The budget threshold below which a market test proves nothing

A market test proves nothing until it has produced enough conversions to tell a result from chance. That number is set first. From it you derive the clicks, then the budget, then the duration. If the amount you land on exceeds what you can commit, narrow the question, not the budget.

On this page

Why half a budget doesn’t give half an answer

This is the most common trade-off when opening a market, and the most costly. A test is presented at $7,000, and the owner offers $2,000. The intent is sound, the mechanics don’t follow.

An advertising test cannot be scaled down like an order. It produces a single number, an observed conversion rate, which only becomes readable above a certain volume. Below that volume, the gap between two cities comes from chance as much as from the market.

The platforms say this in their own way. Google Ads recommends evaluating performance over the last 30 days with at least 30 conversions. Meta indicates that an ad set leaves its learning phase, the period during which the system works out who to show the ads to, after roughly 50 optimization events over seven days. Delivery is less stable there and the cost per result is higher.

Nothing stops a campaign from starting without history, as Google notes. But a campaign that never reaches those volumes stays in the most expensive part of its life.

The backward calculation, in four steps

The sequence reads from the end to the beginning. You start from the proof to be produced, and the amount lands on the last line.

The formula on a worked example

The two values that follow, the conversion rate and the cost per click, are assumptions made by the business and set out for the example. They come from no source. Only the benchmarks of 30 and 50 are documented.

An SMB wants to know whether its offer sells in Ontario. It targets 40 conversions, above the benchmark of 30, to give itself some reading margin. It assumes a conversion rate of 2%. It therefore needs 2,000 clicks. It estimates the cost per click at $3.50. The floor budget comes to $7,000.

That leaves the duration. Spread over six weeks, those $7,000 come to about $1,170 per week, close to seven conversions. On paid search, where the benchmark covers 30 days, that pace holds.

On Meta, the same test changes scale, because the benchmark there is weekly. It would take 2,500 clicks and $8,750 every week. This is why many social tests optimize on an event more frequent than a sale.

There is nothing sophisticated about the calculation, and that is its strength. The amount is not a management decision, it is a consequence of the question being asked.

When the budget doesn’t reach the threshold

This is the normal situation, not the exception. An SMB's means are almost always smaller than its market ambition.

Three exits exist. A fourth, spending anyway, is not one.

Narrow the question being asked

This is the most productive exit and the least practised. The number of conversions required doesn’t depend on the size of the market, but on the number of things you are trying to learn at once.

A test that puts one offer in front of one segment in one city needs a single batch of 40 conversions. Run across three cities, it calls for three batches, so three times the floor. You don’t need a tripled budget, you need one question to pay for first.

Narrowing doesn’t mean aiming small. It means accepting a first answer that is narrow and solid rather than broad and unreadable. Sprinkle the same amount over six combinations and you get six numbers that all look alike.

Switch levers for a cheaper signal

The threshold rests on two movable variables, the cost of a click and the definition of the conversion.

A less contested engine can deliver the same signal for a cheaper click, especially in B2B where the professional audience sits elsewhere. That is a change of platform, not a lowering of the standard.

The second variable is more delicate. A more frequent event, such as a sign-up or an add to cart, reaches the required volume far faster than a sale. But a test proving that people sign up doesn’t prove that they buy, and that link gets verified on your current market.

Postpone rather than buy an unreadable result

This is the exit that goes down worst in a meeting and costs the least. Postponing is not giving up, provided the missing amount and the review date are written down.

That delay is not dead time. You use it to fix conversion tracking and rebuild the landing pages. A test launched six months later on reliable measurement costs less than a test launched tomorrow on false measurement.

Before committing to a market test

The content of the test is covered in the campaign that learns first, the setting that skews the denominator in a badly configured geographic target. For a cheaper signal, see the second search engine in B2B and how long a social test takes. Cost benchmarks are in what advertising asks of an SMB, the profit actually banked in from announced return to real return, measurement in the gaps that skew tracking. The amount is framed in the benchmarks of an annual budget, the opposite decision in the stop criterion for a territory.

The delay to expect before judging is set out in how long before the first sale, and the ability to answer the inquiries generated in response coverage in a new market.

Frequently asked questions about market test budgets