Why You Cannot Judge a Campaign in Week One

A campaign goes live on Monday. By Thursday it has spent a few hundred dollars and produced one enquiry, so somebody turns it off and tries something else.

That feels responsible. It is the reason the account has been busy for two years and has never found a winner.

The system is still learning

Modern ad platforms do not simply show your advert to the audience you selected. They decide, for each impression, whether to show yours or somebody else’s, using a model that gets better as it accumulates results from your specific campaign.

Meta’s engineering team described this directly when they published on Andromeda, their ads retrieval engine, in December 2024. They reported a 6% recall improvement to the retrieval system delivering an 8% ads quality improvement on selected segments, built on an increase in model complexity of roughly ten thousand times for enhanced personalisation. Advertisers enabling Advantage+ AI features reported a 22% increase in return on ad spend.

Two things follow from that, and the second one is the point. These systems are extremely good at finding the right person, and they can only do it once they have enough outcomes from your campaign to learn from. Turn the campaign off at day four and you have paid for the learning phase and collected none of the benefit. Do that ten times and you have funded ten learning phases and never bought a single result.

The numbers are too small to mean anything

Separately from the platform, there is a plain statistical problem.

One enquiry from four hundred clicks and one from six hundred are not different results. They are the same result with noise on top. At the volumes most businesses under fifty million actually run at, a week of data cannot distinguish a good campaign from a bad one, and the decision you make from it is essentially random.

Which produces a specific failure: you will sometimes kill your best campaign because its first week was unlucky, and you will never find out.

The delay nobody accounts for

Then there is your sales cycle. If the average client takes six weeks from first contact to signing, a campaign judged on revenue after seven days is being judged on nothing.

This is the mistake that most consistently kills campaigns that were working. The leads arrived, the platform reported cost per lead, somebody compared it to a target built for a different channel, and the campaign was stopped four weeks before the first deal from it would have closed.

Every account should have a written note of how long the lag is. Almost none do, and it is a twenty minute calculation from your own CRM.

What to actually watch early

Not spending for three weeks in silence, and not judging outcomes in four days. Watch the things that are visible early and genuinely predictive.

• Is it being delivered at all. A campaign not spending is a setup problem, not a performance problem, and it should be fixed on day one.

• Click through rate. Enough data accumulates quickly, and it tells you whether the creative is landing. This is where genuine early signal lives.

• Landing page behaviour. If people arrive and leave in four seconds, the problem is downstream of the advert and no amount of budget fixes it.

• Lead quality, read rather than counted. Ten enquiries all from the wrong sector is a targeting fault visible immediately, well before any cost per acquisition figure is meaningful.

Those four let you kill genuinely broken campaigns fast while giving working ones the room they need.

Where AI helps you wait

The reason people cut early is rarely ignorance. It is anxiety, and anxiety comes from not being able to see what is happening between the click and the closed deal.

That gap is now closable. A system reading the actual enquiries as they arrive can tell you on day three what a cost per acquisition figure will not tell you for two months: that these leads are the right size of business, in the right region, asking the right questions. That is a judgement about content rather than a count, which is why it previously required somebody to sit and read every enquiry, and why nobody did.

It also closes the loop properly. Matching each closed deal back to the campaign that produced it, months later, is unglamorous data work that almost never gets done manually. Done automatically, it means your next decision is based on revenue rather than on lead count, and revenue is the only number that was ever worth optimising.

Waiting is easier when you can see something. Give people a signal that updates daily and is genuinely informative, and the pressure to cut on day four disappears on its own.

How long is long enough

There is no universal number, but there is a usable way to arrive at yours.

Take the number of conversions you need before a result stops being noise, which for most businesses is somewhere around thirty to fifty. Multiply by your current cost per conversion. That is the minimum spend before the question can be answered at all, and for a business with a $400 cost per lead it is considerably more than most people are comfortable committing.

If that figure is uncomfortable, the correct response is to run fewer campaigns rather than shorter ones. Three campaigns each funded to the point of an answer beats twelve that each got a week, and twelve at a week is what most accounts are actually doing.

Then add your sales cycle to the time before you look at revenue. Six week cycle means the revenue verdict arrives in month three, and everything before that is a leading indicator rather than a result.

A rule to write down

Before a campaign launches, agree three things and record them: the minimum spend before any judgement, the minimum time, and the number that decides it.

Decided in advance, those are a discipline. Decided afterwards, they are whatever justifies the decision somebody already wanted to make. That is the entire difference between an account that improves and one that has been busy for two years.

Sources

• Meta Engineering, Andromeda: Advantage automation, next generation personalized ads retrieval engine, 2 December 2024.

AI Optimize reads every enquiry as it arrives and ties closed revenue back to the campaign that produced it, so you can judge on the right number instead of the fast one. That work sits under Paid Ads Management.