Insights

Incrementality: What Would Have Sold Anyway, and How I Find Out

By Thomas Robinson · Sep 14, 2025 · 9 min read

Measurement gauge graphic — incrementality cover, DirectToGrowth Co.

Every budget argument I have sat in ends up at the same question, and none of the dashboards in the room can answer it. Not "which channel got the credit" but "what would have sold anyway." That gap is incrementality: the customers or net revenue a piece of spend produces that you would not have had without it. In my experience it is the number that comes closest to settling whether a channel deserves its budget.

The attribution tool has an opinion about it. Meta has a more generous one. Neither can tell you what happens to new-customer volume when you turn the channel off, because neither has ever watched you do it.

What incrementality is, in the terms I use

Incrementality is the extra customers, or extra net revenue, that a piece of spend produces which you would not have had without it. The question is counterfactual: what would have happened if we had not run this? A spike in sales after a launch is not an answer. Some of that spike is seasonality, some is customers who were coming back anyway, and some is the campaign. Incrementality is the last part only.

Most of the fights I see are about the easy case: spend that captured a sale which was going to happen regardless. An offer ad in front of a repeat customer. Branded search. A retargeting ad served the day before a reorder. All of it reports a beautiful ROAS, and a good share of it is soaking up customers rather than creating them.

Why incrementality matters more than the console

Because budget moves on it, and because the P&L does not care which attribution model you picked. Platform ROAS, the share of spend in learning, the number of ads launched: those are inputs. The scoreboard is total ad spend against total net-new customers, and net revenue growing while marketing spend as a share of net revenue stays inside the guardrail agreed with finance. Say the guardrail is thirty percent, an example rather than a benchmark: the job is to deploy the thirty percent in the way that produces the most new customers, not the best screenshot.

I have watched a channel's console get worse while the business got better. The case is below, and it is the clearest argument I have for keeping the blended number above every platform's. The blended number has blind spots too, a promotion or a stock-out moves it with no help from the ads, which is why the tests below exist.

The other reason it matters is structural. When the share of acquisition arriving through paid channels rises, blended CAC rises with it. Customers who arrive through branded search, direct, organic search and organic social cost far less, and they are what bring the blended number down. That share is exactly what brand spend is supposed to move, and it is exactly what is missing from a click-based view.

How I measure incrementality

Three methods, in rough order of how often I use them. None of them is a dashboard. Each gives one answer to one question, and I keep the attribution model fixed for the whole comparison, because switching models halfway through makes any ranking meaningless.

Holdouts and A/B tests

Split the audience, show the ad to one half and not the other, and compare new-customer orders between them. Email is usually the easiest place to run this: a promotional send to one segment, a control segment that gets nothing, and the difference in revenue is the incremental effect of the email. For a paid channel the same logic is pausing it for a window and watching what total new-customer orders do. Either way the output is one number, incremental orders, and from it an incremental cost per new customer I can put beside the blended one.

Geo uplift studies

When the spend is the kind a pixel sees little of, connected TV, podcasts, out of home, a reach buy on cheap CPMs, I run the channel in some regions and not in others, then compare new-customer volume and net revenue between the two sets of markets. It is the control-group method at the level of a geography, and it is how I back into whether upper-funnel spend is doing anything. It costs real spend and needs enough volume to read a signal, which is why it is a quarterly tool rather than a weekly one. It is also the answer I trust most when the attribution tool shows nothing and finance wants to know why the money is still going out.

Trend reads: brand search, direct and the acquisition mix

Between tests I read the series. Incremental brand search volume after a flight. Direct and organic sales relative to paid. The share of new customers that came through paid channels. If seventy percent of acquisition is paid, to pick a number, and that settles toward fifty while total new customers hold or grow, spend has started to redistribute because the brand is doing some of the work. A before-and-after read like this is weaker evidence than a geo test, because other things move at the same time, so I treat it as corroboration rather than proof.

What this looks like in real accounts

Three situations from accounts I have worked on. None is a controlled study. Each is what I observed and what I concluded from it.

Offer ads out of the top of the funnel

In one account, the top-of-funnel campaign was full of offer-led ads, which is a reliable way to make Meta look good and a poor way to build a brand. We moved the offer creative to the bottom of the funnel and let educational, story-led ads carry the top. For a while Meta's in-console ROAS looked worse. Blended new-customer CPA fell, new-customer ROAS improved and the volume of new customers grew. The console was measuring attributed revenue. The business was measuring incremental customers. They disagreed, and for the decision that mattered, the business number was the one to believe.

Paying for reach the pixel barely sees

The belief I have changed my mind on most is underinvestment in brand and the top of the funnel. The CPM gap is real: three or four dollars on some reach platforms, or a non-skippable YouTube in-stream placement, against the twenty-two to twenty-five I see on Meta in the accounts I run. Those impressions attribute badly and reach far more people. I do not spend and hope. The budget, the time horizon and the success criteria are written down first, the read comes from a geo uplift study and incremental brand search, and the business-level check is whether blended CAC and the marketing-spend ratio moved the right way. For a brand past fifty million, where the cheap incremental wins inside the existing channels are mostly taken, this is where I would be looking hardest.

The AppLovin test that showed no lift

When the AppLovin craze hit we took the credit on offer, moved our best Meta assets across in the right dimensions and ported the Meta strategy with them. We could not find a direct uplift or an incremental halo. I am not saying the channel cannot work; it did not work there, and a null result on one test does not prove zero effect. What it proved was opportunity cost. The same attention on a channel that already worked would have produced a predictable incremental win, and predictable incremental wins beat diverging into the channel everyone is talking about more often than the noise suggests.

How I put incrementality to work in an account

It starts with finance, not with a test. In the first month I look at the P&L, make sure marketing spend is tagged correctly, and agree the guardrails: marketing spend as a share of net revenue, the gap between acquisition cost and twelve-month contribution per customer, and how long the business can wait for payback, six months or twelve, which cash, inventory and repeat behaviour decide rather than preference. Whether the brand needs first-order profitability or can fund acquisition from later contribution changes every answer that follows. Those numbers decide how much incrementality a channel has to show to earn its budget.

Then the commitment is sized to the channel. Linear TV needs enough weight to be visible in total volume at all, which in my experience means something like five to ten percent of the budget or at least a fifty-thousand-dollar commitment, rough figures from the accounts I have run rather than a floor for everyone; below that I have not found a signal to read. Redistributing Meta spend toward educational creative can be done without changing the funnel at all. Shifting toward the top of the funnel is not always a new channel. Sometimes it is changing the content and the funnel inside the channels you already have.

Budget, horizon and success criteria go on paper before the first dollar: what we expect to see in the geo test, in brand search, in the acquisition mix and in blended new-customer CAC, and what would make me scale, extend or stop. I am not an advocate of pushing into channels and committing a lot of spend on principle. Sometimes the incremental move is small, and sometimes, depending on the brand's scale, channels and cash position, it is bigger. It is case by case, and I am suspicious of any rule that says otherwise.

Where incrementality is heading

Toward more of the spend being invisible to the pixel, not less. Consent gaps and shorter cookie windows take touchpoints out of the click path. For brands in retail or on Amazon, the sale drops out of the attribution view the moment a customer who saw the ad buys somewhere other than the site. The answer is not a better pixel. It is getting used to measuring exposure the attribution tool shows little of, through geo tests, brand search and the blended numbers, and being willing to invest in it anyway.

Where this leaves brand spend

In a lot of categories brand is the distinguishing factor right now, because the products are not very different, and the brand is what makes one of them fit a particular customer. If you have to keep buying every net-new customer, and there is no halo of people who saw an ad, remembered you and came back on their own, the business is renting its growth from the platforms. Building that halo through top-of-funnel storytelling is the part I used to underweight. I no longer do, but I measure it the same way I measure everything else.

Eye-level view of a marketer analyzing data on a laptop

None of this means abandoning direct response. It means knowing which part of the spend creates customers and which part collects them, and paying for each accordingly.

What I would tell a CEO reading this

Pick one business number, blended new-customer CAC or marketing spend as a share of net revenue, and make it the scoreboard every channel report answers to. Run an incrementality test on the biggest open question, one a quarter is the cadence I use, with the budget and the success criteria written down first. Treat the attribution tool as an opinion: useful for steering spend between ads and for a consistent trajectory, weak evidence on its own for whether a channel should exist, and weaker the more channels you sell through. And when a test shows nothing, as AppLovin did for us, take the predictable win in the channel that already works. The evidence that would change my mind is a channel that lifts new-customer volume in a geo test and lowers blended CAC at the same time. That is the whole bar.

FAQ

What is the difference between attribution and incrementality?

Attribution assigns credit for a sale that happened. Incrementality asks whether it would have happened without the spend. A channel can win the credit and add nothing, which is why I steer with attribution and decide channel budgets on holdouts, geo tests and blended numbers.

How do you measure incrementality on a small budget?

Start cheap: an email holdout, or pausing a channel for a window and watching total new-customer orders. Geo tests need enough volume to read a signal, so they come later. Either way, write down the budget, the horizon and what would make you scale or stop before the first dollar.

Is branded search incremental?

Some of it, usually less than its ROAS suggests, because those shoppers typed your name. The honest read is a holdout: cap brand bids in some regions and watch whether total branded orders, paid plus organic, fall. How hard competitors bid on your name is what decides it.

Seeing this in your own numbers?

A free 20-minute intro call. Bring what you have and I’ll tell you where I’d look first.

Book an intro call