Click-through rate (CTR)
The share of people who clicked the ad out of those who saw it. If your ad is shown to a thousand people and twenty click, your CTR is 2%.
It tells you whether the image, video, or headline attracts the right audience. But it can mislead: a curiosity-driven ad may collect lots of clicks from people with no intention to buy.
Cost per click (CPC)
What you pay on average for each click: total spend divided by clicks.
A low CPC isn’t always good news. A cheap click from an uninterested audience is effectively more expensive than a pricier click from a customer ready to buy.
Cost per acquisition (CPA)
What you pay for each result that matters: an order, a booking, or a lead. Total spend divided by results.
It’s the first metric I look at for most service businesses, restaurants, and real estate, because it ties directly to the goal, provided the “result” is counted correctly, meaning a confirmed order and not just a button click.
Return on ad spend (ROAS)
The value of sales generated by ads divided by ad spend. If you spend 1,000 and generate 5,000 in sales, ROAS is 5.
It’s essential for online stores that know the value of every order. But remember it measures revenue, not profit: a high ROAS on a low-margin product can still be a loss.
Reading the metrics together
- High CTR, high CPA: the ad attracts, but the page, offer, or audience doesn’t convert. Look at what happens after the click.
- Low CTR, good CPA: a small but interested audience. Try different creative before widening the audience.
- Low CPC, no orders: usually the wrong audience, or a tracking problem.
- High ROAS in the platform, weak profit in the books: check how sales are counted and what your margins are.
Which metric matters most for you?
Online store: ROAS, watched alongside profit margin. Service or real-estate business: cost per lead, then the share of leads that become customers. Restaurant: cost per booking or order, compared with actual bookings in your system.
In every case, don’t judge an ad on its first two days. Platforms need time to learn, and results need enough data to mean something.