The attribution fantasy
Marketing attribution has long held the promise of something that was never fully achievable: a clear, causal line between every marketing activity and every sale. The ideal was a system where you could look at any customer and trace exactly which touchpoints influenced their decision to buy, weight each touchpoint's contribution appropriately, and adjust investment accordingly.
For a period, roughly 2010 to 2018, when cookies were reliable and cross-device tracking was still functional, the promise was partially delivered. Multi-touch attribution models got reasonably good at assigning credit across digital touchpoints within a single-device journey. It was imperfect but useful enough to guide decisions.
Then the environment started to unravel. iOS 14.5's App Tracking Transparency, released in early 2021, disrupted Facebook and Instagram attribution dramatically. Apple's Mail Privacy Protection changed how email open tracking worked. iOS 17's link tracking protection removed UTM parameters from links opened in Mail, making email-driven web session attribution unreliable for a significant portion of audiences. Chrome has been preparing third-party cookie deprecation for years and has finally begun moving on it. The result is a tracking environment that is structurally more difficult than it has been at any point in the past decade, and the trajectory is toward more privacy, not less.
The goal is not perfect attribution. It never was. The goal is good enough information to make better decisions, and that is still achievable.
What the data environment actually looks like now
In the current environment, a realistic picture of what is and is not measurable looks something like this. Direct attribution from ads to conversion is meaningful but overcounted by the platforms, often by significant margins. Last-click attribution in analytics platforms undervalues everything that happened earlier in the customer journey, and misses touchpoints that left no trackable footprint, word of mouth, podcast listening, private sharing, organic social that did not result in a direct click. Dark social, the enormous volume of content shared in private channels like WhatsApp, Slack, direct messages, and email, is essentially invisible to standard analytics but represents a major driver of discovery and consideration for many brands.
The honest response is not to find a better model that makes the invisible visible. It is to acknowledge what can and cannot be measured with confidence, use the measurable signals you do have wisely, and supplement them with measurement approaches that do not depend on perfect individual-level tracking.
A more honest measurement framework
The framework that works best in the current environment combines several layers. Platform-level data, the conversion numbers from paid channels, is useful as one input, with the understanding that it overcounts and requires appropriate scepticism. Analytics platform data, GA4, or a platform like Chronos Analytics, gives you traffic, on-site behaviour, and goal completion data that tells you about the quality of the journeys happening on your own properties. CRM data, where did closed deals come from, what touchpoints are logged in the customer record, gives you a ground truth from the sales side that can be compared against analytics-side attribution. And survey-based attribution, simply asking customers "how did you hear about us?", captures dark social and word of mouth that no tracking technology can see.
None of these alone is sufficient. Together, they provide a triangulated view that is honest about its limitations but useful enough to make better budget decisions than any single model could.
The decisions that still need to be made
The practical question is: given imperfect attribution, how do you allocate a marketing budget? The answer is through a combination of the data you do have and first-principles reasoning. Channels with well-understood, replicable, measurable unit economics, branded search, high-intent search terms, email to a first-party list, get a stable allocation based on known performance. Channels whose contribution is real but less precisely measurable, organic content, PR, brand building, get an allocation based on leading indicators (branded search growth, direct traffic trends, share of voice) and a reasoned case for their contribution to the funnel. Experimental channels get a small protected budget and are evaluated on a longer timeframe with looser criteria.
This is not as satisfying as a single attribution model with a confident number. But it is more honest, and it produces better decisions than trusting any single model's output uncritically in an environment where every model is working with significantly incomplete data.
Investing in what attribution misses
One of the underappreciated consequences of over-reliance on attributable channels is systematic under-investment in channels that work but cannot be easily attributed. Brand investment, word-of-mouth programmes, community building, earned media, all of these create real commercial value that does not reliably show up in attribution models. Teams that have cut them because the attribution is not clean are likely paying higher CAC and seeing slower growth than they would with a balanced approach.
Building a measurement framework that can partially account for these channels, even imperfectly, is one of the more valuable investments a marketing technology function can make right now.