Where, chapter eleven
How the auction spends your money
The search fresh roasted coffee beans delivery
| Spot | Advertiser | Max bid | Quality Score | Ad Rank | Pays per click |
|---|---|---|---|---|---|
| 1 | Kiln (you) | $1.20 | 8 | 9.6 | $1.01 |
| 2 | Fennick Coffee | $1.60 | 5 | 8.0 | $1.51 |
| 3 | Bigcup Warehouse | $2.50 | 3 | 7.5 | $2.11 |
| 4 | Ostrander Roasting | $0.90 | 7 | 6.3 | $0.50 |
The idea
I think of paid media as a machine that turns a budget into delivery against a goal you pick. On Meta the campaign sets the goal, the ad set decides who sees it and how the money gets spent, and the ad is the creative. A lookalike audience finds new people who resemble your best customers, and the delivery system needs conversions before it knows who those people are.
On search you bid on intent, because someone has just typed what they want. The part I find most useful is that the highest bid doesn't automatically win. Relevance and the page behind the ad lower what you pay, so a small advertiser with a better page can sit above a bigger one and pay less for the spot.
This is the chapter where my proof is thinnest. I've planned paid campaigns in class and written the next-round recommendations for one, but I haven't spent my own budget yet. The last section says what I'd do about that.
How the auction is worked out here
Ad Rank = maximum bid × Quality Score
Pays per click = Ad Rank of the ad below ÷ your Quality Score + $0.01, never more than your bid
The lowest ad pays the reserve price.
This is the simplified version, the one Google explains in its own help pages on Ad Rank and on actual cost per click. The real auction also weighs minimum thresholds, the context of the search and the expected effect of extras like sitelinks, so a real price won't match this to the cent.
Two parts are my illustration, not Google's. Google reports three ratings for a keyword (expected click-through rate, ad relevance and landing page experience) and a 1 to 10 Quality Score, but it doesn't publish how one becomes the other. Here below average counts 0, average 1 and above average 2, and the score is 2 × click-through + 2 × landing page + relevance, with 1 as the floor. The $0.50 reserve is also mine.
One thing surprised me when I built it. A better score always lowers the price of the same spot, but it can also lift you into a higher spot, and a higher spot costs more to hold. So a better page doesn't always mean a smaller bill. It means more for the same money.
On Meta, delivery narrows as it learns
Meta's delivery works differently. You don't bid on a search; you pick a goal and an audience, and the system spends the early part of a campaign finding out who in that audience acts. While it learns, delivery spreads wide. As conversions arrive, it narrows toward people who look like the ones who bought.
Day 7 of 7
Conversions 56 of about 50
Status Out of learning
| Day | Conversions so far | Status |
|---|
Where I used it
- The Yard, a group project for a company that matches brands with student creators. My part included the next-round recommendations for the paid section. On Meta, build lookalike audiences from the people who converted in the first round, since a lookalike needs converters to start from. On Google, expand the keywords from the search terms that actually brought clicks, and test new copy against the best ad. Group work, my section
What I'd do now
Three things are still open, and none of them is done yet.
- The paid-media assignment I left open: two ads for a consumer brand and a written Meta plan. I'd do them for Kiln. One ad goes to new people through a lookalike of past buyers, the other retargets people who left a cart, and the plan writes down the objective, the audience, the budget and the schedule for each.
- The Google Ads Search certification, which I skipped at the time.
- A small real test with a hard cap of about $50, the ceiling Avery Colyer suggested for a first boost in a guest session. I'd write down what I expect before it runs and report what it returns, even if that's nothing.