Open Google Ads and it tells you it drove 40 conversions last month. Open Meta and it claims 25. Open LinkedIn and it reports 12. Add them up and 77 people bought from you.
Then you look at your own books. Actual signed customers, actual deposits, actual money in the account. There were 51.
Nobody is lying to you. No pixel is broken. You do not have a tracking problem, and no amount of tag auditing is going to close that gap, because the gap is not an error. It is the predictable result of asking three companies to each report on the same customers using their own rules, their own memories, and their own definitions of who deserves the credit.
This article is about what to do with that. Not how to fix attribution, because you cannot fix it. How to make good decisions anyway.
Every Platform Is Grading Its Own Homework
Here is the thing that took me an embarrassingly long time to explain properly to clients. Your Google Ads account has no idea that your Meta ads exist. None. It cannot see them, it has never heard of them, and it is not accounting for them in any way.
So when a customer sees your video on Instagram on a Tuesday, thinks "oh, that is the roofing company from the school fundraiser," searches your name on Thursday, clicks your Google ad, and books a quote on Friday, here is what happens. Google records a conversion, and it is right to. Meta records a conversion, and it is also right to. Both platforms genuinely contributed. Both are reporting honestly on what they saw.
You just got one customer.
This is not a scandal and it is not a conspiracy. Every platform reports on what it influenced, not what it caused, because causation is not something it can observe. It only ever sees its own slice. Asking Google how much credit Meta deserves is like asking one witness to a car accident what the other witness saw.
The trouble starts the moment you line those three reports up next to each other and start making decisions.
They Are Not Even Using the Same Clock
This is the part almost nobody explains to business owners, and it is the mechanical reason the numbers can never be reconciled.
Every platform has an attribution window. That is the length of time it will keep looking back and claiming credit after somebody interacts with your ad. Here is the problem: those windows are wildly, almost comically different from each other.
LinkedIn lets you set conversion windows to 1, 7, 30 or 90 days, and LinkedIn itself recommends a 90-day click and 90-day view window for most conversions. Ninety days. Three months. Somebody glances at your ad in February and buys in April, and LinkedIn will still put its hand up.
Meta works on a much shorter memory, particularly for views. And Meta changed the rules again this year. Click-through attribution now counts only actual link clicks. Previously, likes, shares, saves and comments all counted as "clicks," which quietly inflated a lot of reports. Those interactions moved into a separate bucket called engage-through. Meta also shortened the engaged-view window for video ads from 10 seconds down to 5, on the reasoning that people now decide much faster than they used to.
Google Ads has gone in a different direction entirely. It now runs data-driven attribution by default, which uses your account's own history to split credit across multiple touchpoints. Google has also retired first-click, linear, time-decay and position-based models. Two models are left standing: data-driven and last click. Anything still sitting on a retired model gets moved to data-driven automatically.
Sit with that for a second. One platform is willing to reach back ninety days to claim a customer. Another has cut its video view window to five seconds. A third is quietly re-slicing credit using a model you cannot inspect.
These reports were never going to add up. They are not measuring the same thing over the same period using the same definitions. Comparing them is not difficult. It is meaningless, in the specific sense that the comparison does not carry information.
And the windows keep moving. Meta changed what counts as a click this year. Google deleted four attribution models. If you have a spreadsheet comparing this year's channel performance to last year's, some of that movement is not your marketing at all. It is the rulers changing length.
The Same Thing Happens Inside One Ad Account
Now here is the version that costs the most money, because at least the cross-platform problem is visible. This one hides in a single account and looks like good news.
Say you run video ads on Meta. The video does its job, people watch it, and Meta builds you an audience out of everyone who engaged. Then you run a retargeting campaign to that engagement audience. Somebody who watched the video last week sees the retargeting ad, clicks, and buys.
Which ad gets the conversion? The retargeting ad. Every time.
And Meta is reporting that accurately. The retargeting ad genuinely was the last thing the customer touched. But the sale did not originate there. It originated with the video that introduced you, earned the attention, and created the audience the retargeting campaign is spending its budget on. The retargeting ad did not find that customer. It was handed that customer.
This is not a Meta quirk. Your brand search campaign in Google does exactly the same thing. Someone sees a Performance Max placement, remembers your name a week later, googles you directly, clicks the brand ad, and buys. The brand campaign posts a gorgeous cost per acquisition. It was closing a sale that something else opened.
We see this on client accounts constantly, and it is the single most common reason a well-intentioned owner makes their account worse.
So You Defund the Ad That Was Actually Working
Follow the logic through, because it is completely reasonable at every step and ends somewhere terrible.
You look at your Meta account. The retargeting campaign shows a cost per lead of 18 dollars. The video prospecting campaign shows 65 dollars. The conclusion writes itself: move the money to the thing that is working. So you cut the video budget and pour it into retargeting.
For about two weeks, this looks like genius. Your blended cost per lead drops. You feel like you finally figured it out.
Then it starts to slide. Cost per lead creeps up. Volume falls. The retargeting campaign that was untouchable is now mediocre, and nothing in its report explains why, because the reason is not in its report. The reason is that a retargeting audience is a pool, and the video campaign was the thing filling it. You stopped filling it. Now you are showing more ads to fewer people, the same people, over and over, until they stop responding. We wrote a whole piece on how quickly an audience wears out when you do this, and the mechanics are the same here.
The prospecting campaign was not your worst performer. It was your inventory. You just could not see that, because it was being judged on a number that belongs to a different job.
The general rule, and it applies to channels as much as campaigns: anything that creates demand will always look worse on a report than anything that harvests it. Always. That is not a performance difference. That is a measurement artifact. If you let the report make the decision, you will systematically defund the top of your funnel until there is nothing left to harvest, and then you will conclude that advertising stopped working.
So what do you use instead?
One Number That Cannot Lie to You
Take all the money you spent on advertising last month. All of it, every platform, every campaign, plus whatever you pay someone to run it. One number.
Now take how many new customers you got last month, and what they were worth. From your books, not from any platform.
Divide. That is your real cost to acquire a customer, and it is the only acquisition number in your business that is not self-reported by a company with an interest in the answer. If you have never worked out what a customer is actually worth to you, start there instead, because this number means nothing without it.
This is usually called blended cost per acquisition, or if you prefer the revenue version, marketing efficiency ratio: total revenue divided by total ad spend. Same idea either way. You are stepping outside the platforms entirely and asking the only question that actually pays your rent: I put this much in, how much came out?
A few things to get right when you calculate it.
Count all the spend. Ad spend, management fees, the landing page tool, the creative you paid for. If it exists to bring in customers, it counts. Owners who leave out the unglamorous costs end up with a number that flatters them.
Count new customers only. Repeat business from existing customers did not come from this month's ads. Mixing them in will make your advertising look far better than it is and will send you scaling something that is not working.
Give it a sensible period. A month is usually right. A week is noise for most small businesses. And be careful comparing months that are not comparable: a month with a holiday week in it is not the same month as one without, and that alone can look like a decline that has nothing to do with your ads.
Then track that one number over time. That is the trend that tells you whether your marketing is working.
Now, the honest limitation, because this is where a lot of advice oversells. Blended cost per acquisition tells you whether the whole machine is working. It does not tell you which part to cut. If it goes up, you know something got worse, and you do not know what. That is a real weakness, and it is why you need the next two things as well. It is also why picking a small number of channels makes this easier: the fewer moving parts, the more a blended number can tell you.
Ask the Customer. Genuinely.
There is a measurement tool sitting right there that no platform can offer you, and most businesses either skip it or bury it. Ask people how they found you.
Yes, it is imperfect. People misremember. They say "Google" when they mean the Meta ad that made them google you, which is ironic given everything above. Some skip the question entirely.
Use it anyway, because it is the only instrument you own that can see the things no pixel will ever capture. The customer who came from a referral. The one who saw your van outside a job. The one who heard you on a podcast eighteen months ago. Your ad platforms are structurally blind to all of it, and for a lot of small businesses that invisible traffic is a serious share of the business.
How to actually do it:
Use an open text box, not a dropdown. A dropdown gives people your list of guesses and they pick the nearest one. An open box gives you their words, and their words are where the useful stuff lives. "My neighbour used you for her extension" is worth more than a radio button labelled Referral.
Ask once, at the right moment. On the enquiry form is fine. On the phone, when a call comes in, is better, because you can ask a follow-up. Do not ask three times across the journey. You will annoy people and you will not get better data.
Make it optional. A required field costs you conversions, and a conversion is worth more than a data point.
Actually read them. This is where it falls down. The answers pile up in a CRM nobody opens. Set aside twenty minutes a month, read every response, and count them roughly by theme. That is the whole process.
Read directionally, never precisely. If eight percent of customers volunteer that they saw you on Instagram, that does not mean Instagram produced eight percent of your revenue. It means Instagram is doing something real that your reports are probably underselling. That is enough to act on.
The Only Real Test Is Turning It Off
Everything so far tells you what happened. Only one method tells you what a channel actually caused, and it is refreshingly blunt: turn it off and see what breaks.
This is called a holdout test, and it is the closest thing to truth available to a business your size. Pause a channel for two to four weeks. Watch your blended number and your total lead volume. If nothing moves, that channel was taking credit for demand that existed without it. If things get noticeably worse, it was doing real work that its own reporting could not prove.
Doing it without hurting yourself:
One channel at a time. Two at once and you learn nothing about either.
Give it long enough to matter. Two weeks minimum, ideally three to four. Anything shorter and you are reading noise. If your sales cycle runs long, the effect will lag the pause, so extend accordingly.
Pick a boring stretch. Do not run a holdout across a holiday, your busy season, or a week you are running a promotion. Normalize for the calendar or you will draw a confident conclusion from a week that was never going to be normal anyway.
Write down your expectation first. Before pausing, write the number you think you will see. It stops you retrofitting a story onto whatever happens.
And this is exactly how you settle the retargeting question from earlier, which is otherwise unanswerable. Pause the video prospecting campaign, leave retargeting running, and watch what happens to retargeting over the following three weeks. If its volume holds, fine, the pool was deeper than you thought. If it dries up, you have just proved the dependency that no report in the account will ever show you. That is a genuinely important thing to know about your own business, and it costs you three weeks of one campaign to find out.
Most owners never run one, because deliberately switching off something that appears to be working feels reckless. It is the opposite. It is the only time you get a straight answer.
Your Platform Numbers Are Still Good for Something
None of this means the platform reports are worthless. Do not throw them out. It means they are good at one job and terrible at another, and you need to know which is which.
What they are good for: comparing like against like, inside one platform. Ad A against ad B, both prospecting, same audience, same period. Audience 1 against audience 2, same creative. This week's landing page against last week's. In every one of those, the measurement error is the same on both sides, so it cancels out and the comparison holds. This is genuinely useful and it is where most of your day-to-day optimization lives.
What they are not good for: comparing things doing different jobs. A retargeting ad against a prospecting ad. A brand campaign against a cold campaign. Meta against Google. In all of those, one side is being handed warm demand the other side created, so you are not comparing performance. You are comparing position in the funnel, and the one nearer the finish line always wins.
The rule is short enough to remember: compare inside a platform, and only between things doing the same job. Everything else goes to the blended number.
Where does that leave you? Judging your account by three numbers instead of thirty. Your blended cost to acquire a customer, tracked monthly, tells you whether the machine is working. Your how-did-you-hear-about-us answers tell you what your pixels are missing. A holdout test, once or twice a year per channel, tells you what is actually causing what. Everything the platforms report sits underneath those as tactical detail, useful for choosing between two ads, and never for deciding whether a channel deserves to exist.
You will still never know exactly which ad made the sale. That was never available. What is available is knowing whether the whole thing is making money, which is the question that mattered all along.
We spend our days inside Google Ads, Meta and LinkedIn accounts watching owners get talked out of good channels by bad numbers. If you would rather have someone sort out what your reports are actually telling you, that is what we do. And if you would rather run it yourself, take the three numbers above and you are ahead of most people already.
