The lecture is taught by Kwon Jun-seong, who runs the Meta Ads Masterclass. Over the past 10 years he has met, trained and delivered results with more than 300 employees, and this course lays out the Meta ads know-how he built up along the way. What you take away depends on your level.
The course was built on one standard: what would I teach an employee who has to deliver results with me? I wrote it imagining an employee put on a project that must succeed, where I have to pass on every skill from A to Z so we can get results together.
It starts with perspective on Meta ads, that is, how to approach them, then explains real campaign combinations and how to use them, and campaign operating strategy, with real cases. Take in just this part properly and you will know well enough how Meta ads are run and what Meta ads are.
The next chapter covers how to judge data and how to design around it so your ads perform.
The last part covers how to plan and produce high-performing creatives, and how to automate with AI.
| Chapter | Details (as written on the table-of-contents screen) |
|---|
| Redesigning the performance perspective and a campaign operating framework | The right perspective for mastering Meta ads. The essence of Meta ads: people, not the system. The same product is used differently depending on who consumes it. What is machine learning? Setup standards by campaign objective. The basic structure of Meta ads. Demo of what not to miss in ad set settings (audience, placements, budget) and creative setup (cropping, optimization options, text limits) |
| Using campaign combinations | Real combinations such as Traffic + conversion and Engagement + conversion. Which combination to use in which situation |
| Campaign operating strategy | Applying the 3-step strategy 1-1-1 → ABO (ad set budget optimization) → CBO (campaign budget optimization) in practice |
| How to read data, judge it and apply it | A data-reading approach that creates results (CPM, CTR, content response). ON/OFF criteria for judging creatives. What is a funnel? How to design a funnel |
| Online commerce (products) | Beauty, food, high-involvement, fashion, household goods |
| Online non-commerce (services, digital products and lead gen) | Beauty, food, high-involvement, fashion, household goods (exactly as shown on screen) |
| Offline | Restaurants, cafes, etc. |
| Ad design by budget |
| Content planning / production / usage flow | A 5-step process for planning content that takes off. Content production flow. Getting 1,000% out of content (OSMC) |
| AI automation |
Each cell is one lecture of the full Masterclass and its length in minutes. Cells marked with a chapter id (1-1 to 7-3) are the 13 chapters of this Standard Edition; the blue one is the chapter you are reading. Faded cells are Masterclass-only, and Appendix B gives their key points and checklists. Group headings follow the instructor's order. MCP = Model Context Protocol.
What is the difference between ads that work and ads that don't? Is it detailed targeting? Is there some secret campaign structure everyone knows but me? Is it money?
Or is the product just better? What secret lets that person win while I don't?
The difference is whether you go looking for the right answer or work toward an answer. The right answer is about the result. Working toward an answer is about the process. You have probably heard plenty of lines like the ones on the left below. That is the right-answer way of looking at ads.
The red text under the slide's title in the lecture reads "There is no right answer."
To do Meta ads well, work toward an answer. There is no right answer, so approach it by asking why people act the way they do, and how the structure and message have to be built to draw out the behavior you want.
This drawing checks the lecture's "there is no right answer" against the official documents. The documents give no number for how big a budget increase restarts machine learning.
Put Meta ads into one sentence and you get: add options, then optimize.
You have to add options first before you can optimize by turning things off, on and up
Sometimes no number of options helps. Then the product most likely doesn't fit DA (display ads, that is, image ads), and the funnel has to be redesigned. If you've run a lot of Meta ads and spent a lot of money but efficiency won't come, the cause is one of three things.
Almost no product is a total mismatch for Meta ads. But sometimes the resources are better spent elsewhere. That was the case with a diet food brand the instructor ran himself.
Its annual revenue was about ₩5B (≈ $3.76M), and while building that revenue it spent ₩3M (≈ $2,250) on Meta ads. Over 3 years it sold ₩4–5B (≈ $3.0–3.8M) a year, about ₩15B (≈ $11.3M) in total, and spent about ₩3.5–4M (≈ $2,630–3,010) on Meta in those 3 years. This is not a story about turning ₩3M into ₩5B or about a huge ROAS (return on ad spend). It's a story about trying it and getting no results.
He knew how to make content and had gotten results on many other ads, yet with his own product it didn't work. Testing ₩4M worth is, from an expert's view, a lot of testing. The conclusion was that the funnel was wrong. A wrong funnel means redesigning it and going in a different way, and compared with that, selling on Coupang (Korea's Amazon-type marketplace), Naver (Korea's main search portal) or Google, or getting influencers to post, fit far better for the resources.
Ads always have to weigh output against input. So he didn't run Meta ads. This is a case of a product that doesn't fit image ads. In cases like this, weighing everything and turning your ads toward another channel is also an option.
| Diet food brand | Value |
|---|---|
| Annual revenue | ₩4–5B (≈ $3.0–3.8M) (about ₩15B (≈ $11.3M) over 3 years) |
| Meta ad spend over 3 years | ₩3.5–4M (≈ $2,630–3,010) |
| Judgment | The funnel didn't fit. Output was low for the input, so it stopped |
| Channels used instead | Coupang, Naver, Google, influencer posts |
One square (₩100M) blown up. The blue part, ₩4M, is 4% of one square
The lecture says not to read this number as a ROAS brag. It is the third of the three things to check when more options still don't work (content, ad structure, funnel). A case the instructor ran himself, cited in the lecture
Go around the loop from ① in the direction of the arrows. At ③ check output against input, then add options again
If the loop still doesn't deliver efficiency, narrow the cause to one of three. The instructor's diet food case was judged a funnel problem, so Meta ads were stopped and the work moved to other channels
There are two ways to add options: finding audiences and designing campaigns.
Without options, you just keep the ads running. And you pray. Yesterday's ROAS was good, so you hope today's will be too: so-called prayer Meta. With no options, praying is all you can do. That's why you need to add lots of options.
Making dozens of pieces of content is one example. It's not the right answer, but it's the simplest way to add options. With dozens, you can boldly kill the content that doesn't perform. So the first thing to do is add options.
This content side is how you find audiences.
If you only build Sales campaigns, then when Sales stops delivering there's nothing to do but make new content. Mix Traffic or Engagement campaigns into your Sales campaigns and you have many more moves.
The three campaigns affect each other and help revenue and ROAS. This is how you add options.
The upper branch on the previous screen is finding audiences: make lots of content and find the people who respond to it. The lower branch is campaign design: of Meta's 6 campaign objectives, add Engagement and Traffic to Sales. ① to ④ is the order the lecture describes: adding an Engagement campaign improved Sales campaign efficiency, and finally Traffic joined so all three run together and affect each other. The names of the 6 objectives are Ads Manager's official labels.
Split Meta ad results into shares, and in the past content got 80 and ad design 20. Today machine learning works very well, so content still has a big share, but with the same content, slightly better ad design can change results a lot. So ad design gets 30%. Either way, content is what matters. This part, though, focuses on ad design rather than content.
As machine learning got good, ad design's share grew from 20 to 30
Each bar is the whole result (100%), and length is proportional to share. The top two rows are the shares the lecture gave; the bottom two are Nielsen's sales contribution from about 500 campaigns analyzed from 2016 to Q1 2017, where creative is largest at 49% (including 2% context).
Nielsen splits out a 15% brand share and measured consumer-goods campaigns including TV, so its numbers can't be laid directly against the lecture's.
Let's look at a case to see which ad is better. The ad dashboard shows one campaign with a cost per result of $0.65 and one at $0.76. The daily budget is the same.
The amount spent is almost the same too. Yet the results differ by more than 100.
| Daily budget | Amount spent | Results | Cost per result | |
|---|---|---|---|---|
| First campaign | $686.00 | $688.18 | 1,062 | $0.65 |
| Second campaign | $686.00 | $690.54 | 904 | $0.76 |
On the dashboard screen alone, the first campaign at $0.65 looks better.
Looking at this screen alone, most people would have to pick the first. The metrics you can see are limited, and the first has the lower cost per result. If cost per result were the customer acquisition cost, meaning cost per purchase, that call could be right. But this metric isn't a purchase metric; it's a custom metric. It's a metric before the purchase that correlates with purchases. Think of it as something like add to cart. When cost per result tracks some other acquisition-related metric rather than purchases, deciding on it alone is likely to be wrong.
Start from the question and answer the box in the middle to pick a branch. By Meta's definition, the result depends on the objective and settings chosen for each campaign. If the result is a purchase, cost per result is cost per purchase, so you take the purchase branch and judge by profit range. Case 1's result is not a purchase but a custom event called Favorite (following the store), so it's the earlier-action branch, and the lecture says picking on $0.65 vs $0.76 alone is likely to be wrong.
Let's open up more columns. Cost per result is lower for the top one and higher for the bottom one. CPM (cost per 1,000 impressions) is the reverse: higher on top, lower below. Online there's a lot of talk that high CPM is bad, so many people try to manage CPM. You can, but the instructor barely looks at CPM. That's because an expensive CPM can mean two things.
The full dashboard shows, from left to right: cost per result, CPM, cost per ThruPlay and 3s→15s, CPC and unique outbound clicks, CTR, cost per landing page view, Instagram profile visits, and post shares, with the order for reading metrics noted below them.
To tell which it is, you have to look further down.
If CPM is high but the ad really reached an audience with buying power, purchases come along, and content response like likes, shares or comments follows.
Meta ads are a competitive auction. So good audiences are expensive and poor audiences are cheap. Say brand A's Sales ads performed very well. B, C, D and E in the same industry see A and come in to advertise in the same place. Without B, C and D, A had the gains to itself, but now everyone has to bid against each other to get their ads delivered.
So CPM keeps rising. B, C, D, E and F all came in with Sales campaigns, so they compete to reach the audience with buying power, and the price goes up.
This is why mixing campaigns came up earlier. Sales campaigns are already a huge red ocean. Mix in even a little Traffic or Engagement and you pull in a pool of people for less money, so cost per result goes down.
| CPM | Conversions | Verdict |
|---|---|---|
| Expensive | Coming in | High-quality audience |
| Expensive | Not coming in | Low-quality audience, or the content is wrong |
Sales campaigns are already a red ocean, so mixing in Traffic and Engagement pulls in a pool of people for less money (lecture). The three factors in the top row of the previous screen are the auction factors in Meta's official documents. Meta writes that more relevant ads cost less and get more results. That's why the lecture says it barely looks at CPM and looks further down instead.
Meta's ad impressions and average price per ad, year-over-year change. Annual, in %. Bar length is proportional to the change. The bottom row is that year's ad revenue
These are two numbers Meta reports at every earnings release. Impressions are ad impressions served across its Family of Apps. The filings show that price is set by supply and demand together. In 2022 and 2023, when ad space grew fast with things like Reels and impressions rose 18% and 28%, prices fell 16% and 9%; in 2024 and 2025, when impression growth slowed to 11–12%, prices rose 10% and 9%. These are Meta-wide averages, not the same number as your ad set's CPM. But when your CPM rises, they're the backdrop for telling whether the cause is your content or the whole market.
Keep reading the metrics on the same screen. First, cost per ThruPlay. The first is $0.03 and the second is $0.02. You can guess the second had a more attractive opening.
3s→15s is a custom metric built to calculate retention from 3 seconds to 15 seconds. The first campaign is 48% and the second is 68%. On this number alone, it's easy to think 48% means people watched far too briefly and the second is much better. But there are possibilities here too. The content may have been so attractive that people clicked within 3 seconds.
If they went to the site that way, the 48% one could be better. You have to keep your thinking open in every direction like this.
CPC (cost per click) is cheaper for the bottom one. So how do you know whether the 48% went to the homepage because of the appeal, or just dropped off? The metric to check is unique outbound clicks: the number of people who actually went to the website. The first is 4,180 and the second is 4,806.
The first is lower. So they didn't leave because of the appeal; they just dropped off. As a result, cost per landing page view is also cheaper for the second. Cost per result is higher for the second, but the lecture sums up that every other step favors the second (only CTR (all) on the screen is higher for the first). Content response such as post shares does too.
| Metric | First | Second | Ahead |
|---|---|---|---|
| Cost per result | $0.65 | $0.76 | First |
| CPM (cost per 1,000 impressions, Korean market) | $9.75 | $7.73 | Needs reading |
| Cost per ThruPlay | $0.03 | $0.02 | Second |
| 3s→15s retention | 48.52% | 68.31% | Second |
| CPC (all) | $0.12 | $0.11 | Second |
| Unique outbound clicks | 4,180 | 4,806 | Second |
| CTR (all) | 8.16% | 6.74% | First |
| Cost per landing page view | $0.17 | $0.15 | Second |
| Instagram profile visits | 209 | 387 | Second |
| Post shares | 27 | 122 | Second |
Cost per 1,000 impressions
Click-through rate: interest in the content (impr. = impressions)
Cost per click
Cost per result
Plug in the screen values from the lecture's Case 1 and the formula holds
| Campaign | CPM | CTR (all) | CPM ÷ (CTR × 1,000) | CPC (all) on screen |
|---|---|---|---|---|
| First | $9.75 | 8.16% | 9.75 ÷ 81.6 = $0.119 | $0.12 |
| Second | $7.73 | 6.74% | 7.73 ÷ 67.4 = $0.115 | $0.11 |
The official glossary gives definitions but no formulas, so the formulas are the definitions written out. The four counts are the raw material and the metrics are divisions of them. That's why CPC equals CPM divided by CTR, and plugging in the lecture's Case 1 screen numbers actually works. This drawing puts in numbers what the lecture said: don't judge on CPM alone; look further down too.
Cost per ThruPlay is $0.03 for the first and $0.02 for the second.
3s→15s retention is a custom metric the lecture built; the definitions of ThruPlay and outbound clicks were checked in Meta's documents.
Read metrics in this order: cost per result, CPM, response, action. Of these, look hardest at action, and use CTR and action as supporting metrics. When these metrics are good, you can often scale up the budget without breaking efficiency.
| Order | Metric | What it checks |
|---|---|---|
| 1 | Cost per result | Is it within the profit range? What is each campaign achieving? ROAS and cost per purchase are a given |
| 2 | CPM | Which audience is it reaching? |
| 3 | Response | From ThruPlay to CTR. CTR shows how attractive the content is |
| 4 | Action | Landing page views, Instagram profile visits, post shares, etc. |
Meta as a platform is a place people come to hang out, to share and to chat.
Then the post actually being consumed matters more than something like clicks. Cost per result, ROAS and cost per purchase are a given; after that, look at CTR, and put metrics like likes, shares, comments and saves first. If you have several campaigns, scale up the one that leads on these. In this case the second was better, with a cost per landing page view of $0.15, so the second campaign was scaled up.
After scaling up, the second campaign's cost per result rose from $0.76 into the $1 range (the screen shows a $1,300 budget and $1.09 cost per result), and the existing campaign came to $0.78.
The first campaign, at $0.65, rose to $0.9 over time.
The second campaign was picked for scaling on the strength of cost per landing page view, Instagram profile visits and post shares, and its budget was then raised to $1,300.
Ads have this trait: efficiency is good at first, then frequency rises and efficiency drops as time goes on. In Meta ads, efficiency often drops when frequency rises or the audience pool gets bigger. Of course, rising frequency doesn't always mean a drop. For some products, higher frequency has an even more positive effect on purchases.
What comes out of this one case may not make much sense to beginners. That's fine. Take away just one thing here: you don't look at one metric, you look at two, three, four. Think in combinations and keep doubting. Doubt the dashboard. Is this really right? Should I look at something else too? Hold on to that perspective and you're set.
Case 1 screen values. Before the "/" is the first campaign, after it the second
Cost per result favors the first, CPM needs reading and response is mixed, but in the last box, action, cost per landing page view, shares and profile visits all favor the second. The lecture weighs actions where the post is actually consumed more heavily because Meta is a place to hang out, share and chat, and so it scaled up the second. The post-scaling screen values show how cost per result rises over time as frequency climbs.
Numbers are Case 1 screen values. With several campaigns, scale up the one that leads on these metrics
A month earlier, in April 2024, Instagram was already ahead of Naver (20.9B minutes vs 20.0B). KakaoTalk = Korea's main messenger
App usage time estimated by WiseApp·Retail from a sample of Korean smartphone users. On total usage time for May 2024, Instagram ranks third after YouTube and KakaoTalk. It has become an app people stay on longer than Naver, where they go to search. This time spent is the background to the lecture calling Meta "a place to hang out" and saying to check whether posts are consumed before clicks. People aren't there to find something; they're there to spend time, so ads are judged by traces of consumption like likes, shares, comments and saves. The April 2025 value is 25.877B minutes, rounded.
You have to doubt the dashboard because dashboard data is quantitative data. It is fragmented, modular data: in other words, partial data. It can't explain the conversion environment, and it can't reflect qualitative data.
Dashboard data is quantitative and fragmented. It can't explain the conversion environment, so you have to dig a little deeper.
Dig a little deeper, and the campaigns above turn out to be live-commerce campaigns. The value used as the result was "Favorite" (following the store), set up as a custom event. Tap Favorite and you're asked to log in with Kakao Login (sign-in through KakaoTalk, Korea's main messenger). Log in with Kakao and you're asked to sign up as a member. In a setup like this you can't capture the conversion, so the ads ran on a metric related to the conversion.
That's what lets machine learning work. This is the method to use when Meta can't track conversions.
On the dashboard, favorite_sign_in, shown under cost per result, is the custom event (tracking of a specific value). Below is the actual path: the Favorite button, the Kakao Login prompt, Kakao Login, then entering member details.
Cost per result = spend ÷ Favorite clicks. The sign-ups after that aren't on the dashboard
Per Meta's docs, conversions are tracked only on pages that have the pixel. So the real goal, a completed sign-up, can't be captured, and the Favorite click just before the boundary becomes the last trackable action. The lecture's reason for setting it up as a custom event, "that's what lets machine learning work," ties to the official standard that the learning phase needs about 50 results a week. So the dashboard's cost per result counts only up to Favorite, not up to sign-up.
Here is the comparison of the first and second campaigns. Impressions: about 70K for the first, about 89K for the second. The rate from landing to Favorite was higher for the first campaign at 25%, so its cost per result was about $0.1 cheaper ($0.11 on screen). But every metric after that, content response and consumption alike, strongly favored the second campaign.
| First campaign | Second campaign | |
|---|---|---|
| Impressions → landing | 70,579 → 4,108 (5.82% landing rate) | 89,316 → 4,730 (5.30% landing rate) |
| Landing → Favorite | 1,062 (25.85%) | 904 (19.11%) |
| 3s→15s retention | 48.52% | 68.31% |
| Shares | 27 | 122 (4.5x) |
| IG profile visits | 209 | 387 (2x) |
| Cost per landing | $0.17 | $0.15 |
The first campaign leads on landing → Favorite (25.85%, $0.11 cheaper per result). The second leads on retention, shares, profile visits and cost per landing.
The path people take from consuming an ad to taking some action is called the funnel. The ad is shown and watched. At that stage, people decide whether to enter the landing page.
After entering the landing page, they either tap Favorite or leave. Tap Favorite and they go through four inputs; finish those and they become a member.
Funnel of the live-commerce campaign. The dashboard's cost per result counts only up to Favorite, the middle stage. The funnel widths show only the order of stages and are not proportional to the numbers
The "4 inputs" stage corresponds to Kakao Login and entering member details.
The first campaign has lower watch time and fewer unique outbound clicks too. Its people were likely to have come in out of light curiosity, and it was a campaign with a somewhat weaker conversion rate. This is how you have to look at a funnel: as a whole. Put the two campaigns side by side from a funnel view and it gets clearer. Below is a hypothesis of what happens all the way to the end, applying each stage's rate.
| Stage | First | Second |
|---|---|---|
| Ad viewed | 20,522 (48.52% retention) | 37,058 (68.31% retention) |
| Landing page entry | 4,108 | 4,730 |
| Favorite | 1,062 (25.85% click rate, includes light curiosity) | 904 (19.11% click rate) |
| Past the 4 purchase steps (sign-up rate) | 478 (45%, average) | 633 (70%) |
| Final sign-ups | 478, $1.44 per sign-up | 633, $1.09 per sign-up |
The first leads only at the Favorite stage; the second leads at every other stage.
Cost per member is 24% cheaper for the second campaign. This is the kind of math you do to decide whether to scale up.
Bar length in ① = impressions. On cost per result alone the first wins, but the lecture scaled up the second because of these extra impressions and the further-down response in the gray box
The pass-through rates in ② are the lecture's hypothesis (45% average for the first, 70% for the second). The first's Favorite clicks are assumed to include light curiosity. The dashboard counts only up to ①, so without this math the first looks cheaper. All numbers are lecture screen values; only the rates in ② are hypothetical.
The second case also asks which ad is better. One ad spent ₩300K (≈ $225), the other ₩700K (≈ $526). The ₩300K ad's CTR (click-through rate) is 4.5% (4.57% on screen). Just as CPM can be read two ways, so can CTR.
| Amount spent | CTR (all) | Post shares | |
|---|---|---|---|
| First ad | ₩308,656 (≈ $232) | 4.57% | 105 |
| Second ad | ₩768,031 (≈ $577) | 1.34% | 67 |
On the three columns amount spent, CTR (all) and post shares alone, the first ad looks better.
CTR is basically an interest metric. If CTR is high but nothing converts, it's clickbait. Say there's a post that reads "Lotte World (a Seoul theme park) for ₩100 today." CTR would be huge. Out of 100 people who see it, maybe 70 would click. But you go in, and far from ₩100, the price is unchanged. No conversion happens. So a high CTR isn't automatically good. The metric you must always read alongside CTR is the result metric.
On this screen alone, spend is ₩300K and ₩700K, the first has a much higher CTR and more post shares, so the first looks like the better ad.
You could think that, but you have to look at it as a whole. This case has no cost per result, because the landing was moved out to a blog. So all you can do is infer from CTR and the metrics further down. Even without cost per result, results still happen.
This is an ad for a boarding academy for students retaking the college entrance exam. It's a lead-generation business model: customers come in and leave an inquiry. Comparing the two ads, the first got a good response, but its inquiry rate dropped sharply. The inquiry rate was tracked by splitting the landings. The instructor used a solution he planned, built and runs himself; it's explained later in the course. Here is what kept bothering him while running Meta ads.
If you could just match the landing to the ad content people came in on, the conversion rate would go up. But you can't make unlimited landing pages. Each one costs money, and once there are many, they're a hassle to run. He looked for a solution to this, found none, built one himself and made it trackable. The "thought I was at a hotel staycation" content was linked to a landing page called "Dorm Up," and the "lower-ranked students retaking the exam" content was linked to a blog landing that matched its content.
The share metric shows the difference. Both are lead-gen ads, but on the second ad's landing, 111 shares happened. Prospects actually came in and shared it 111 times. Add in the reaction on the product/landing page further down like this, and the second ad was far better than the first.
| First ad | Second ad | |
|---|---|---|
| Ad message | "Came here to retake the exam, thought I was at a hotel staycation" | "Lower-ranked and retaking the exam? Do this and you're 1000% doomed" |
| Dashboard (CTR, post shares) | 4.57%, 105 | 1.34%, 67 |
| Reactions on the landing post (screen) | 4 likes | 22 likes, 3 comments, 111 shares |
| Verdict | Good response, but the inquiry rate dropped sharply | Counting the reaction further down, a far better ad |
Comparing the reaction counts under each ad's landing post, the second post has 111 shares.
Why was the second the better ad? Content needs a value proposition underneath it. A value proposition says, "This bothered you, right? I'll solve your problem or meet your need." The first ad's message looked like a hotel staycation. But the real offer didn't match those staycation photos. So performance fell off somewhat.
The ad's message showed a hotel-like room; the real offer was the actual dorm. The two weren't the same.
Good response, but the inquiry rate dropped sharply
Once in, the content was different: close to reading 1 (clickbait)
Counting the reaction further down, a far better ad
Interested people came in: close to reading 2 (targeting)
On the dashboard row alone, the first ad leads with a 4.57% CTR and 105 shares, but the lecture rules that, counting the reaction further down, the second is the far better ad. Meta also says it factors attributes like sensationalized language and engagement bait into ad quality, and ad quality is one of the three factors that set an ad's total value in the auction.
US internet users, one month in March 2009. Band length is proportional to share of clicks
Halved in under two years. Bar length is proportional to the rate (full width = 100%)
On these results, comScore wrote that click-through rate is not the most appropriate metric for measuring campaign performance, and that visits, searches and purchases rose even among people who didn't click. That points the same way as the lecture ruling the 1.34% CTR ad better than the 4.57% one. Lots of clicks can mean the ad reached a small group who click readily, so confirm with result metrics such as inquiries and purchases. The study covered US web banner ads in 2009, so its numbers can't be carried over as-is to today's Meta ads.
Write down the "right answers" you follow now (CTR of 3% or more, turn off if CPM is high, scale up 20%, etc.) and turn each into a question that asks why.
Add more content and more campaign objectives (Sales, Traffic, Engagement) to create options you can turn off when they don't deliver.
If cost per result is a metric before the purchase rather than a purchase, compare CPM, cost per ThruPlay, retention, unique outbound clicks, cost per landing page view and shares side by side.
When CPM or CTR is high, check conversions and cost per result to tell which of the two readings applies.
If conversions can't be tracked, set an action tied to the conversion as a custom event, and work in the rates through the end of the funnel to calculate the final cost.
Check that the ad message matches the landing's real offer.
Next: 1-2 is also free.
Read 1-2 free →