start here

An auction runs every time you open the app

In the half second your feed loads, advertisers you have never heard of just fought over the right to reach you, and one of them won.
Nobody rings a bell. It happens billions of times a day. Here is what actually goes on in that half second.
the normal state

Your feed is not sitting there waiting

There is no fixed page. It is assembled fresh, for you, in the moment you pull it down, and a few of the slots are for sale.
So an advertiser is never buying a page. They are buying a chance at one person, one slot at a time.
the only rule

The biggest bid does not win

Meta gets paid when you are shown something, but only makes money long term if you actually engage. So it does not rank bids. It ranks bid times how likely you are to act times how good the ad is.
That number is called total value. Everything else in this story is a consequence of it.
TOTAL VALUE
Big Brand
bids £20.00
0
will they act? ?
ad quality ?
WINNER
You
bids £6.00
0
will they act? ?
ad quality ?
WINNER
Spam Co
bids £12.00
0
will they act? ?
ad quality ?
WINNER
THE BIGGEST WALLET LOST
watch it happen

Three advertisers, one slot

Big Brand bids twenty pounds. You bid six. Add in whether you will actually click, and whether the ad is any good, and the money stops mattering so much.
0in the mix
›››
drag to run the auction
👈
the nice surprise

You rarely pay what you bid

The winner is charged only enough to have beaten the runner up. Bidding high mostly sets your ceiling, it does not set your price.
Which is why bidding harder is a weak lever. The other two parts of that sum are where the real money is.
the part doing the work

It is guessing what you will do next

Before the ad is shown, Meta predicts the odds that you specifically will click, buy, or fill the form. That prediction comes from what people like you have done after seeing things like this.
Two advertisers with identical budgets pay completely different prices, because one has a better prediction attached.
the bit everyone gets wrong

Picking the audience is the small lever

Hand picked interests used to be the craft. Now the system finds the buyers itself, and the audience you type in behaves more like a suggestion than a fence.
Give it a wide net and clean signal and it will out-hunt you. Box it into a tiny audience and you are just making its job harder and its guesses worse.
CONVERSIONS THIS WEEK
guessing wildly, prices all over the place
LEARNING COMPLETE
the awkward first week

A new ad set starts out clueless

It needs roughly fifty conversions in a week before its guesses steady up. Until then the costs you are staring at are noise, not results.
0events
›››
drag to feed it a week
👈
how people ruin it

Every nervous tweak sends it back to the start

Change the budget hard, swap the audience, switch the goal, and learning restarts. You are the gardener pulling the seedling up to check the roots.
If a campaign has been rebuilt three times in a fortnight, it has never once shown you what it can do.
the feedback loop

It only learns if you tell it what happened

The pixel on your site, and the server side feed alongside it, report back which click became a sale. That is the only way the prediction ever gets sharper.
Broken tracking is not a reporting problem. Blind bidding is the actual cost, and by default it counts a sale within seven days of a click or one day of a view.
so what moves the needle

The ad itself is the targeting

The thumb stop, the first line, the offer. They are what change the likelihood you act, and that sits right in the middle of the sum that decides the auction.
Same budget, same audience, different creative, and the cost per sale can halve. That is not a slogan, it is just the maths above.
SAME AD, SAME PEOPLE, WEEK AFTER WEEK
times each person has seen it
1.0
click rate
100%
what you pay per sale
£12
week 1
NOTHING BROKE. THEY JUST GOT BORED.
and then it fades

Nothing lasts, because they have seen it

As the same people see the same ad again and again, the click rate slides, the prediction drops with it, and you quietly pay more for the same sale.
0cost
›››
drag through eight weeks
👈
the whole thing

Six things you now know

  • an auctionRuns in the moment your feed loads, for one slot, for one person.
  • bid x odds x qualityTotal value wins the auction, not the fattest wallet.
  • you pay lessJust enough to beat the runner up, so bidding harder is a weak lever.
  • ~50Conversions a week before an ad set stops guessing. Tweak it and that resets.
  • signalNo tracking, no learning. The prediction is only as good as what you feed back.
  • creativeThe strongest lever you own, because it changes the odds you act.
You are not buying space on Facebook. You are winning a guess about a stranger, over and over, all day.
a prune · regent