The mistake I see a lot of paid marketers make is waiting until performance gets messy to decide what success means.
That works when everything is up and to the right. You can show ROAS, CAC, CTR, MER, GA4, Shopify, Triple Whale, platform reporting, all of it. Everyone nods because the business is happy and nobody wants to spend the meeting arguing about attribution.
Then sales slow down.
Maybe it is the economy. Maybe it is seasonality. Maybe the offer is getting tired. Maybe the media buying is not working. Usually it is some annoying combination of all of it.
That is when the number matters.
Not the number you can defend in a report. The number the owner or CEO actually believes when the room gets tense and every expense is suddenly on trial.
Pick the number before the campaign starts
Before I start scaling a new channel, I want to know the one number we are going to judge it against.
Not twenty numbers. One.
That does not mean I ignore everything else. I still look at Google Ads, Meta, GA4, Shopify, Triple Whale, Impact, post-purchase surveys, internal revenue reports, and whatever else the business has. I want the mess. The mess is where the useful stuff usually is.
But there is a difference between numbers you use to diagnose the account and the number everyone agrees to trust.
The diagnostic numbers help me understand what is happening. The trusted number decides whether the work is working.
If you do not separate those two, every review eventually turns into the same boring fight:
- Google says this
- Meta says that
- GA4 says something else
- Shopify does not match either of them
- The agency says attribution is complicated
- The owner hears "I want you to keep paying me"
Nobody wins that meeting.
Where in-platform numbers fit
In-platform reporting is useful. It is just not enough by itself.
I care about Meta ROAS if I am deciding which ad set to cut. I care about Google Ads conversion value if I am looking at search terms or campaign structure. I care about CTR when I am judging creative angles. I care about CPC when traffic quality changes.
Those are operating numbers.
The problem is when marketers treat operating numbers like business truth.
Every ad platform has an incentive to tell you it is working. That does not make the data useless, but it does mean you need to stop acting surprised when the CEO does not fully trust it.
They can feel when the report is a little too convenient.
They might not know the difference between click-through attribution and view-through attribution, but they know when revenue is not matching the story in the deck.
That is why the trusted number usually lives closer to the business.
For ecommerce, MER can be good. Total revenue divided by total ad spend. If we spend more, does the whole business move? Pretty hard to hide from that.
Shopify new customers can be good too. If the goal is acquisition, show me actual new buyers, not only purchases or attributed conversions.
Contribution margin can be good if the business is mature enough to track it cleanly.
But for demand creation, my favorite is still the post-purchase survey.
Ask the buyer
The question is painfully simple:
Where did you first hear about us?
That is not perfect attribution. People misremember. Some people pick the first option. Some channels are easier to recall than others. If you are looking for perfect measurement, congrats, you will be miserable forever.
But it tells you something very hard to fake.
It tells you what the buyer remembers as the thing that introduced them to the company.
That matters a lot when you are testing channels like YouTube, Demand Gen, Meta prospecting, creator ads, podcasts, or anything else that might not convert the same day someone first sees it.
I set this up with a client before we scaled spend into a newer channel. Before the test, that channel had basically no survey signal. Now the weekly meeting is not a debate about whether the platform is taking too much credit. We can look at the survey answers and ask a much better question:
Are more buyers saying they first heard about us there?
That changes the whole conversation.
You will not get every buyer to answer. In a lot of accounts, a good response rate might be around 20-30%. That is fine. You do not need perfect. You need enough signal to make a decision.
Ask right after purchase, while the memory is fresh. Do not bury it in an email three days later. Do not make them type a paragraph. Do not give them 19 choices and a homework assignment. Make it easy.
For a paid media test, I usually want options that map cleanly to how the business actually buys media:
- YouTube
- Friend or family
- Podcast
- Other
If we are testing YouTube and YouTube was basically zero before, I want to know if that answer starts moving.
If we are testing Meta and the survey says Facebook is doing the work while Instagram is getting too much spend, that tells me something I might not see clearly in Ads Manager.
If we are testing creator ads and buyers start naming the creator, that is a different kind of signal than CTR.
Again, imperfect. Still useful.
The 25% rule
The way I read survey data is simple.
If roughly 25% of buyers answer the survey, multiply the answer count by four to get a directional read.
If 10 buyers say they first heard about you from YouTube, that is not "10 YouTube customers."
It is probably closer to 40 YouTube-first customers, directionally.
Then compare that against spend.
If the campaign spent $1,200 that week and the survey-adjusted read says YouTube created about 40 first-touch customers, you are looking at roughly $30 CAC from that demand source.
Is that exact? No.
Is it useful enough to decide whether the test deserves more time, more creative, or more budget? Usually, yes.
You are trying to get 80% of the way to the truth so the business can make a call.
That is the whole job more often than people want to admit.
The metric hierarchy I like
If I am walking into an account, I like to separate metrics into three buckets.
Level 1 is the trusted business number. This is the number the CEO cares about. It might be MER, new customers, contribution margin, total profit, or post-purchase survey acquisition by channel.
Level 2 is the channel operating number. This is what helps me manage the channel. Google Ads conversion value, Meta ROAS, CPA, CPC, CTR, conversion rate, search term quality, creative fatigue, audience overlap.
Level 3 is the diagnostic mess. This is where I look for why something is happening. Landing pages, offer changes, inventory, email calendar, promo schedule, survey comments, GA4 paths, Shopify cohorts, device mix, geography, customer type.
Most marketers blend all three together and then wonder why the meeting gets confusing.
Keep them separate.
Use Level 1 to align with leadership. Use Level 2 to manage the account. Use Level 3 to figure out what is broken.
That is how you avoid turning every performance conversation into an attribution argument.
Why this matters
Performance marketing is a weird job because you are never that far from being blamed.
When numbers are green, everyone wants more budget. When numbers are red, everyone wants to know what marketing is doing.
That is fair. We are spending the money.
But if the source of truth was never agreed on, you end up defending yourself instead of making decisions. That is bad for the business and bad for you.
The better move is to pick the number early, before the test starts and before anyone is emotional about the result.
Then every week you can come back to the same simple structure:
This is the number we agreed on. This is where it started. This is what changed. This is what I would do next.
That is not fancy. It is not the kind of reporting setup people brag about on LinkedIn.
But it is the kind of thing that keeps you trusted long enough to do the real work.