Nobody writes about the messy middle, the actual decision process between "we got a bad result" and "we did the right thing about it." Here's the playbook I wish someone had handed me.
The tracking industry is constantly improving its ability to pinpoint problems. You can run a geo holdout, isolate causal lift, and calculate the incremental ROAS, and there it is: a number below 1.0, staring back at you from a channel that has been eating a meaningful slice of the budget.
Having lived through this a few times, here's what I've learned about the part nobody writes about.
The instinct after an ugly result is binary: the channel failed, cut it.
Sometimes that's right. But an incrementality test tells you the channel wasn't incremental as you were running it. We should be careful of overgeneralizing, responding to that specific case should take less time than the overgeneralized response.
A negative result is still useful information, but a point estimate below 1.0 isn't enough by itself. If the test was underpowered, poorly randomized, contaminated by spillover, or executed during an unusual period, the real conclusion may simply be that the evidence is inconclusive.
One of the biggest mistakes isn't what happens after the test, it's what happens before it starts. Before the experiment launches, always align on four things:
Measurement without predefined decisions is just expensive reporting.
Once you have a result, several different problems can lie behind one bad number.
Saturation: the channel works, but you've pushed beyond the efficient frontier. Early dollars generated meaningful incremental demand; the last dollars bought very little. The answer is not to stop spending, the response is to spend less.
Pull back spend toward a point where demand becomes incremental again, and retest. Remember that incrementality measures the value of the next dollar, not your historical average.
Execution: spend was not positioned correctly, the creative, audience, bidding strategy, landing page, or frequency was off, but the channel may still be working. Make meaningful changes, not cosmetic ones, and retest. If you're continuously adjusting the campaign while testing, you won't know what the experiment actually evaluated.
Structural mismatch: the audience doesn't fit your product. The buying journey doesn't align with the channel. Or it's fundamentally a demand-capture channel taking credit for demand that would have arrived anyway. This is the clean kill.
Strategic exceptions: a channel may have poor short-term iROAS but still build value in ways you can't see during a two-month test, brand building, market expansion, new product launches, or acquiring unusually valuable long-term customers.
My rule is simple: strategic exceptions have to be declared before the test, along with the longer-term KPI they'll be judged against.
Strategy changes made after disappointing results indicate a lack of capability to execute a strategy.
The question we should be asking is whether a channel is the best use of the next marketing dollar.
An iROAS of 0.9 may still deserve funding if every realistic alternative produces 0.6, or it's a defensive strategy to not lose your brand spot on search results. Likewise, an iROAS of 1.5 may deserve cuts if another channel consistently delivers 3.0.
Incrementality should improve capital allocation across the portfolio. It shouldn't become a pass/fail test for individual channels, think reallocation, not elimination, most of the time.
When the answer is a cut, how you cut matters almost as much as the decision. A channel should be gradually cut, and the impact of the cut should be monitored.
A decision has to be made about where the marketing funds get redirected. Otherwise, a cut turns into a negotiation, and the negotiation always favors the same marketing strategy. Pre-decide where the money goes.
The hard part of all of this is that there's no complex analysis required to determine any of it.
What's hard to accept is that poor iROAS puts many channels, and people, in the loser category, and organizations are good at shielding themselves from being put in that category. The channel owner insists on more optimizations. The agency presents yet another deck. Someone suggests waiting until after peak season. Finance demands three more months of data.
I've watched rigorous, well-designed experiments change absolutely nothing because nobody attached a decision deadline to the result. So attach one. Set the decision rule before the test even launches: if iROAS comes in below X, action Y gets taken by this date.
It's the highest-leverage conversation in the entire testing process, because it transforms measurement into accountability.
In the majority of cases, companies don't waste marketing dollars because they lack measurement. They waste them because they fail to act on measurements they already trust.
A bad iROAS, handled well, is one of the best things that can happen to a marketing budget. It's found money. The experiment found it. Your organization's willingness to act is what determines whether you get to keep it.