Learning Phase

I used to check the account at 11am and again at 4pm. If the number looked bad at 11, I changed the audience. If it still looked bad at 4, I changed the creative.

I thought I was being on top of it. I was actually paying for a fresh first week, twice a day, and wondering why nothing ever settled.

Here is what I did not understand back then.

A new delivery rider joins on Monday. He is good at the job. He rode for years in another city. He just does not know this one.

First week he takes the long way. He guesses which lane moves at 6pm and guesses wrong. Three drops in the time he will later do six. The parcels arrive. They just cost you more, and you cannot guess what the next one will cost.

Nobody watches that week and says the rider is bad. Everyone knows he is learning the city.

That is the learning phase. Every time you switch on a new ad set, Meta puts a rider on the road who does not know your city yet.

It is spending your money on tests, on purpose.

Meta is not trying to get you the cheapest result today. It is trying to find out where the cheap results live. So it explores. Different people, different placements, different hours. Every conversion that comes back is a street name the rider writes down.

That is why week one is expensive and jumpy. Monday looks fine, Tuesday looks like a disaster, Wednesday looks fine again. You are not watching performance. You are watching someone learn.

Meta puts a number on how many runs he needs: a minimum of 50 events over a 7-day period, for one ad set.¹

Two words there matter. Events means your optimization event, not activity in general. If the ad set optimizes for purchases, it needs 50 purchases. Add to carts do not count. And it is per ad set, not per campaign. Five ad sets are five riders, each needing their own 50.

Editing is the expensive part.

A significant edit restarts the phase.² You moved the rider to a new city. Everything he learned is worth nothing now.

The clock restarts too. Meta counts the 50 in the week after your last significant edit.³ So the finish line moves to seven days from right now.

This is the most expensive habit in a young account, and it never feels like a mistake while you are doing it. That was my 11am and my 4pm. Two resets a day, every day, each one a first week I paid for and threw away.

Learning limited is not a punishment.

Meta's own words: "Learning limited isn't a penalty."³ It means the ad set is unlikely to reach about 50 events in the week, so your budget is not being spent well. You gave the rider two drops a week. He will still be guessing in month three.

Three things people do not realize about the learning phase:

1. Two of the causes are not money. Meta lists high auction overlap and too many ads running at once, right alongside small audience and low budget.³ Those two are structure, not spend. The account is not underfunded, it is split up, and your own ad sets are bidding against each other.
2. Shops ads use a different number. Not 50. It is 17 purchases through your website and 5 through Meta after 7 days.³
3. You can see the count. Optimization events is a reporting column almost nobody turns on. Columns, Customize columns, tick Optimization events, apply.³ Then you know if you are close instead of guessing.

Work out your floor before you build the structure.

50 events in 7 days is about 7 a day. Multiply that by what one result costs you and you get the least an ad set needs per day to finish learning at all.

Run this before you decide how many ad sets to have, not after. If the honest answer is one ad set, build one.

One honest caveat.

Getting out of learning is not the goal. Profit is. An ad set can leave learning and be reliably unprofitable. It works the other way too. An ad set sitting in learning limited can still be making you money, and rebuilding the account to chase a green label is how you break a campaign that worked.

Be careful with Meta's easiest fix, switching the optimization event from purchases to add to cart. It clears the status, because add to carts happen more often. It also changes what you are buying. That last judgement is mine, not Meta's.

The one line to walk away with: the rider learning your city does not mean you sent him to the right city.