When to kill a campaign, and when it just needs a week

There is a piece of folk wisdom that goes: nobody ever regrets killing a campaign early. It sounds like discipline, and it is popular because the money you did not spend is easy to feel good about. It is also wrong often enough to be worth arguing with. Stopping too early and feeding too long are both mistakes, and the reason the first one feels virtuous is simply that its cost never shows up on a statement. You do not get an invoice for the channel you abandoned in week one because it had not warmed up yet.

So there are two ways to be wrong here, and they fail in opposite directions. Feeding a campaign that already answered you is a slow leak — obvious in hindsight, painful to admit, and at least it is visible while it happens. Killing a campaign that had not answered yet is invisible: nothing bad appears anywhere, the spend stops, and you file away a conclusion about a channel that you never actually tested. The second mistake is more common precisely because it costs nothing you can see, and because it comes with a story about being decisive.

The question underneath both is one question: has this campaign said anything yet? A campaign talks to you in two very different voices, and they arrive on completely different schedules. The top of it — what a click costs, how often people take one — settles within a few hundred impressions, which is days. The bottom of it — whether visitors sign up, whether signups pay — is built from a handful of events and stays noisy far longer than anybody wants it to. We have written before about how a first hundred dollars buys you a real opinion about your ad and almost none about your product. The kill decision runs on the same split: you may already have enough to judge the ad, and nowhere near enough to judge the offer.

Put numbers on it, invented but plausible. Two hundred dollars at a dollar fifty a click is roughly a hundred and thirty visits, and suppose you got two signups. That is one and a half percent, and it looks like a failure sitting next to the three percent you were hoping for. But if the true rate really is three percent, a hundred and thirty visits has an expected value of about four signups, and landing on two — or on one, or on seven — is completely ordinary variation. Run the identical campaign again next week and you might get five, having changed nothing. Two signups is not evidence that the page converts badly. It is not evidence of anything. Killing on that number is a coin flip you will remember as a judgment.

a campaign that has not spoken yet is not failing. it is just quiet, and quiet is easy to mistake for an answer.

Which is not an argument for patience in general, because some kills are correct on day two and cost you nothing to make. Three signals settle fast enough to act on early. If a click costs three or four times what your economics could ever support, no amount of waiting negotiates the auction down, and that result arrived cheaply — take it. If almost nobody clicks, the ad is the problem rather than the market, and running it longer only buys more evidence of the same. And if the seam is broken — the ad promises one specific thing and hands people a generic page that never mentions it — you are not testing a campaign at all, you are paying to watch a mismatch. Fix or stop, but do not wait it out.

Then there is the cost most people never count, which is thrash. Ad platforms need a stretch of stable delivery before their targeting settles, and a meaningful edit — new audience, big budget change, swapped creative — sends that back to the start. A campaign that is adjusted every second day is not a campaign that has run for three weeks. It is ten campaigns that each ran for two days, none of which finished a sentence, all of them billed to you at the beginner rate. If you are going to change something, change one thing, and then leave your hands off it for a defined stretch. The tinkering feels like work. It is mostly a way of never running the test.

Judge the result on your number, not the platform's. The ad account will report conversions generously — it counts people who saw the ad and arrived later by another route, over attribution windows it chose — and those numbers are useful to the platform's own optimization and misleading as a verdict on your spend. What matters is what happened in your own house: visits, signups, revenue, sitting on the same screen so the comparison is a glance rather than an afternoon of reconciling two exports. When the platform says twelve conversions and your side of the glass says three, three is the number your bank account agrees with.

The genuine case for one more week is narrower than optimism wants it to be, but it is real. A campaign that has been running a few days has usually not seen a full weekly cycle, and weekday and weekend audiences can behave like different countries. A campaign whose cost per click is falling as delivery settles is mid-sentence. A campaign that is expensive but converting is a pricing problem, not a viability problem, and a week of data tells you which lever to pull. What does not earn another week is a campaign whose top-of-funnel numbers are plainly bad and unchanged — that one has spoken clearly, and waiting is just a way of not hearing it.

All of this is far easier if the decision was made before the money moved. Campaigns we build arrive paused and spend nothing until a person starts them, and the quiet minute before starting is the right time to write down two lines: the number that would make this a keep, and the number that would make it a kill, with the amount of spend that has to happen before either line counts. Ten minutes of that beforehand is worth more than any amount of staring at a chart later, because once real money is moving you will read the same chart as hope on Monday and panic on Thursday.

So: kill on a number you wrote down in advance, and wait on a number you did not. Everything else — the folk wisdom, the tinkering, the conversions the platform is happy to credit itself with — is a way of making a decision feel earned when it was really just made on a Tuesday, by a mood.