Attribution Today: What Still Works – and What You Need to Replace

Attribution was long a promise: every conversion can be seamlessly assigned to the ad that triggered it – and budget simply flows to where the numbers look best. That promise is broken. Consent banners, browser protection and cross-device purchase journeys have thinned out the data basis on which the classic measurement logic rests. Still, it would be wrong to write attribution off completely: for certain tasks it continues to work well. This guide shows why the old logic is crumbling, what it still delivers, where it deceives you – and which pragmatic triad replaces it.

Why classic attribution is crumbling

Four developments are hollowing out seamless assignment at the same time:

Consent gaps. A relevant share of your visitors simply declines tracking. These users still buy – but their journey is invisible to measurement. Attribution therefore never sees the whole picture, only the consented excerpt.

Browser protection mechanisms. Modern browsers limit the lifespan of cookies and strip tracking parameters. For many measurement systems, a click from two weeks ago simply can no longer be connected to today’s purchase.

Cross-device journeys. The ad is seen on the phone in the evening, the purchase happens the next day on the laptop. Without a login bridge, the chain breaks – and the purchase appears as direct traffic out of nowhere.

Platforms measure themselves. Meta evaluates Meta’s contribution, Google evaluates Google’s contribution. Every platform has a structural interest in attributing conversions to itself – and does so generously. Added up, the self-reports regularly amount to more revenue than actually exists.

In everyday practice, this shows up as a familiar annoyance: Ads Manager, web analytics and the shop backend name three different figures for the same period – and each is correct according to its own logic. Whoever tries to bring these views fully into agreement loses time on a question that can no longer be answered cleanly.

The platforms increasingly fill the resulting gaps with modelling: where data is missing, conversions are estimated. That is better than nothing – but it turns a measurement into a projection. Part of the gaps can be mitigated technically: whoever runs the Pixel and the Conversions API in parallel and keeps an eye on their Event Match Quality gives the systems the best possible data basis – how to do that, our guide Setting up Meta Pixel and CAPI shows. But even the best setup changes nothing about the fundamental problem: the seamless, neutral assignment across all channels no longer exists.

What platform attribution still does well

You should not write platform measurement off because of this – you just need to know its strengths. Two tasks it continues to handle reliably:

Relative comparisons within a channel. Whether creative A performs better than creative B, platform attribution answers robustly. Both run under the same measurement conditions, with the same attribution logic and the same blind spots – the comparison stays fair, even if the absolute numbers are overstated. The same applies to campaign against campaign or audience against audience within the same account.

Short-term optimisation signals. The algorithm needs fast feedback to steer delivery and budget. For that, the platform conversions are the best available signal – not because they are the truth, but because they arrive fast, consistently and in large volume.

For the same reason, a short attribution window like 1-day click has established itself as a deliberately conservative convention: it only counts conversions that immediately follow a click, and thus overstates the ad’s contribution far less than long windows with view-through attribution. With our own brand SASSYCLASSY, for example, we steer towards a ROAS target above 4 on a 1-day-click basis – not because this window captures the full effect, but precisely because it is strict and the number therefore stays reliable.

What it can no longer do

Two tasks, by contrast, you must no longer entrust to platform attribution:

Cross-channel truth. Which channel deserves which share of total revenue, no platform can answer – each only sees itself and attributes generously to itself. This applies all the more the closer the channels interact: the ad sparks the interest, the search closes the purchase – and both platforms celebrate the same order. Whoever distributes budgets between channels based solely on the in-platform numbers lets referees decide who are playing in the match themselves.

Incrementality. Attribution answers which conversions had contact with an ad – not which would have failed to happen without the ad. The classic case is the retargeting ad shortly before a purchase that was planned anyway: perfectly attributed, zero additional revenue. Assignment is not causation.

The pragmatic triad as a replacement

What takes the place of the one big attribution truth is not a new tool, but a division of labour across three levels:

1. Platform attribution for optimisation within the channel. Within Meta, Google or TikTok, platform measurement decides which creatives, campaigns and audiences receive budget. Here it is strong, fast and sufficiently fair.

2. MER and the shop backend for the overall view. Whether your marketing works efficiently as a whole, MER answers: total revenue from the shop backend divided by total ad spend – immune to any attribution logic. Why this metric is the better north star, we explain in the article MER instead of ROAS.

3. Occasional incrementality tests as a reality anchor. Selective experiments – holdouts, geo splits, lift tests – check whether a channel really generates additional revenue or merely attributes successfully. How to run such tests without a data science team, the article Incrementality testing shows.

Each level answers its own question: what works within the channel? How efficient is the whole? And is the effect real? None of the three replaces the others – together they replace attribution of the old school.

In practice this also means: the three levels have different rhythms. Platform optimisation runs continuously, you read MER weekly, and incrementality tests you run selectively – for instance before a channel is about to be scaled significantly. Whoever treats all three at the same frequency overburdens their team and dilutes the significance of each individual level.

Post-purchase surveys: the inexpensive corrective

Complementing the triad, one tool is worth using that is technically banal and surprisingly revealing: the question “How did you discover us?” right after the purchase. A short selection in the checkout or on the thank-you page is enough.

The answers are methodologically messy – people misremember, name the last contact instead of the first, or click anything. But as a corrective they are valuable: if a channel barely appears in attribution but is constantly named by customers – typical for podcasts, TikTok or recommendations –, that is a strong hint at a blind spot in your measurement. The reverse applies equally. It is not about exact percentages, but about systematic discrepancies between what the measurement says and what customers say.

For the implementation, a few fixed answer options plus a free-text field are enough. What matters is keeping the question and the options constant over time – only then do shifts between the channels become visible. And: evaluate the answers regularly instead of merely collecting them. A survey nobody reads is as useful as a dashboard nobody maintains.

Attribution windows: choose deliberately, keep consistent

One final lever is regularly underestimated: how you handle the attribution window itself. Depending on the window, the reported conversions change considerably – the same campaign can look strong or weak depending on the setting, without anything having changed in reality.

Two rules follow from this. First: choose the window deliberately and in line with your purchase cycle – short, click-based windows for conservative steering, longer ones only where long decision journeys justify them. Second, and more important: stick with it. Whoever switches the window as soon as the numbers disappoint makes their time series unusable and lies to themselves in slow motion. A consistently measured, deliberately conservative window is worth more than the theoretically perfect one that keeps being readjusted.

Also document the decision in writing – including the reasoning. That sounds bureaucratic, but it prevents the creeping readjustment when numbers come under pressure and keeps reports comparable across months.

Conclusion

Attribution is not dead – it has been demoted: from a system of truth to an optimisation tool. Within a channel it remains useful, for creative comparisons and as the algorithm’s signal source. The questions that decide budgets – which channel truly contributes, how efficient is the whole – are answered today by the triad of platform attribution within the channel, MER based on the shop backend and occasional incrementality tests, complemented by post-purchase surveys as a corrective. If you want to know how robust your measurement really is – from the tracking setup to the window configuration –, our free account check shows you where you stand.