Why Your Campaign's Worst Week Is Often the Week It Started Working

Quick Takeaways

1. A campaign that climbs hard and then falls for three days is usually behaving exactly as designed. The warm audience gets used up first, and what follows is the campaign reaching people who have never heard of you.
2. When Harvest Bloom's revenue dropped, the ads were reaching four and a half times more people than the week before. The drop was a side effect of growth, not a fault.
3. Changing the audience, swapping the creative or moving the budget all reset the learning that was already underway. The fix can cost you more than the problem.
4. Before you touch anything, check reach, frequency, audience temperature, sample size, tracking and the day of the week. Six checks, in that order.
5. A zero on tiny numbers is meaningless. An ad shown to 26 people at Harvest Bloom's normal conversion rate would be expected to produce 0.06 sales, so zero tells you nothing at all.
6. AI dashboards will hand you a confident looking zero and let you act on it, because nothing in them knows that a number can be too young to mean anything.
7. Patience and neglect look identical from the outside. The only thing separating them is whether somebody is actually watching.

The dip is where most campaigns die

Almost every campaign I have inherited from another agency died at the hands of a person having a bad Sunday night. Somebody looked at a chart, felt sick, and changed something. Then they changed something else on Tuesday, and within a week the account had been reset so often that nothing in it had ever been given long enough to prove itself.

The belief underneath that behaviour is reasonable. Revenue going down feels like evidence that something has broken, and taking action feels responsible. In a campaign that is running properly, though, the early shape of the graph is almost always a climb followed by a fall, and that fall is a consequence of the campaign working. Knowing the difference is most of the job, and here is the week that taught me to explain it properly.

What happened at Harvest Bloom

Harvest Bloom Flowers is a cut flower farm in Hamilton run by Rachel, who grows the seed varieties she sells and packs the boxes herself. Between January and July the online shop had taken NZ$723. We rebuilt the site, sorted the foundations, then put spend behind it.

The first week was excellent. Revenue climbed steadily into the weekend and kept climbing. Then Sunday, Monday and Tuesday fell off a cliff, three days almost flat after a run that had looked like the business finally taking off.

Rachel messaged me in the middle of it, and it was a fair question. When you have grown the seeds yourself, a quiet Sunday feels personal in a way no spreadsheet captures. She had watched something go from almost nothing to genuinely promising, and then watched it stop.

The anxiety had plenty of suggestions. Change the audience, swap the creative, move the budget somewhere it looks like it is doing more. Every one of those resets what was already working, which is what makes them expensive. So I went and looked at the numbers underneath the revenue line.

The reason the graph did that

The ads had scaled. In the three days that looked like a collapse, they were reaching four and a half times more people than in the week that had looked like a triumph.

That is the whole explanation, and it is worth sitting with because it runs against instinct. Every audience has a warm layer at the front of it: people who already know the business, people who have visited the site, people whose interests line up so neatly they were always going to buy. A new campaign finds them first, because that is what the delivery system is built to do, and they convert at a rate you will never see again.

Then they run out. There are only so many of them, and once the campaign has been through that layer it starts showing your ads to strangers. Strangers convert more slowly. They look, they leave, they come back four days later. Your revenue line drops while your reach climbs, and if you are watching revenue alone it looks exactly like failure.

That big first weekend was never a peak we fell off. It was the warm crowd getting worked through, which is what is meant to happen. So we held. No new audience, no new creative, nothing clever. It recovered, and the campaign went on to run at NZ$2.31 per conversion with sales up 580%.

The bravest thing I did that week was nothing at all, which is the part nobody warns you about in this job. Everyone expects the big swing and the bold save. Most of the time the skill is recognising you already hold a decent hand and choosing not to overplay it.

Six checks before you touch a campaign

This is the sequence I run whenever a client account has a bad three days. It takes about twenty minutes and it settles the question nearly every time.

1. Check reach before you check revenue

Pull the number of people the campaign reached in the bad window and compare it to the good one. If reach has grown significantly while revenue dropped, the campaign has moved beyond its warm audience. That is a scaling pattern, so hold.

If reach has collapsed alongside revenue, something mechanical is going on: budget capped, an ad rejected, an approval stuck. Go and look at delivery.

2. Check frequency

Frequency is the average number of times each person saw your ad. If it has climbed steeply while reach stayed flat, the campaign is circling the same small group and wearing them out. That is one of the few dips that genuinely calls for new creative, and it is rarer in the first fortnight than people assume.

3. Check whether the sample is big enough to mean anything

This is the one that catches experienced operators, and it caught me a few days before all this. Three of Rachel's video ads had produced no sales, so I turned them off after one day.

Then it nagged at me and I went and looked underneath the chart. Those three ads had been shown to 350 people, 176 people, and 26 people. At the store's normal conversion rate, an ad put in front of 26 people would be expected to produce 0.06 sales. Zero is what you would get from a shocking ad, an average ad, or the greatest ad ever made. It is a coin landing tails once.

Two went straight back on. Before you judge any ad, work out how many conversions its traffic should have produced at your site average. If that number is below one, you have no information yet.

4. Check the tracking

Before you conclude that customer behaviour changed, confirm that your measurement did not. A broken pixel, a changed checkout, an expired payment gateway, a plugin update on Thursday night. I have seen a three day revenue drop turn out to be a cookie banner blocking its own conversion tag. Load your own site, buy something small, and watch whether it lands where it should.

5. Check the calendar, including the boring parts

Day of week patterns are real and they are stronger than most people expect in ecommerce. Weekends, paydays, school holidays, a public holiday in your main market, the week everyone in Queensland is at the Ekka. Compare this Monday to the last four Mondays. If the shape repeats weekly, you are looking at a rhythm.

6. Check how long the current setup has actually been live

Write down the date of your last change, then count the days. A change made on Thursday has not had a fair run by Sunday. If the answer is fewer than seven days and nothing in checks one to five looks broken, you have not yet earned an opinion about performance. Wait.

If all six come back clean, do nothing. That is the answer, and it is harder to give a client than a new strategy is.

When a dip does mean something

Holding is the right call more often than people think. Some dips are real, and they look different from an ordinary scaling dip.

Frequency climbing while reach stays flat is genuine creative fatigue. Cost per click rising steadily across a fortnight while your site conversion rate holds says the auction has got harder. Site conversion rate falling while traffic quality stays constant points at the website, which is usually where I end up. Traffic arriving and bouncing within a few seconds means the page has broken or the ad promised something the landing page does not deliver.

The common thread is that a real problem shows up in more than one metric and gets worse over time. A scaling dip shows up in revenue alone and then stabilises. Twenty minutes in the account will usually separate them, and three days will do it with certainty.

Patience and neglect look the same from outside

Here is the line I keep coming back to. Sitting on your hands during a dip and abandoning an account for a month produce identical activity logs. The only thing separating them is whether somebody is actually watching.

Most retainers get this backwards. There is an entire industry built on a beautifully designed PDF arriving on the third of the month, describing something that stopped being true on the eleventh. That is neglect with good typography. If your marketing partner cannot tell you what changed in your account last Tuesday, the calm they are recommending is a guess.

I look at client accounts every day, because a week is long enough to lose a season and I am not spending somebody's money without watching where it goes. Watching is also what makes holding defensible. I could tell Rachel to sit tight because I could show her the reach number, and a reason beats reassurance every time.

There is a second trap here, and it is newer. The AI reporting layer sitting on top of most ad accounts will hand you a confident looking zero and let you act on it, because nothing in it knows that number is too young to mean anything. That is what happened with the 26 person ad. My gut said keep them running, the dashboard said zero, and I went with the screen, because a chart looks like a fact and a feeling looks like an excuse. Use the feeling to decide where to dig, and use the numbers to find out whether anything is down there.

The short version

A campaign that climbs then falls is usually a campaign that has finished with its easy audience and started on the hard one. Check reach, frequency, sample size, tracking, the calendar and the age of your last change, in that order. Hold when nothing is wrong and dig when something feels off, and do both from inside the account.

All six checks assume the traffic is landing somewhere that converts, which is why at Infinity1 we won't run your ads without doing an audit and getting the score above 75. You would not hand the kicking tee to a middle forward when you have a fullback landing 85%. The ad budget is the shot at goal, the website is who takes the kick. Stack the odds, then kick.

If you are staring at a graph right now and cannot tell which kind of dip you are in, send it over and let's talk it through.

FAQ

How long should I leave a new campaign before judging it? Long enough for the conversion maths to be capable of producing a result, which depends on your own conversion rate and traffic. Calculate it. As a floor, no ad has earned a verdict after a single day.

Does more reach always mean a lower conversion rate? Usually in the short term, because the warm audience converts first. What matters is total revenue over a fortnight against spend. A campaign can have a falling conversion rate and rising profit at the same time.

What if my client is panicking mid dip? Show them the reach number beside the revenue number. A client who can see that four and a half times more people saw the ads that week will sit with it. A client handed only reassurance will keep asking, and fairly so.

Is this specific to Meta ads? The audience temperature pattern shows up on any platform with an auction and an optimisation loop, which includes Meta, Google and TikTok. Search campaigns move more slowly because intent is steadier, so the dip is shallower and the same checks apply.

Should I ever make more than one change at once? Only when you have stopped caring what you learn from it. Change one variable, give it a fair run, then read the result. Change three and you will never know which one did it.