Attribution Disagreement Across Three Channels
An e-commerce apparel brand spends $60K/month across paid search, paid social, and email. Last-click attribution credits paid search with 62% of conversions, social 18%, and email 20%. The team is about to shift budget from social to search based on this. A marketer notices that branded search terms drive most search conversions, that social is the first touch in a large share of new-customer journeys, and that email mostly reaches existing customers. Blended CAC is $41; the target is $35. A geo holdout test in two matched markets paused social for four weeks and saw total conversions fall 9% in the test markets versus control, with branded search conversions falling alongside.
1. Explain why the last-click view misleads here and what the holdout test reveals about the true role of paid social.
Read the worked answer
Last-click credits the channel that closes, not the channel that creates demand. Branded search conversions are largely demand that another touch generated - people search the brand name because they were already persuaded. Crediting search for those conversions overstates search's causal contribution and understates social's.
The geo holdout is the decisive evidence because it measures incrementality: pausing social cut total conversions 9% in test versus control, and branded search conversions fell with it. That means social was feeding branded search demand - social's true contribution is larger than its 18% last-click share, and search's is smaller than 62%. Shifting budget from social to search based on last-click would cut the demand that search harvests.
So the proposed reallocation is backwards. The holdout shows social is incremental; last-click shows only who got the last cookie.
2. Given a $41 blended CAC against a $35 target, what should the team do with the budget, and what would they measure to confirm the decision?
Read the worked answer
Do not cut social on last-click logic. Use the incrementality evidence to reallocate toward what the holdout proved creates demand.
Actions:
- Hold or modestly increase social, but optimize it on new-customer acquisition (first-touch, new-customer CAC) rather than last-click ROAS, since social's value is demand creation.
- Reduce over-crediting of branded search by separating branded from non-branded search in reporting; judge search on incremental branded lift and non-branded acquisition, not on branded closes it didn't cause.
- Treat email as retention - measure it on repeat rate and LTV, not blended CAC, so it isn't wrongly penalized or credited.
Measurement to confirm: run further geo holdouts per channel, track new-customer CAC by channel, and watch whether social reductions produce the 9% total-conversion drop seen before. Close the $41→$35 gap by cutting genuinely non-incremental spend (likely some branded search) and improving social's new-customer efficiency, not by starving the demand engine.