Attribution Dispute After a Channel Mix Shift
An online home-goods merchant moved 40% of its paid search budget into influencer affiliate codes six months ago. Since then, reported revenue is up 18% but net profit is flat. The marketing team credits the affiliates; the CFO is skeptical. The affiliate program pays a 12% commission on any order using the code, including orders from customers who were already in the loyalty program. The merchant's analytics show that 60% of affiliate-coded orders came from returning customers, and average order value on coded orders is 9% lower than non-coded orders. There is no holdout or control group. The merchant asks whether to double the affiliate budget.
1. Evaluate whether the 18% revenue lift can be attributed to the affiliate channel, and identify what the data does and does not tell you.
Read the worked answer
The 18% revenue lift cannot be attributed to the affiliate channel with the data available, and the practitioner should say so plainly.
What the data does tell you:
- 60% of coded orders are from returning customers. Those customers already had a relationship with the brand; some fraction would have bought anyway. The code is, for those orders, a discount on a sale that was already going to happen - pure margin leakage at 12% commission plus the discount.
- AOV is 9% lower on coded orders, which is consistent with the code being used as a price-shopping tool rather than as an acquisition driver. It is not proof, but it points the same direction as the returning-customer finding.
What it does not tell you:
- How many of those orders were incremental. Without a control group, you cannot separate 'the affiliate caused this sale' from 'the sale would have happened anyway.' The 18% revenue lift is a correlation, not a causal estimate.
- Whether the 40% of new customers acquired via affiliates are high-LTV. That is the one cohort that could justify the spend, and it is not broken out.
Conclusion: the flat profit despite revenue growth is exactly what you would expect if a large share of the spend is subsidizing existing demand. That is a hypothesis the data supports but does not prove.
2. Propose a measurement design and a budget decision that the CFO could act on, given no budget for a long experiment.
Read the worked answer
You do not need a long experiment to get an actionable answer; you need a holdout and a cohort split.
Measurement design:
- Geo or audience holdout. Hold the affiliate code out from a random 10–15% of the eligible audience for 3–4 weeks. Compare conversion and revenue between holdout and exposed groups. The difference is your incremental lift. This is the single most valuable number and it is cheap to run.
- Split the code by cohort. Issue one code visible to new customers only and one to existing/loyalty customers. Track them separately. New-customer-coded orders are plausibly incremental; loyalty-coded orders are almost certainly not.
- Compute contribution margin per coded order, not revenue: (AOV × gross margin) − commission − discount. Compare to the same figure for non-coded orders. If contribution margin is negative or below non-coded, the channel is destroying value regardless of revenue.
Budget decision: do not double the budget. Hold it flat for one quarter while the holdout runs. Then reallocate toward the slice that shows incremental lift - most likely new-customer acquisition - and stop paying commission on loyalty-coded orders (or cap it). The framing for the CFO: 'We will move from a revenue-credited channel to an incrementality-credited channel, and we will know within a quarter whether the affiliate spend earns its commission.' That is a decision the CFO can approve today without a long experiment.