Short answer
CPM is what 1,000 views cost, CTR is the share of views that click, CPC is what a click costs, and CPA is what an order costs. Enter your numbers into the free ad metrics calculator on this site and it works out all of them, plus conversion rate and ROAS.
Free tools for this
The five numbers
CPM is your spend divided by impressions, times 1,000. CTR is clicks divided by impressions, as a percentage. CPC is spend divided by clicks. Conversion rate is orders divided by clicks, as a percentage. CPA is spend divided by orders. ROAS is sales divided by spend.
Read them in order, left to right
They follow the customer's path: they see the ad, click it, land on the page, and buy. A problem shows up at one step, and the metric at that step is the one to look at.
A high CPM means you are paying a lot to be seen. A low CTR means the ad is not interesting to the people who see it. A high CPC usually follows from those two. A low conversion rate means the page, price or offer is not convincing. A high CPA is the result of all of them.
Try it now: Ad Metrics Calculator → · Break-even ROAS Calculator →
Which one to fix first
Start at the end. If CPA is below what an order earns you, you are making money and the funnel works well enough. If it is above, work backwards: is it the conversion rate or the cost of clicks? Fixing the page often does more than rewriting the ad, because every visitor you already pay for benefits.
A worked example
Spend ₹10,000, get 2,00,000 impressions, 3,000 clicks and 60 orders worth ₹48,000. CPM is ₹50. CTR is 1.5%. CPC is about ₹3.33. Conversion rate is 2%. CPA is about ₹167. ROAS is 4.8. Whether that is good depends on what an order earns you after costs, which is why ROAS alone never settles it.
Why your numbers disagree with your store
Ad platforms count an order if it follows a click or a view within their attribution window. Your store counts every order, including those that came from other sources. The two will rarely match, and checkout apps that hide the sale from the ad platform make the gap bigger.
Compare trends rather than single numbers, and fix the tracking before trusting a small difference.