Facebook Ads Metrics: Which Number Answers Which Question

Read five metrics, one per question: CPM for delivery, hook rate for attention, link click-through rate for action, cost per purchase and ROAS against your break-even for money, and agreement with your store's orders for trust. The table gives each formula and what it cannot tell you.

Meta groups its Ads Manager columns into Performance, Delivery and Engagement presets, and guides sort metrics into those groups differently, so use the question, not the group, to decide what to read. The other metrics you will meet:

The right headline metric also depends on the campaign goal. Judge each campaign on its own goal and use the other metrics to diagnose it:

Read link CTR. Meta defines CTR (link click-through rate) as link clicks divided by impressions, and CTR (all) as clicks (all) divided by impressions. Clicks (all) includes link clicks plus likes, comments, shares, Page or profile clicks and clicks to expand a photo or video.

CTR (all) is therefore usually the higher figure. For an ad meant to send people to your store, link CTR is the one that measures that.

For context, AgencyAnalytics' January 2026 guide reports a median Facebook Ads link CTR of 1.92% and a median cost per click of $0.57 for agency accounts (CPC as of December 2025). Treat those as context, not targets: whether a given link CTR is good depends on your margin and what a click costs you, which we work through in what is a good CTR for Facebook ads.

Cost per click (CPC) equals CPM ÷ (1,000 × CTR), with CTR as a decimal (2.3% = 0.023). CPM is spend ÷ impressions × 1,000, CTR is link clicks ÷ impressions, and CPC is spend ÷ link clicks, so when all three count the same clicks they must fit that equation. If they do not, two of them count different clicks.

Our calculation from the March 2026 test's posted figures, all in Canadian dollars (CAD):

• A CAD $26 CPM and a 2.3% CTR imply CAD $26 ÷ (1,000 × 0.023) = CAD $1.13 per click. The post shows CAD $1.80.
• A CAD $26 CPM and a CAD $1.80 CPC imply a CTR of CAD $26 ÷ (CAD $1.80 × 1,000) = 0.0144, or 1.44%.

The likeliest explanation is CTR (all) sitting beside a cost per link click. The thread does not confirm that, so it is our inference. If it holds, the real link CTR was about 1.4%, and the test bought roughly 39 link clicks, not the roughly 62 that 2.3% would suggest (2.3% × 2,692 impressions).

Far more than 39. At about CAD $70 and a CAD $1.80 cost per click, the March 2026 test, which ran about three days in Canada, bought about 39 clicks (CAD $70 ÷ CAD $1.80 = 38.9). At a 1% to 3% click-to-purchase rate, the chance of zero sales was 30% to 68%, even for a store that converts normally.

This is our model, not a benchmark. The chance of zero purchases is (1 − rate)^39, treating each click as an independent chance to buy. The three rates are for illustration.

The rest of the delivery picture was fine. CAD $70 ÷ CAD $26 × 1,000 gives about 2,692 impressions. With reach of 2,600, the poster reports a frequency (impressions ÷ reach) of 1.11, and our estimate gives 1.04 (2,692 ÷ 2,600). Either way almost everyone saw the ad once, and fatigue was not the issue. Five add-to-carts from 39 clicks is 12.8%. The leak is more likely after the click: a product page without a clear buy button, or a Purchase event that never fires, which Meta Events Manager will show.

Our recommendation: before you cut or scale an ad, aim for enough clicks to expect about 10 purchases. Clicks needed = 10 ÷ click-to-purchase rate. That is 1,000 clicks at 1% (about CAD $1,800 at the test's CAD $1.80 a click), 500 at 2% (CAD $900) and about 333 at 3% (CAD $600). At 10 expected purchases, the count still moves by roughly a third either way about two times in three (the square root of 10 is about 3.2), and by close to 60% either way at 95% confidence. That is enough to separate ads that differ by about 2×, not ads that differ by 20%, so collect more purchases before making finer calls. To plan a test budget, use our Meta ads budget and creative test calculator.

A ROAS of 4 is good only if it beats your break-even ROAS, which equals 1 ÷ your margin before ad spend. A store keeping 25% of each sale before ads breaks even at 4.0. A store keeping about 58% breaks even at about 1.71. ROAS is revenue divided by spend, not profit.

Here is a worked example with illustrative numbers, not a benchmark. A product sells for $60, costs $18 to make and $7 to ship:

• Contribution per order before ads: $60 − $18 − $7 = $35.
• Break-even cost per purchase: $35.
• Break-even ROAS: $60 ÷ $35 = 1.71.
• At a $40 cost per purchase (a ROAS of 1.5), each sale loses $5.
• At a $15 cost per purchase (a ROAS of 4), each sale leaves $20.

This excludes payment fees and returns, so a real break-even sits a little higher.

This also shows why CTR alone decides nothing. A store keeping $35 per order and paying $1.80 a click needs 1.80 ÷ 35 = 5.1% of clicks to buy just to break even.

Ads Manager credits a purchase to an ad inside an attribution window, and some windows count people who only saw the ad. Your store records orders and has no record of who merely saw an ad. So the two totals differ even when nothing is broken.

As of March 3, 2026, Meta counts only link clicks as click-through attribution for website and in-store conversion campaigns. Likes, shares, saves and comments now fall under engage-through attribution, which has a 1-day window. Engaged view was renamed engage-through. Reported purchases can therefore drop after March 2026 without any change in performance, so compare your store's orders for the same dates before concluding an ad got worse.

Separately, on January 12, 2026, Meta stopped returning its 7-day view and 28-day view windows through the Ads Insights API, the feed that reporting tools and spreadsheets pull from. That change applies to the API only. Meta announced it on October 16, 2025. The 1-day click, 7-day click, 28-day click, engage-through and 1-day view windows remain, and Meta described the change as consistent with Ads Manager behavior. If a sheet still requests 7-day or 28-day view data, expect blanks. Comparing a report from before that date with one after may mean comparing different windows.

Our recommendation: do not try to make the two match. Compare Ads Manager purchases with store orders for the same dates each week. A steady ratio is normal. A sudden break in the ratio means tracking has changed or failed.

Run seven tests in order and stop at the first fail. Each has a pass line you can read straight off Ads Manager and your store.

When delivery, columns, break-even, sample size and tracking all pass and the money metrics still fail, the problem often sits in the creative, and the next question is which single variable to change. In our view, change one thing per test, so the result teaches you something. Our guide to how to analyze Facebook ads and choose your next test walks through the decision.

This is also why data-backed creative decisions matter more on Meta now than they used to. With less manual control over bidding and audiences, the creative is the lever you still control, and the teams that keep finding new angles to test pull ahead of the teams that guess. Metrics tell you an ad failed, not what to try next. That gap is what creative intelligence is meant to close.

That is the step we built Deepsolv for. It breaks a whole ad library down by offer, language, landing page, creator share and format, and uses rank data to separate ads that are performing from ads that are merely still running. Every insight comes with a specific action, so you end up with a clear list of what to test next.