A marketing team moves from last-click attribution to a more sophisticated multi-touch attribution model, distributing conversion credit across every touchpoint in a customer's journey rather than crediting only the final one. This is presented, reasonably, as a more complete and rigorous view of the customer journey — and it's easy to mistake for a more causally accurate one. It isn't, necessarily: multi-touch attribution still distributes credit according to a chosen rule, not according to evidence that specific touchpoints actually caused the conversion.
Every multi-touch attribution model is built on an arbitrary crediting rule
Linear attribution splits credit evenly across every touchpoint in the journey. Time-decay attribution weights more recent touchpoints more heavily. Position-based attribution assigns extra weight to the first and last touchpoints specifically. Each of these is a defensible modeling choice, reflecting a different assumption about how influence probably works across a customer journey — none of them is derived from actual evidence, for the specific journeys being analyzed, about which touchpoints genuinely moved the customer toward converting and which happened to occur along the way without exerting real influence.
Different rules produce meaningfully different answers from identical data
Applying linear, time-decay, and position-based attribution to the exact same set of customer journeys will typically produce noticeably different credit allocations across channels — a channel that looks strong under time-decay attribution can look comparatively weak under linear attribution, and vice versa, purely as a consequence of which rule was chosen, without any change to the underlying customer behavior at all. If the choice of rule alone can swing which channel appears most effective, none of the individual rules can be said to be measuring the channel's actual causal contribution — they're each expressing a different assumption about how credit should be distributed.
Why multi-touch attribution feels more rigorous even though this problem persists
Multi-touch attribution genuinely improves on last-click attribution's most obvious flaw — ignoring every touchpoint except the final one, which clearly misses real influence from earlier touchpoints. But fixing that specific flaw doesn't fix the deeper, separate problem: correlational co-occurrence in a customer journey (a touchpoint happened to occur before a conversion) is still not evidence of causation (that touchpoint caused the conversion), no matter how many touchpoints the model accounts for or how sophisticated its weighting scheme is. Accounting for more of the journey makes the model more complete; it doesn't make the underlying credit assignment more causally grounded.
What would actually establish causal channel effectiveness
Incrementality testing — deliberately withholding a specific channel or touchpoint from a randomly selected holdout group and comparing conversion outcomes against a group that received the normal marketing exposure — directly measures the actual causal lift a channel provides, because it observes what happens with and without that channel under otherwise comparable conditions, rather than inferring influence from co-occurrence patterns in observational journey data. This kind of testing is more expensive and slower to run than an attribution model, which is exactly why attribution models remain the default despite not answering the causal question attribution is often implicitly assumed to answer.
What this means for how marketing effectiveness should actually be measured
- Treat multi-touch attribution as a description of correlated touchpoint co-occurrence, not a measurement of causal channel contribution
- Be aware that switching between reasonable attribution rules can meaningfully change which channels appear most effective, independent of any real change in customer behavior
- Use incrementality testing and holdout experiments, where the budget and channel allow for it, to establish genuine causal effectiveness for high-stakes channel investment decisions
- Present attribution model outputs to stakeholders with the actual caveat attached — that the credit split reflects a chosen modeling rule, not a discovered causal fact
Multi-touch attribution is a real improvement over last-click for describing the customer journey more completely. It remains, at its core, a rule for splitting credit — not a method for discovering which touchpoints actually caused anything.