Why marketing attribution cannot prove what caused a purchase

Better tracking will not solve attribution’s deeper problem. A perfectly recorded click can tell a marketing team that an interaction happened. By itself, it cannot establish that the interaction caused a customer to buy. That boundary should shape how executives read attribution reports.

For CEOs, marketing leaders, and CFOs, this is a measurement problem. A useful response is to combine signals showing where marketing was present and how buyers or sellers used it. Those signals can provide directional evidence for management decisions. Evidence of presence and use does not establish causal lift.

Better tracking cannot establish causation

Tracking asks whether an event was recorded and associated with a person, account, or opportunity. Causal measurement asks whether the outcome would have changed without the marketing activity. Attribution rules can distribute credit across recorded touches, but that calculation cannot answer the causal question.

An attribution chart shows part of the buying process

Consider one B2B purchase. A prospect can discover an issue through content, discuss it internally, ask an AI assistant for options, talk to a salesperson, read material that salesperson sends, raise an objection with another employee, and eventually complete a form that enters the CRM. Depending on the company’s systems, only some of those events may appear in attribution data. The dashboard describes the events recorded under its collection rules.

First-touch identifies the earliest recorded interaction, last-touch the final recorded interaction before a defined conversion, and weighted-touch divides credit among several recorded events. Those labels become risky when executives read them as a reconstruction of the buyer’s decision process. A lead-source field has the same limitation: it is a database property populated under specific measurement rules. A buyer may encounter several pieces of content and people before generating the event stored in that field.

Missing observations should not automatically receive credit either. An untracked conversation could have been decisive or irrelevant, and an observed click has the same causal limitation. Observation establishes that an event occurred. Determining whether it changed the outcome requires stronger evidence.

Change the question from credit to evidence of presence and use

A more defensible operating question asks where marketing appeared at relevant decision points and whether buyers or sellers used it. A decision point is a stage where a buyer needs information to continue evaluating a purchase, such as resolving an early objection, establishing value during a deal, or building trust near the close. This approach gathers evidence about marketing’s role in buying and selling activity. It avoids treating a rule for allocating recorded touches as a causal estimate.

Here, “contribution” means directional evidence that marketing material was present or used during the sales process. A salesperson might report that a case study helped answer an objection, while a customer later mentions the same study when describing the purchase. Together, those observations show that the material entered the process. They still cannot establish what would have happened if the case study had never existed.

That boundary supports questions the available records can answer. Executives can ask where material was available, where sales used it, and where customers reported encountering it. Tracking records, sales reports, and customer accounts can answer different parts of that question. Management gets a richer evidence base without reducing it to a single causal credit number.

Attribution becomes one input to management decisions

Existing attribution infrastructure can remain useful within this approach. For marketing operations, provenance is an important design principle. Provenance means recording where a claim came from: what a system observed, what a salesperson reported, what a customer said, or what an AI analysis inferred. Keeping those categories distinct makes disagreements visible and reduces the risk that an inference will later be treated as a directly observed event.

A CFO can then evaluate marketing evidence at the level each measurement supports. Attribution data can document recorded interactions, while usage records and reported journeys add context around activity the attribution system may not represent. A revenue number produced by an attribution rule does not gain causal authority simply because it appears in a dashboard. Claims of incremental revenue require methods capable of supporting causal inference.

Key takeaways for leaders

Better tracking does not prove causation: Attribution can show that a marketing interaction occurred, but not whether it caused a purchase. Leaders should reserve incremental revenue claims for methods designed to support causal inference.
Treat attribution as a partial view: First-touch, last-touch, weighted-touch, and lead-source data describe recorded events rather than the complete buying process. Executives should interpret these measures within their collection rules and limitations.
Measure presence and use: Combine tracking data with evidence of how buyers and sellers encountered and used marketing material at relevant decision points. Keep these signals directional rather than treating them as proof of causal lift.
Make attribution one input to decisions: Preserve the provenance of system observations, sales reports, customer accounts, and AI inferences instead of blending them into one metric. This gives management a broader evidence base while keeping causal claims separate from observed or reported activity.