Same journeys, same window, opposite errors

Here is the comparison. First click asks "who started this sale?" Last click asks "who was standing there when it closed?" Both are looking at the exact same set of purchases. If attribution error were random, some weeks Google would gain credit under last click and some weeks it would lose it. That is not what happens. In this week's data, Google's ROAS reads 1.53 on first click and 1.71 on last click. Meta reads 0.39 on first click and 0.30 on last click. Last click hands Google a 12% raise and hands Meta a 23% pay cut, on the same customer journeys. 

The pattern holds across the grid. Microsoft, a classic bottom-of-funnel search channel, gains the most:2.32on first click becomes2.79on last click, a 20% boost from the measurement method alone. YouTube, a discovery channel, loses 16%. The channels that introduce your brand to a stranger get their credit taken and handed to the channels that catch that same stranger a week later when they search for you by name. Now connect this to cost: Meta acquired new customers at$93nCAC this week versus Google's$143. The channel last click punishes is the cheaper place to buy a new customer. The channel last click flatters costs 54% more per new customer. If you rebalance budget on last click ROAS, you move money toward the expensive closer and away from the cheap opener, and your new customer growth stalls while your reported ROAS improves.

Is this a calendar artifact or a structural pattern? Structural. There is nothing special about a mid-August week, no holiday, no seasonal spike, and the direction of the credit shift matches how these channels function: search intercepts demand that already exists, social and video create it. Because the cause is the mechanics of the channels themselves, expect this same directional bias in your account next week and the week after, at whatever magnitude your mix produces. And because this data is aggregated across hundreds of accounts, no single brand's promotion can move these numbers. Pinterest is marked as a thin channel this week and is not used to anchor any conclusion here.

03 / New Customer Acquisition 

The full acquisition picture, by channel

Look at the % new column next to the credit shift from section 02. Meta is 89% new customers, the highest share of any major paid channel, and it is the channel last click penalizes hardest. Microsoft is 58% new, the lowest, and it is the channel last click rewards most. The measurement method is systematically shifting credit from the channels doing your acquisition work to the channels harvesting demand that already existed. Meta's nCAC also improved about 4% week over week while its new-visit conversion ticked up, so the penalized channel got cheaper, not worse. 

04 / First Click vs Last Click 

Who starts the sale vs who takes the bow

This is the week's grid to sit with. Every search channel gains credit when you switch from first click to last click. Every social and video channel loses it. Six channels, one direction of error. If someone tells you last click is "close enough," ask them close enough in which direction, because the answer determines whether you are quietly starving the channels that bring you new customers. 

 Nobody performed differently. The credit moved.

Take every order in the trailing 90 days and give 100% of the revenue to the first click that touched it. Then run it again and give 100% to the last click. Same orders, same revenue, same spend in the denominator. The only thing that changes is who gets paid on paper. Here is what that swap does to each channel, with every channel indexed to its own first click result so you can see the direction rather than the absolute level. 

Meta loses 30% of its measured return on the switch. Google gains 12% and Microsoft gains 22%. YouTube loses 16%. The pattern is not random and it is not about ad quality. Social channels get found early in a buying decision, and search gets typed in at the end, so a last click model quietly hands social’s work to search and calls it search performance. The acquisition grid says the same thing from the other direction: 90% of Meta’s customers this week were brand new, against 66% on Google and 58% on Microsoft. The channel bringing you the most first time buyers is the channel last click punishes hardest.

Is this a calendar effect? The first week of August is a quiet stretch, past the July promo cycle and not yet into back to school and Labor Day, so a soft social week would be an easy explanation. I do not think that is what this is. The first click versus last click comparison runs on a rolling 90 day window, not on the week, and the direction of the gap has been consistent: prospecting channels open sales and search closes them. That makes it structural, and structural means it will be true again next week. The number I would treat as genuinely calendar sensitive is Google’s nCAC, up 12% week over week to $148, with the click price up about 5% and new customer conversion down about 4%. That is a mix and auction move worth watching, not a verdict.

03 / New Customer Acquisition 

The full acquisition picture, by channel

Read the last two columns together. Meta converts new visits to new customers at 2.5% and delivers new customers at $96, and 90% of what it delivers is a first time buyer. Google converts at 2.3% but a new customer costs $148, and a third of its customers were already yours. Microsoft is the most expensive new customer in the set at $160 and the least new at 58%. YouTube and TikTok are prospecting channels that are honest about it: 82% and 77% new, at $395 and $344. Pinterest is marked with an asterisk because it produced very few new customers this week, so its rates swing wildly and it anchors nothing in this issue. 

04 / First Click vs Last Click 

Who starts the sale vs who takes the bow

The credit gap column is last click minus first click. Negative means the channel starts sales it does not get paid for. Meta and YouTube are both under credited, Google and Microsoft are both over credited, and TikTok is close to even. Microsoft has the widest single gap at 0.56, which is a good reminder that a 3.07 ROAS on a closing channel is not the same claim as a 3.07 ROAS on a channel that found the customer. If you only ever look at one of these two columns, you are not measuring performance, you are choosing a winner in advance. 

 Rank by ROAS and the new-customer share runs the other way

Take this week's channels and line them up by ROAS, best to worst. Microsoft leads at 2.76, then Google at 1.54, then the video and social channels trail with Meta at 0.51 and TikTok at 0.27. Standard picture. Now add one column next to it: the share of each channel's customers who are actually new to the brand.

The two columns point in opposite directions. Microsoft, the ROAS leader, is only 59% new. Google is 68% new. Meta, near the bottom on ROAS, is 89% new. YouTube is 85%. The higher a channel sits on ROAS, the smaller the fraction of its customers who are people you did not already have.

This is not a coincidence, it is how the metric is built. A channel like Google or Microsoft captures a lot of demand at the moment of purchase, and a good chunk of that demand is existing customers coming back to buy again. Those repeat orders are cheap to win and they land in the channel's revenue, so its ROAS looks strong. Meta and the video channels spend most of their effort in front of people who have never bought from you, which is harder and converts lower, so their ROAS looks weak even though they are doing the actual work of growing the business.

Put plainly, ROAS quietly rewards a channel for reselling to people you already have. The two dominant channels this week make the point on their own. Google's ROAS is three times Meta's, but roughly a third of Google's customers are repeat buyers, against barely one in ten on Meta. If you shift budget toward the higher ROAS number, you are partly paying to harvest demand you would likely have captured anyway, and starving the channel bringing in new humans.

The honest caveat, because this cuts both ways. A high repeat share is not a sin, closing existing demand efficiently is a real job and someone has to do it. The point is not that Google is bad and Meta is good. It is that ROAS alone cannot tell you which job a channel is doing, and if growth is your goal, the share of new customers belongs right next to the ROAS number, not three columns away. This is a quiet, no-drama week across the board, so read this as a standing structural feature of the data, not a one-week event.

03 / New Customer Acquisition 

The full acquisition picture, by channel

 Sorted by new share this time, so the pattern from the chart reads top to bottom. Meta and the video channels sit up top doing the prospecting, and by the time you reach the bottom of the list you are looking at the channels with the strongest ROAS. Meta held steady with conversion up a couple of points, the leveling off mentioned earlier. YouTube and TikTok conversion slipped on small volume, so treat those as wobbles. Pinterest ran thin again and carries an asterisk. 

04 / First Click vs Last Click 

Who starts the sale vs who takes the bow

This grid reinforces the same idea from a different direction. The channels that close, Google and Microsoft, gain ROAS on the last click model, the one closest to what platforms report. The channels that open, Meta and Pinterest, lose it. So the channel doing the most new customer work is also the one most likely to be under credited when you look at platform numbers. Two ways of measuring, same conclusion about who gets shortchanged. 

ROAS ranks your channels backwards

Here is the trap. You open your dashboard, you sort by ROAS, and you make a decision. Microsoft is at 2.60. Google is at 1.59. Meta is sitting at 0.52 and looks like a problem. The obvious move is to pull from Meta and feed the winners.

Now sort the exact same channels by nCAC, the cost to acquire a brand new customer, verified against first order IDs. Meta is the cheapest in the set at $84. Google is $117. Microsoft is $141. The order you just trusted has flipped on its head.

So which sort is right. Both, and that is the point. ROAS is not lying to you. It is doing something worse. It is folding two very different numbers into one and hiding both.

Look at what ROAS is actually made of. Google closes a lot of demand that other channels created, so it books high revenue against its spend and posts a strong ROAS. Meta introduces people who have never heard of the brand, many of whom buy something small first, so it posts a low ROAS even while it is doing the hardest and most valuable job in the funnel, finding new humans.

The two numbers that ROAS smears together are cost and value. Split them apart and the fog clears. Cost is nCAC, what you pay to acquire a new customer. Value is nLTV, what that new customer becomes over the next year. This week Meta is the cheapest to acquire at $84 but the lowest one year value at $97. Microsoft is the most expensive to acquire at $141 but the highest one year value at $341. Neither of those facts survives inside a single ROAS number.

One honest note, because this report only works if the numbers are trustworthy. This is the week after the July 4 selling season, so conversion softened a little across most paid channels, the normal post holiday give back. But the story this week is not a calendar story. It is structural. Search and Microsoft close demand and read high on ROAS. Social and video prospect and read low. That pattern does not need a holiday to show up, and it will look the same next week. That is exactly why you cannot budget on ROAS alone.

 03 / New Customer Acquisition 

The full acquisition picture, by channel

Meta is carrying the prospecting load. It brought in more than 40,000 new customers this week, and89%of the customers it touched were brand new to the brand. That is the profile of a channel finding people, not milking a list. It is also the channel your dashboard told you to cut.

Conversion softened for most channels this week, the expected step down after the holiday selling week. YouTube was the exception, up 11%. Pinterest ran on a tiny base this week and is marked with an asterisk, so I am not using it to anchor anything.

 04 / First Click vs Last Click 

Who starts the sale vs who takes the bow

This grid is the mechanism behind the ROAS trap. Last click is the model closest to what the platforms report, and it inflates Google and Microsoft, the channels that close, while it shrinks Meta and Pinterest, the channels that open. Meta gives back 0.14 of ROAS on the last click. Microsoft gains 0.61. Judge a discovery channel on last click alone and you cut the thing that started the sale, then wonder why new customer growth stalled.