Facebook Ads Testing Budget: How Much to Spend

"How much should we put into the first test?" is the question we hear most from store owners who have never run Meta ads, or who ran them once, spent $300 and decided "Facebook doesn't work for us". Both answers usually come from the same mistake: the budget was picked from a gut feeling, not from what the test needed to prove. This guide shows how we size a first ad test for an online store, with the arithmetic, three worked examples and the stop rules we agree with clients before a single dollar is spent.

What a test budget actually buys

A first test does not buy sales. Some sales will come, but what you are really paying for is an answer to a short list of questions: which angle makes strangers buy, which audience responds, and roughly what a customer costs. If the budget is too small to answer those questions, the money is not "a cheap test". It is simply lost, because you finish with numbers you cannot trust.

So the right question is not "what can we afford per month" but "how many results does each thing we compare need, and what will those results cost?" Answer that and the budget falls out of a simple formula.

The formula we use

Test budget ≈ number of variants × results needed per variant × expected cost per result

Three inputs, each worth a minute of thought:

Number of variants. Everything you want to compare head to head: angles, audiences, offers. Three angles in one audience is three variants. Three angles in two audiences is six, and it doubles the budget.
Results needed per variant. The number of purchases (or another event, see below) each variant needs before you trust its cost. We plan for at least 10, and 15 to 20 when the money allows.
Expected cost per result. Your honest guess at what a purchase will cost in week one. Start from your break-even cost per acquisition and assume the first week runs above it, because nothing is optimized yet.

If you do not know your break-even number yet, work it out first. Our break-even ROAS guide walks through it with a worked example; it takes about twenty minutes with your last month of orders.

Why 10 purchases, not 2

Small numbers lie. A useful rule of thumb from basic statistics: when you count random events such as purchases, the typical swing in the count is about the square root of the count. Relative to the result, that is roughly 1 ÷ √n.

Purchases per variant	Typical noise in cost per purchase	What you can conclude
2–3	±60% or more	Almost nothing. Luck dominates.
5	about ±45%	Spot a disaster, not a winner
10	about ±30%	Separate a clear winner from a clear loser
20	about ±22%	Compare two decent variants with some confidence

In plain terms: if angle A got 3 sales at $25 each and angle B got 2 sales at $38, you have not learned that A is better. One extra sale for B on day six would flip the result. With 10 or more each, a gap of 40% or more starts to mean something. This is also why spreading a budget across twenty ads fails: every one of them stays in the "luck" row.

There is a platform reason too. Meta's delivery system learns from the event you optimize for, and Meta has long said an ad set needs around 50 of those events in a week to leave the learning phase. Exact thresholds and wording change, so treat that as direction, not a rule. Most small stores will not reach it in a first test, and that is fine. It just tells you that fewer, better-funded ad sets learn faster than many starved ones.

When purchases are too expensive to count

For a $300 product with a $120 cost per purchase, 10 purchases across 3 angles is a $3,600 test. Many stores cannot or should not risk that on a first round. The answer is not to give up on testing; it is to read the result one step higher in the funnel.

We still optimize the campaign for purchases whenever volume allows, because optimizing for clicks or add-to-cart tends to find people who click and add to cart. But we judge the early rounds on cheaper signals that appear earlier:

Click-through rate and cost per landing page view tell you whether the angle stops people at all. You get hundreds of these quickly.
Cost per add to cart tells you whether the product page and price survive contact with the visitor.
Cost per initiated checkout is the closest cheap signal to a sale, and it is where shipping costs and surprise fees show up.

Plan for 10 or more of whichever event you will judge by. A variant can lose on clicks within two days and be cut early; a variant that wins on checkouts earns the budget to prove itself on purchases in round two. Just do not crown a final winner on proxy events alone. We have seen angles with great add-to-cart numbers that converted poorly, often because the ad promised something the product page did not.

Want us to size your first test on a call? Book 30 minutes with Nick. You leave with a concrete 7-day ad test, whether you hire us or not.

Book a free call →

Three worked examples

The numbers below are illustrative, not client results. Swap in your own order value and margin.

	Snack brand	Skincare brand	Premium equipment
Average order	$40	$85	$420
Break-even CPA	$15	$40	$140
Expected week-one CPA	$20	$55	$180
Variants	3 angles	3 angles	2 angles
Judged on	Purchases (10 each)	Checkouts first, then purchases	Checkouts (10 each)
Test budget, 7 days	~$600 (~$85/day)	~$900 (~$130/day)	~$800 (~$115/day)

Snack brand. Purchases are cheap enough to count directly: 3 × 10 × $20 = $600. Seven days at about $85 a day. If the week ends with fewer purchases than planned, that itself is information: the real cost per purchase is higher than assumed, and the next step is a better offer or bundle, not more spend.

Skincare brand. Counting purchases would need 3 × 10 × $55 = $1,650. Instead we plan the first week around initiated checkouts, say $30 each in this example, so 3 × 10 × $30 = $900. Purchases still get tracked; the best angle by checkouts then gets a second week focused on confirming the cost per purchase.

Premium equipment. Here even checkouts cost real money, so we cut the comparison to two angles: at about $40 per initiated checkout in this example, 2 × 10 × $40 = $800. The trade-off is a slower answer. With very few sales a month, retargeting site visitors and capturing emails often matter as much as the cold test itself.

If you run a local service and sell through leads rather than a cart, the same logic applies with leads as the result. Our guide on how much a small business should spend on ads has a table for that case.

Daily budget per ad set, and how long to run

Two rules of thumb keep the test readable:

Give each ad set at least your expected cost per result per day. At a $20 expected CPA, an ad set on $8 a day can go several days without a single purchase, and you will not know whether the angle failed or simply did not get enough delivery.
Run a full 7 days. Weekdays and weekends behave differently, and the platform needs a few days to settle. We do not touch budgets or ads for the first three days unless something is broken, because edits can restart learning.

Two more details people miss. First, purchases are often reported with a delay because of attribution windows, so look again two or three days after the test ends before final decisions; check the attribution setting in your own account, as Meta's defaults have changed over the years. Second, compare Meta's numbers with your shop's orders. If Ads Manager reports 30 purchases and Shopify shows 18 from that traffic, your tracking needs fixing before your budget does.

Agree the stop rules before you launch

The single best protection for a test budget is writing down the decision rules in advance, when nobody is emotionally attached to an ad. Ours look like this:

Cut early: a variant that has spent two to three times your target cost per purchase with no purchases, or whose click-through rate sits far below the others after a couple of days, is switched off.
Hold: a variant near break-even, or with too few results to judge, gets more time or a fresh creative in the same angle.
Promote: a variant below target with at least the planned number of results, confirmed by shop orders, gets the budget in round two.
Cap the total: the test has a maximum spend. If you reach it without a promoted variant, you stop and change the offer, page or product, not the ads.

Rule four is the one that saves stores. When no angle works, the problem is rarely the targeting. It is usually the price, the shipping cost, the product page or the product itself, and no amount of extra ad spend fixes those.

What happens after round one

A good first test ends with one or two promoted angles and a list of things you learned about your buyer. Round two keeps a small fixed budget for new variants, often 10 to 20 percent of spend, while the winners get the rest and grow gradually. This is the same loop we describe step by step in our e-commerce testing playbook.

It is also how larger accounts are built. For Biocoiff, an organic hair-care brand, we tested 28 campaigns, cut 9, and scaled the best 4 by about 20% a month, reaching up to 8x ROAS on purchase campaigns. None of that started with a big budget; it started with tests sized to give honest answers.

Full breakdown: Biocoiff case study.

Your test budget worksheet

Write down your break-even cost per purchase.
Estimate week-one cost per purchase at 20 to 40 percent above it.
Choose 2 to 3 variants. Not more.
Pick the event you will judge by: purchases if 10 per variant is affordable, otherwise initiated checkouts or add to cart.
Multiply: variants × 10 × expected cost per event. That is your test budget; divide by 7 for the daily budget.
Check that each ad set gets at least one expected result's worth of budget per day.
Check that tracking matches your shop orders.
Write the four stop rules and the spending cap down before launch.

If the number on line five is more than you are comfortable losing, reduce the variants, not the results per variant. A smaller test that answers one question beats a bigger one that answers none.

More on how we run tests for online stores: e-commerce ads.

FAQ

How much should I spend on my first Facebook ads test?

Multiply the number of variants you want to compare by about 10 results each and by your expected cost per result. For a store with a $20 expected cost per purchase testing 3 angles, that is around $600 over 7 days. If that is too much, test fewer variants rather than spreading the same money thinner.

How long should a Facebook ad test run?

We run a full 7 days so weekdays and weekends are both included, avoid edits for the first three days, and look at results again a couple of days after the end because purchases can be reported with a delay.

What if my product is too expensive to get 10 purchases in a test?

Judge the first round on a cheaper event that comes earlier in the funnel, such as initiated checkouts or add to cart, and cut obvious losers on click-through rate. Then give the best variant a second round to confirm its cost per purchase before scaling.