Borrowed Against Tomorrow

A high ROAS doesn't mean you created a new customer. It might just mean an existing one arrived today instead of later.

If any company could track advertising cleanly — from impression to click to purchase, all on one platform — it should know exactly what its ads are worth. Amazon sits in exactly that position. But when Amazon looked closely at its own data, it found that precise measurement doesn't guarantee an accurate read. The effect captured by immediate clicks and purchases and the effect captured by real, long-term shifts in buying behavior didn't line up. They diverged sharply.

That gap has a name: demand harvesting. It describes what happens when an ad pulls forward a purchase a customer was already going to make — tomorrow, next week — and makes it happen today instead. A sale did occur. The number isn't lying. What the number doesn't say is whether that sale was new demand or simply future demand, relocated to today's column. The dashboard doesn't know the difference.

Everything this series has covered so far — harvesting versus planting, discounts and targeting, price premiums — collapses into this one illusion. Whatever the tactic, if you can't tell whether a good-looking number represents value you created or value you borrowed from the future, you fall into the same trap again. Amazon and Trader Joe's are two extremes that make this distinction visible, each in its own way.

What ROAS Didn't Tell Amazon

The pattern Amazon confirmed was this: many ads with high immediate ROAS were simply moving up the timing of a purchase a customer had already decided to make. Ads with lower immediate conversion but broader reach — built to raise awareness, not force a click — ended up bringing in more new customers once enough time had passed. Short-term efficiency and long-term marketing value pointed in opposite directions. Measured at one moment, the same campaign could look brilliant; measured at another, it could look wasteful.

A high ROAS doesn't always mean new demand — it may just mean tomorrow's sale, pulled into today.

After this discovery, Amazon moved away from evaluating ads on pure short-term efficiency and toward measuring long-term contribution alongside it. The company best positioned in the world to track advertising with precision confirmed, using its own data, that precision alone doesn't produce the full picture. It didn't stop measuring — it widened what it measured.

One more mechanism amplifies this illusion. Customers pass through several touchpoints before buying. They see a brand ad, encounter content on social media, run a search, and finally click a retargeting ad that closes the sale. But attribution models that credit the last click with the entire sale treat every earlier touchpoint as worth zero. Build a budget on that math and retargeting looks like the most efficient channel, brand advertising the least. The hand that planted the seed disappears; only the hand that harvested it is visible. A flaw in the measurement tool itself quietly, systematically starves brand budgets.

P&G's Narrow-Targeting Wall

Another company arrived at a similar conclusion by a different route. Around 2016, Procter & Gamble sharply increased its digital performance advertising budget, narrowing its targeting to reach only the most obviously relevant consumers. Efficiency metrics improved. Sales did not.

Marc Pritchard, then P&G's Chief Marketing Officer, said as much publicly: targeting one- and two-second ads too narrowly had been the mistake. P&G subsequently shifted back toward broader reach. This wasn't a small experiment — it was one of the largest consumer goods companies in the world, testing the theory with billions of dollars. Which is why the case reads less like one company's misjudgment and more like an industry-wide data point about the structural limits of narrow targeting. What Amazon found inside its own numbers, P&G confirmed in its own revenue.

The scale of both companies is what makes this pairing worth taking seriously. Amazon has arguably the best measurement infrastructure in the industry; P&G has spent more on marketing, for longer, than almost anyone. Neither precise data nor deep budgets protected either company from the illusion that narrow, well-targeted efficiency creates. If anything, precision made the illusion more convincing — the finer the number, the harder it is to doubt it.

When Good Numbers Are Borrowed

Demand harvesting turns genuinely dangerous at one specific moment: when a company with a strong existing brand shifts its center of gravity toward performance marketing.

The early phase of that shift is deceptively smooth. Consumers who already hold a favorable view of the brand respond well to ads. ROAS looks good; customer acquisition cost looks low. Every number seems to confirm that performance marketing works, and on the strength of that evidence, more budget moves from brand to performance. But that efficiency is really just brand equity built up earlier, now being spent down. What performance marketing harvests is demand that brand advertising planted long before. Planting less doesn't show up right away — not while there's still equity in the ground to harvest.

The trouble is that this structure reinforces itself. Brand equity erodes gradually while performance metrics stay strong for a while longer. That strength reads as confirmation that cutting brand and growing performance was the right call. Confirmation justifies doing it again. Only after a year or two, once equity has been meaningfully depleted, does efficiency start slipping too — and by then, the cause is usually filed under 'the competition got tougher' or 'the market got harder,' not 'we stopped planting.' When the cause is invisible, the company keeps walking in the same direction.

What makes this especially hard to catch is that the performance team's ROAS data is real. It's not wrong. What the same data can't tell you is whether that ROAS comes from the strength of the performance campaign itself, or from brand equity accumulated years before. Because that distinction is invisible inside the numbers, the decision to keep funding performance and keep cutting brand repeats itself naturally — with no one ever feeling like they made the wrong call.

Trader Joe's and Sears

One company avoided this self-reinforcing pattern entirely. Another fell into it precisely. Both operated in the same industry — retail — which is what makes the comparison fair: not two unrelated sectors forced together, but two different choices, under similar conditions, that led to very different endings.

Trader Joe's is known for running almost no performance advertising. Instead, it invests steadily in product quality, a distinctive in-store experience, and employees who are genuinely friendly. Customers experience the store and talk about it, and the brand spreads through that conversation. There's no ad to run ROAS on, so there's no efficiency metric to point to. And yet the company's customer loyalty ranks among the highest of any major retailer. The absence of a metric wasn't the absence of value.

On the other side stands Sears. Once the owner of a genuinely powerful brand, Sears spent decades cutting brand marketing and leaning harder on short-term promotions and discounts. Less brand investment led to depleted brand equity; depleted equity weakened pricing power; weakened pricing power drove deeper reliance on discounting. That cycle contributed to the company's eventual bankruptcy. It wasn't the only cause — but the pattern of chasing short-term efficiency while consuming long-term equity is unmistakably present in its story.

What separated the two companies wasn't skill. It was where they chose to look. Trader Joe's kept investing in what was hard to measure. Sears kept migrating toward what was easy to measure. Given how organizations actually make decisions, that migration makes sense: when 'this campaign's ROAS is 5.2' competes against 'this investment probably built brand trust,' the sentence with a number attached wins by default.

Being hard to measure doesn't make something worthless. The value that lasts longest is usually sitting exactly where no one is measuring.

There's an old joke about looking for keys under the streetlight — not because that's where you dropped them, but because that's where the light is. Marketing measurement has the same problem. Cost-per-click and conversion rate sit under the streetlight; they're easy to see. Whether a brand makes it into a customer's consideration set, the trust that makes someone less sensitive to price, the impression that only pays off six months later — these live in the dark. Invisible isn't the same as absent.

It's worth naming what's actually sitting in that dark. Whether the brand made it onto a customer's shortlist. The emotional associations a customer carries about it. How much that trust reduces price sensitivity. What today's perception does to customer lifetime value years from now. None of this is impossible to measure — brand tracking studies, lifetime value analysis, and marketing mix modeling all exist for exactly this. But none of them are as immediate or as clean as cost-per-click. The numbers vary study to study, need interpretation, and take time to resolve. That ambiguity is precisely what gets these values quietly pushed out of the room. A small, certain number beats a large, uncertain one.

This bias isn't a personal failure — it's structural. The performance team can prove, in one clean sentence, 'this budget generated this much revenue.' The brand team can't prove 'this budget built tomorrow's sales foundation' with the same confidence, using the same sentence structure. That's not a gap in talent between two team leads; it's a game where proof itself is unevenly distributed from the start. Miss that unfairness, and an organization keeps funneling budget toward whatever is easiest to prove — while believing, the entire time, that it made a fair call.

Trader Joe's conclusion is, in a way, a paradox. Because it had no metric for advertising efficiency, it had nothing to distort its behavior around. No metric meant nothing to optimize — which meant nothing pulled its attention away from the actual product and experience. The absence of a metric wasn't a weakness. It was a shield against optimizing the wrong thing.

Managing Both Clocks at Once

Escaping the efficiency trap doesn't mean abandoning efficiency. It means tracking the time horizon efficiency can't see. You need one eye on this campaign's ROAS and another on what several years of investment have done to brand equity. Many organizations check brand health metrics quarterly, or even annually — too slow. By the time a decline shows up a year later, real damage has already been done. Awareness, purchase consideration, and full-price purchase rate should be tracked at least monthly, and reported in the same document, the same meeting, as performance metrics. As long as the two numbers live in separate rooms, no one sees the whole picture.

This shift also requires time. Brand marketing's effects typically show up somewhere between six months and two years out. Pull the budget because nothing visible happened yet, and you stop planting before the harvest was ever due. Patience here isn't sentiment — it's an accurate read of the timeline.

Two questions are worth asking honestly. When was the last time you ran an ad that was purely about the brand, with no performance goal attached? And when budgets get cut, which one goes first? In almost every organization, the answer is the same: whatever can't prove its return immediately gets cut first. That order is itself a signal — a tilt toward demand harvesting that was already there before anyone noticed it.

Which narrows the last question to one. Is the good number in front of you new demand, or an invoice for tomorrow's sale, paid early? If you can't answer that, you also don't yet know what today's efficiency is quietly eating into. The crack Amazon found in its own data, the wall P&G hit with its own revenue, the fork where Trader Joe's and Sears went separate ways — all of it is the same story told three times. Selling well today is not evidence that you'll sell well tomorrow.

A good number today doesn't prove tomorrow — it might just be tomorrow, spent early.