Conversion Attribution: Where Your Results Really Come From

End of the month, time to close the report. Meta Ads says it drove 620 sales. Google Ads claims 410. GA4 shows 780 total conversions. And your e-commerce platform recorded 850 actual orders. Add up what each channel claims and you get more sales than the company actually made. Is somebody lying?

Nobody is lying. Each tool is answering a different question, under different rules, and calling everything a "conversion". The name of this problem is attribution.

Conversion attribution is the method that determines which channel or ad gets credit for a sale or lead. Since most customers interact with several touchpoints before converting, the attribution model you choose decides how that credit is split among them, and it directly changes the reported ROAS of every channel.

What is conversion attribution in practice?

Picture a typical journey: someone sees a sponsored Reels video on Tuesday, googles your brand the following week, clicks the search ad, joins your email list, and buys ten days later from a promo email. Three channels participated. Which one "generated" the sale?

The honest answer is uncomfortable: all of them and none of them, in proportions nobody knows for sure. Conversion attribution is the attempt to turn that uncertainty into an operational rule. It is not exact science; it is an accounting convention. And like any convention, it serves some decisions well and others very poorly.

This matters because the model changes the number on your dashboard. The same month, with the same spend and the same sales, can make a channel look like a hero or like wasted budget depending purely on the credit rule. If you allocate budget from a single report without knowing which model sits behind it, you are deciding in the dark.

What are the main attribution models?

The classic models split into fixed rules and statistical models. It is worth knowing all of them, even the retired ones, because they still show up in CRMs, BI tools, and agency debates.

Model	How it splits credit	When it makes sense
Last click	100% to the final touch before conversion	Simple operations, short cycles, closing-channel focus
First click	100% to the first touch of the journey	Evaluating discovery and top-of-funnel channels
Linear	Equal credit to every touch	Long journeys where everything counts a little
Time decay	More credit to touches closer to the conversion	Mid-length sales cycles, promotions
Position-based	40% first, 40% last, 20% to the middle	Valuing discovery and closing at the same time
Data-driven	An algorithm estimates each touch's weight by comparing converting and non-converting journeys	High conversion volume; Google's current default

A worked example. A three-touch journey (Instagram, Google search, email) ends in a $1,000 purchase. Under last click, email takes the full $1,000. Under first click, Instagram takes it all. Under linear, each channel gets $333. Under position-based, Instagram and email get $400 each and search keeps $200. Same sale, four different stories.

Here is a detail many teams missed: in 2023 Google removed first click, linear, time decay, and position-based from Google Ads and GA4, citing low adoption (fewer than 3% of conversions used those models, according to Search Engine Land). Only last click and data-driven survived. In practice, Google decided the debate over manual rules was finished and pushed everyone toward its own algorithmic model.

Why do Meta, Google, and your CRM never show the same number?

Back to the opening example. The differences are not bugs; they are different rule sets operating on the same reality.

First, attribution windows. By default, Meta Ads counts conversions up to 7 days after a click and 1 day after an ad view. So someone who watched your video, never clicked, and bought the next day becomes a Meta conversion. GA4 cannot see that view, so it credits the sale to the last-click channel, which might be organic search. Both reports are "right" within their own rules.

Second, every platform takes full credit for itself. Meta and Google do not talk to each other. If both touched the same sale, both count the entire sale. That is why the sum of channel reports almost always exceeds the real total in your CRM.

Third, signal loss. Since iOS 14.5, cross-app tracking requires explicit consent. The global average opt-in rate sits at 35%, according to Adjust (Q2 2025); Brazil, at around 50%, is among the most permissive markets, and even there half of iPhone users vanish from the radar. Platforms fill that gap with modeled conversions: statistical estimates, not measurements. Part of the number you read in the ads manager is a projection.

A mini-case for scale. An online fashion store does $60,000 a month in revenue on $18,000 of ad spend. Meta reports 620 conversions, Google Ads reports 410, the sum is 1,030, yet the store recorded 850 orders, of which roughly 180 came from organic repeat purchases and WhatsApp. The overlap is obvious. If the manager cuts Meta because "Google converts cheaper", they may be cutting exactly the channel that starts the journeys Google closes. That mistake is expensive and invisible in any single report.

How do you choose an attribution model?

The right question is not "which model is most accurate" but "which decision am I trying to make". Different models answer different questions.

• To compare campaigns inside one platform, use its default model (data-driven on Google, 7-day click on Meta) and stay consistent. Comparing campaign A and B with the same yardstick works, even if the yardstick is imperfect.
• To allocate budget across channels, trust none of their reports. Use a total-efficiency view: revenue divided by total ad spend (often called MER). For the store above: $60,000 / $18,000 = 3.3. That figure will not tell you which channel is best, but it tells you whether the whole machine is healthy, and it cannot be inflated by double counting.
• To understand funnel roles, look at conversion path reports in GA4 and at each channel's share as an opening or middle touch, not just as a closer.

What about Google's data-driven model? It beats last click for distributing credit inside the Google ecosystem, but it has two practical limits: it is a black box you cannot audit, and it only sees what Google sees. A sale influenced by a Meta ad nobody clicked remains invisible to it. The whole approach breaks down in businesses with long cycles and human-assisted sales, where much of the journey happens outside any pixel.

A pragmatic 4-step approach

1. Accept imperfection. No model will reveal "the truth". The goal is to reduce decision error, not eliminate it.
2. Measure the whole before the parts. Track MER and blended CAC every month. If overall efficiency drops while every channel claims to be efficient, attribution is masking the problem.
3. Standardize UTMs and windows. Half of all attribution arguments die when everyone uses the same campaign naming and knows which window each report uses.
4. Run incrementality tests for big bets. Before cutting or doubling a major channel, run a geo test or a controlled pause for 2 to 4 weeks and watch total sales. It is the closest a marketing team gets to causality without becoming a statistics department.

Frequently asked questions

What is the best attribution model?

There is no universal "best". To compare campaigns within one platform, data-driven usually beats last click. To allocate budget across channels, no platform model is enough: combine a total metric such as MER with incrementality tests.

What is the difference between an attribution model and an attribution window?

The model defines how credit is split among the touches in a journey. The window defines how long after a touch a conversion still counts. They are independent settings, and both change the numbers in your reports.

Why does Meta Ads show more conversions than Google Analytics?

Mostly because of view-through conversions, which GA4 does not count, plus different windows and the modeled conversions Meta estimates for untracked users. Each tool applies its own rules to the same sale.

What is data-driven attribution?

A statistical model that compares converting journeys with similar non-converting ones and estimates each touch's contribution. It has been the default in Google Ads and GA4 since 2023, but it works as a black box and only considers touchpoints Google can observe.

Does attribution still make sense after iOS 14.5?

Yes, with adjusted expectations. With a 35% global opt-in rate for app tracking, reports increasingly depend on statistical modeling. Attribution has become a trend compass rather than an exact scoreboard, and that is how it should be used.

Stop chasing the perfect number

The hunt for the definitive attribution model usually ends in paralysis or false confidence, and both are expensive. The more honest path is to operate with multiple lenses: total efficiency for business health, platform models for tactical optimization, and tests for the big decisions. Understanding what each number can and cannot tell you already puts you ahead of most of the market.

That cross-referencing of sources is exactly what slows lean teams down: paid traffic data, GA4, and the CRM live on separate screens. With Sherlok, you connect those sources and ask questions in plain language, like "which channel started the journeys that converted this month?" or "is my true ROAS above break-even?", and get the analysis ready, no spreadsheet required.