When to stop a marketing campaign: decide before launch

You are looking at a campaign that has spent enough to be annoying, but not enough to make the answer obvious.

One more week might rescue it. One more week might also buy you a slightly more expensive version of the same bad news.

This is where founders usually ask: How long should a marketing campaign run before I stop it?

The useful answer is not three days, three weeks or three months. The useful answer should have been written before the campaign began.

Every marketing bet needs two boundaries: a minimum amount of evidence before you judge it, and a maximum amount of exposure before it must earn another round. Without both, “give it more time” becomes a recurring invoice.

The decision gets worse after you spend

Before launch, you can be reasonably calm. You have a hypothesis, a budget and several other things competing for attention.

After launch, the bet becomes personal. Someone argued for the channel. Someone wrote the copy. You have already spent money, held meetings and checked the dashboard far too often. Stopping now feels like admitting the work was wasted.

So the standard quietly moves.

The original goal was qualified enquiries. Then the campaign produced clicks, so clicks became encouraging. Then it produced a few weak leads, so the conversation moved to “learning”. Nothing has technically failed because nothing was defined well enough to fail.

Founder time makes this worse. In a recent discussion about moving from organic to paid marketing, one founder put it neatly: “My hours are the budget, and they’re maxed.” Free distribution is not free when the owner is the production system.

The answer is not to become ruthless with every weak number. It is to remove as much improvisation as possible from the decision.

A stop condition is a decision contract

A stop condition is not “pause if ROAS is below two”. That may be one threshold inside it, but a useful condition says more:

• what you believe;
• what business result would support that belief;
• which early signal is worth watching;
• how much evidence you need before judging it;
• the most money and time you will expose;
• when the decision will be made;
• who can make it.

Write those points before the first euro or founder hour is spent. That is the contract.

It protects the budget, but it also protects a decent bet from a nervous Tuesday afternoon.

Set an evidence floor and an exposure ceiling

These are different controls. You need both.

The evidence floor is the earliest point at which the result is credible enough to influence a decision. It could be a number of qualified conversations, a full buying cycle, a defined volume of searches, or enough traffic to observe a meaningful conversion pattern.

The exposure ceiling is the most you will risk before the bet must justify more. Count cash, founder hours, team capacity, opportunity cost and any reputational risk.

If you only set the ceiling, you can kill a good idea on noise. If you only set the floor, an inconclusive bet can keep feeding indefinitely.

Advertising platforms make this distinction in their own way. Google Ads sets initial durations of 2–12 weeks for several experiment types and advises advertisers to keep the rules stable between the test groups. Its reporting also explains that a result may remain unclear because the test lacked time, traffic or a large enough split—not because the idea was proved bad. (Google Ads experiment guidance, reporting guidance)

Optimizely uses product-specific minimums too. For binary metrics, it requires at least 100 visitors or sessions and 25 conversions in both the variation and baseline before it declares a winner. Those numbers are not targets for your business. They are evidence that even tools built for testing refuse to call a result before their own floor is met. (Optimizely statistical significance guidance)

Your evidence floor will depend on the bet. A local service business testing a new offer has a different data problem from a high-volume shop testing a checkout button. False precision is still false.

Write these seven fields before launch

Keep the document short enough that people will actually use it.

1. The belief

Write one sentence that can be wrong.

We believe owners of multi-location clinics will respond to a fixed-scope website assessment because they suspect their current site is losing enquiries.

“We should try LinkedIn” is not a belief. It is an activity looking for a reason.

2. The business outcome

Name the result that would matter outside marketing. Qualified enquiries, booked assessments, accepted proposals, repeat purchases or recovered staff time all count. Impressions do not become revenue because the chart is colourful.

3. The early signal

Some outcomes take too long to observe directly. Choose an earlier signal with a credible connection to the outcome: visits from the right search terms, replies from the target buyer, completed diagnostic forms or sales conversations that mention the tested problem.

An early signal earns more observation. It does not get promoted to revenue.

4. The evidence floor

Define what must happen before the team is allowed to judge the bet. Use the buying cycle and available volume, not a borrowed internet rule.

For low-volume work, this may be a set of ten well-targeted conversations and the objections heard in them. For a paid campaign, it may be enough spend to give the target acquisition cost a fair test, across a full conversion delay. For content, it may be a consistent publishing window plus evidence of the right readers arriving.

5. The exposure ceiling

State the maximum in plain units:

• cash;
• founder hours;
• team hours;
• calendar time;
• customer or brand risk.

One consultancy recommends reserving 5–10% of the marketing budget for controlled experimentation, with a hypothesis, owner, cap, metric, timeline and stop-loss rule. Treat that percentage as a reference point, not a commandment. The useful part is separating experimental money from the core engine and capping it before launch. (Pedowitz Group)

6. The decision date

Put the review in the calendar now. “We monitor continuously” usually means everybody watches and nobody decides.

The date can include an exception: review earlier if the exposure ceiling is hit, tracking breaks, or the activity creates a material customer risk.

7. The owner and possible next states

One person owns the call. Their options are:

1. Scale because the result cleared the success condition.
2. Continue because the evidence floor has not been met and the ceiling has room.
3. Change one variable because the diagnosis points to a specific weakness.
4. Pause because the environment or measurement is unreliable.
5. Stop because the hypothesis lost or the likely upside no longer justifies the exposure.

“Keep tweaking” is not a sixth state.

Three bets, three different stop conditions

The framework stays stable. The numbers should not.

Notice what is missing: a universal number of days.

A direct-response ad can produce an early commercial signal. A positioning change may need repeated exposure and sales conversations. A search-led article can take time to be discovered. Treating them as the same test makes the spreadsheet tidy and the decision useless.

Do not confuse a losing hypothesis with a broken test

A clean test can show that the idea was wrong. That is a useful result.

A broken test tells you very little. Tracking failed. The offer changed halfway through. Sales did not follow up. Three variables moved at once. The target audience never saw enough of the work. Calling that a failed channel is convenient, but inaccurate.

At the review, ask two questions in this order:

1. Was the test credible? Check execution, tracking, audience, timing and whether the agreed evidence floor was met.
2. Did the belief hold? Compare the result with the success and stop conditions written before launch.

If the first answer is no, fix the test only when the remaining upside justifies another capped round. Poor setup does not create an automatic right to more budget.

The 15-minute pre-launch ritual

Before the next marketing activity begins, open a blank page and complete this:

We believe: The business outcome is: The earliest useful signal is: We will not judge it before: We will not expose more than: We decide on: The decision owner is: If it stops, the learning we keep is:

If the team cannot complete the page, it is not ready to spend. It may be ready to research, interview customers or fix measurement. That is different work.

Store the brief next to the work and its results. A system such as SharpOS makes the useful part visible: the original decision, the work in progress and the evidence used at review. The point is not another dashboard. It is stopping the rationale from being rewritten after the result arrives.

Marketing patience should be chosen, not improvised

Good marketing often needs time. That is not an argument for open-ended activity.

Choose the time. Name the signal. Cap the exposure. Then let the bet run without a founder changing direction every time the chart moves.

The discipline cuts both ways: weak work loses the right to continue, and credible work gets the room it was promised.

If your marketing still depends on late-night judgement and moving targets, talk to SharpHaw. We build and run the website, advertising, content and automations inside one visible operating rhythm—so the next decision is already clear before the spend begins.