Situation. A durable-goods e-commerce company running a single acquisition channel — paid search — with poor results and a widespread internal belief that the channel simply did not work for their category. The obvious recommendation was to abandon it and look elsewhere.

The number nobody had computed. Whether the channel was failing or the execution was. These produce identical-looking results and lead to opposite decisions. A structural review of the account — bid strategy, term selection, negative keywords, budget allocation across campaigns — indicated the problem was operational, not structural.

The test and how we made it readable. We replaced the operator before killing the channel. Once paid search was performing, it became the baseline against which new channels could be judged, and we tested two: paid social and affiliate.

Each required different measurement. Paid social offers no clean incrementality read, so we set the required return on ad spend well above the true hurdle rate to absorb the platform's known over-attribution — clearing an inflated bar on flattering numbers means it works on honest ones. Affiliate needed no such adjustment, because attribution there is clean by construction: affiliates only get paid for sales they can be shown to have caused.

Result. All three channels performed. Revenue tripled over two years, driven primarily by the two new ones.

What transfers. Do not kill a badly run channel before fixing the operator — the failure modes look the same from the outside and the cost of the mistake is asymmetric, because a channel written off wrongly is rarely revisited. And where only a dirty measurement is available, raise the threshold rather than abstain from testing. Demanding a return well above your real hurdle is a workable substitute for measurement you cannot get.

‍