Second-Order Thinking: The Marketing Decision That Looks Right in Week One

The most expensive marketing mistakes look like wins for the first few weeks. The cost arrives later, usually after the person who made the call has already been praised for the first result. I don’t think that’s a character flaw in marketers. It’s what happens when the report that judges a decision only looks one step ahead.

Why every dashboard rewards the first order

A dashboard reports what moved in the period it covers. A discount, a retargeting push or a new ad format shows its result inside that period. The cost, if there is one, turns up in a later period, on a different line, often in someone else’s report.

That’s the whole problem. First-order effects are quick and easy to credit to someone. Second-order effects take months, and nobody owns them. So a team that runs on its dashboard keeps approving first-order decisions, without anyone ever intending to. A gamed metric works the same way: the 4.9 star rating that means nothing looks excellent in the week it’s reported.

Marks sets the bar plainly: “In order to outperform, your thinking has to be different and better.” He’s writing about investing, not marketing, and I’m borrowing the idea rather than claiming he wrote about our work. The borrowed version holds up, though. If your team reads the same kind of dashboard as every competitor and reacts to the same week-one numbers, you will make the same decisions they make. If you want a practical place to start, the marketing metrics I stopped reporting is a list of numbers that only ever answered the first-order question.

Three decisions where the bill comes later

These are the three I’d check most carefully on any marketing plan, because each one reports well early and costs later.

The discount

First order: sales jump in the week of the offer. It’s easy to report and easy to celebrate.

Second order: who actually bought? In a study of leading grocery brands published in the Journal of Advertising Research in 1994, Ehrenberg, Hammond and Goodhardt found that price promotions “generally did not affect a brand’s subsequent sales or brand loyalty”, and that the extra buyers during a promotion had almost all bought the brand before. It’s an old study of grocery brands, and I treat it as a strong prompt rather than a law for every category. The question it raises applies everywhere, though. If the people who bought on discount were going to buy anyway, you sold the same product to the same customers for less, and the week-one chart called it growth.

The activation-heavy budget

First order: performance spend reports results within days, so it keeps winning the budget argument.

Second order: Binet and Field’s work for the IPA, as summarised on the IPA’s blog, put the optimum split at 60:40 between brand building and activation in The Long and the Short of It in 2013, and 62:38 in Effectiveness in Context in 2018. The same piece notes that as the “tide of activation” rose, effectiveness fell, with effectiveness measured as the number of very large business effects reported in IPA cases. Those cases are established brands, not startups, so don’t lift the ratio as a rule. The startup version of the budget argument is a separate piece. What I take from it is simpler: the budget line that reports fastest isn’t automatically the one doing the most work.

The AI-optimised channel mix

First order: an optimisation tool moves money towards whatever converts this week, and the cost per result falls.

Second order: look at where large marketing teams are putting their media money. The Gartner survey behind the opening figures covered 401 marketing leaders, most at companies above $1 billion in revenue. It doesn’t say AI optimisation caused the shift away from loyalty and retention, and I won’t claim it did. What I will say is what an optimiser does. It moves money towards whatever reports a result fastest. Point one at a dashboard where conversion reports in days and retention reports in quarters, and you can guess where the money ends up. Gartner’s Ewan McIntyre put it more simply: “AI can help marketers optimize faster, but optimization is not the same as strategy.”

What changes when someone asks “and then what?”

None of this needs a framework. It needs one question asked out loud before sign-off, by someone in the room whose job that day is to ask it.

Marks’s second-level questions translate almost word for word. He asks “What does the consensus think?” and “How does my expectation differ from the consensus?” In a marketing review, that becomes: what does everyone here expect this to do in week one, and what do I expect it to do in month four that nobody has mentioned yet?

Applied to the three decisions above, it gets specific fast. For the discount: how many of these buyers had already bought from us at full price, and what will they expect to pay next time? For the budget: if brand spend drops this quarter, which number would show the damage two quarters from now, and is anyone looking at it? For the optimiser: what is it optimising towards, which line in the report would tell us if retention slipped, and who reads that line? None of these questions is clever. They just get skipped, because the week-one number is already on the screen and looks good.

Two habits make the question stick. Write the second-order risk next to the first-order target on the plan itself, so it lives in the document and not only in the meeting. Then set the review for the month the second-order effect should show up, and write down the name of whoever made the call. That isn’t about blame. It’s so the decision gets judged on its whole result, not on the part that showed up first.

This is also why compounding work is so hard to defend in a quarterly review. Content marketing that compounds is the clearest example I know: weak in the first order, strong in the second, which is the whole argument of that piece.

So here is the question I’d take into your next decision meeting. Which item on the current plan would you still approve if its result were reported in month six instead of week one?