stopping rule before the start: "zero purchases after UAH 1,107, we stop". A rule written before the start does not have to be debated once the money is spent.
The test requires purchases to be counted correctly: a purchase event with an amount must reach your analytics from the ads, not just button clicks. How to check this is in the article on end-to-end analytics.
What to do after the test
- Did the stopping rule trigger?As many purchases as in the chart above, or fewer, at the matching spend. Yes → the channel is worse than target; change the offer or the landing page, not the budget
- Is the cost per order clearly below target?For example UAH 150–180 against a target of 300, with at least as many purchases as the second table requires for that result. Yes → scale gradually, watching the cost per order
- Is the result close to target?Cost per order within ±20% of the allowable cost. A test of this size will not tell whether the channel is profitable. Yes → improve site conversion or average order value, rather than buying more statistics
The third case is the most common and the most expensive to get wrong. Extending the test to "gather statistics" will take tens of thousands of hryvnias. It is cheaper to change something that moves the result well away from the line: the delivery price, the average order, the conversion of the product page. How to choose between ads and SEO for testing demand is in the article SEO or Google Ads.
Our position
Do not plan a test from a daily amount. Plan from 30 purchases at the allowable cost per order and a written stopping rule. If the budget does not cover 30 purchases, it is better not to run the test: a smaller one will show spread, not a result, and the decision will be made by chance.
This position does not hold for very expensive goods with rare purchases, such as furniture or appliances costing tens of thousands of hryvnias, where 30 purchases in a reasonable time are out of reach. There the test is built on an intermediate event reliably linked to purchase in your shop: a call or a lead with a known share that reaches payment. That share has to be measured on your own data first, not taken from someone else's case study.
The Meta and Google help pages were checked on 13 October 2026. The ranges in the tables and the chart were calculated using the Poisson distribution with exact 95% limits (one-sided 97.5% for the stopping rule and the "channel is better" table), assuming purchases are independent and their rate is stable during the test. For a different allowable cost per order all amounts scale proportionally: the percentages and multipliers do not depend on it.