Marginal ROI vs Average ROI

The number that always says yes

I want to talk you out of a number you probably love. Average return on investment, the ROI or the ROAS your dashboard prints in bold green, is the most flattering figure in marketing, and it will happily talk you into spending too much or too little on the very same channel. Here is the uncomfortable truth sitting underneath it. The profit a channel produces does not climb in a straight line as you feed it money. It follows an S shaped curve, slow at the start, then a steep middle where every pound works hard, then a long flattening where extra pounds do almost nothing. Average ROI smears all of that into one comfortable number. It tells you what your money did on average across the whole of that spend, and average is exactly the wrong lens when you are deciding what to do with your next pound. The question is never how the whole budget performed. The question is what the next pound will do. For that you need marginal ROI, not average ROI.

I am going to build this up with a worked table you can rebuild in a spreadsheet in ten minutes, because once you watch average ROI stay high while marginal ROI collapses beneath it, you cannot unsee it. You will start looking at every green ROAS in your account and asking the only question that pays: not what did this spend earn on average, but what would the next pound earn. Those two numbers part company faster than almost anyone expects, and the gap between them is where budgets get quietly wasted and where they get won.

Before we get into the numbers, let me tell you why I keep coming back to this one idea with nearly every client I sit down with. It is not because the maths is clever. The maths is barely more than primary school arithmetic once the table is on the screen. It is because the average is emotionally sticky. It is the number that got the channel approved, the number in the quarterly deck, the number a founder repeats at dinner. Nobody wants to hear that their proudest ROAS is also the reason they are leaving money on the table or, worse, quietly setting fire to it at the top end. So I am going to make the case slowly and with real rows you can check, because you will not drop a habit this comfortable on my say so. You will drop it when you see it in your own numbers.

A channel is not a flat rate

The hidden assumption behind average ROI is that a channel pays out at a constant rate. Put in twice the money, get back twice the profit. If that were true, average and marginal would be the same number and I would have nothing to write about. But no real channel behaves like that.

Think about what actually happens as you scale spend on a single channel, say paid search or a Meta campaign. At tiny budgets you are still learning, the algorithm has little to work with, and the return per pound is modest. As you grow into the sweet spot you reach the people who were basically ready to buy, the audience is warm, and every extra pound brings back a lot. Then you start running out of those people. To spend more you have to reach a colder audience, bid on looser keywords, show the same person the same ad a fifth time. The extra pounds still do something, but less and less, until they do almost nothing. That is the S curve, and it is not a quirk of one account. It is the shape of nearly every marketing response, and it is the backbone of the response modelling in Cutting Edge Marketing Analytics by Venkatesan, Farris and Wilcox, which is where I would point anyone who wants the proper treatment.

It helps to give the curve three named regions, because you manage each one differently. The first is the toe, the slow low part where you are still buying data and the machine is still learning who your buyer is. The second is the body, the steep stretch where each pound is met by warm demand and marginal return is at its best. The third is the shoulder, the long flattening where you have exhausted the easy demand and every extra pound has to work harder for less. Almost every argument about budget is really an argument about which region a channel is sitting in, and average ROI hides that from you completely because it blends all three into one figure.

The moment you accept that a channel has an S shaped response, average ROI becomes actively dangerous, because it averages the brilliant early pounds together with the useless late ones and hands you a single number that describes neither.

Average ROI and marginal ROI are different questions

Let me pin down the two numbers precisely, because the whole argument lives in the difference.

Average ROI asks: across everything I have spent on this channel so far, what did I get back per pound. If $S$ is your spend and $R(S)$ is the gross profit that spend brings back, then

$\text{ROI}_{\text{avg}} = \frac{R(S) - S}{S}$

Marginal ROI, sometimes written ROMI, asks a much sharper question: for the last block of spend I added, what did that block bring back per pound. If you increase spend by a small amount $\Delta S$ and profit responds by $R(S + \Delta S) - R(S)$, then

$\text{ROMI} = \frac{R(S + \Delta S) - R(S)}{\Delta S} - 1$

and in the smooth limit that is just the slope of the response curve minus one,

$\text{ROMI} = \frac{dR}{dS} - 1$

Average ROI is the height of the curve divided by how far along you are. Marginal ROI is the steepness of the curve right where you are standing. On an S shaped curve those two things can point in completely opposite directions, and that is the whole game.

If you want a single closed form to play with, the S shape is captured nicely by a saturating response, often called a Hill or logistic form,

$R(S) = \frac{R_{\max}\, S^{\alpha}}{S_{50}^{\alpha} + S^{\alpha}}$

where $R_{\max}$ is the most the channel could ever return, $S_{50}$ is the spend at which you reach half of that ceiling, and $\alpha$ controls how sharp the S is. You do not need to fit this to run your budget, but it is worth knowing the curve has a name and a ceiling, because the ceiling is exactly what average ROI pretends does not exist.

One more idea earns its keep here: elasticity, the percentage change in return for a one percent change in spend. Near the toe elasticity is high, near the shoulder it falls toward zero, and it is a cleaner early warning than the raw average because it is already telling you how tired the next pound will be.

A worked table you can copy

Here is a single channel, numbers kept round so the pattern jumps out. Spend is monthly. Gross profit returned is what that level of spend brings back before you subtract the spend itself. Each row adds another one thousand pounds.

Monthly spend	Gross profit returned	Extra from the last block	Net profit	Average ROI	Marginal ROI
1000	2000	2000	1000	100 percent	100 percent
2000	6000	4000	4000	200 percent	300 percent
3000	11000	5000	8000	267 percent	400 percent
4000	14500	3500	10500	262 percent	250 percent
5000	16500	2000	11500	230 percent	100 percent
6000	17500	1000	11500	192 percent	0 percent
7000	18000	500	11000	157 percent	minus 50 percent

Read down the average ROI column. It rises, peaks at a glorious 267 percent, then drifts down but stays high all the way to the bottom. Even at seven thousand a month, where you are clearly scraping the barrel, average ROI still reads 157 percent. Any dashboard would paint that green. Any agency would screenshot it. On average ROI alone this channel looks wonderful at every single spend level.

Now read down the marginal ROI column, the profit the last block actually brought back. It climbs to 400 percent in the steep middle, then falls off a cliff, hits break even at six thousand, and goes outright negative at seven thousand. The last block of spend at seven thousand did not earn 157 percent. It lost you money. Same channel, same month, two columns telling opposite stories.

That is the first thing worth sitting with. A channel with a brilliant average ROI can be a terrible place to put your next pound. The average is high because of pounds you spent ages ago in the steep part of the curve. Your next pound does not get that rate. It gets the marginal rate, and the marginal rate has quietly collapsed while the average was still smiling at you.

Where the two curves cross

There is a small piece of maths hiding in that table that is worth making explicit, because it turns a vague worry into a precise signal. The average ROI peaks at the exact spend where the marginal return has fallen back to meet the average return. That is not a coincidence, it is what has to happen at the top of any average.

In symbols, average ROI stops rising when its slope is zero,

$\frac{d}{dS}\!\left(\frac{R(S)}{S}\right) = 0 \quad\Longrightarrow\quad \frac{dR}{dS} = \frac{R(S)}{S}$

which reads in plain English as: your ROAS is at its highest at the very moment the next pound is only as good as your running average, and from there on the next pound is worse than the average that is still flattering you. So the peak of the ROAS is not the sign of a healthy channel with room to grow. It is the last moment the marginal and the average agree, and every step past it, the two part ways.

Keep that picture in your head. Everyone celebrates arriving at the ROAS peak. It is actually the moment your best pounds are already behind you, and the interesting profit lives just past the point where the pretty ratio starts to fall.

The flat maximum, and why precision is a trap

Look at the net profit column now, because it hides the second surprise. Net profit climbs, reaches eleven thousand five hundred at five thousand of spend, and then sits there. At six thousand of spend it is still eleven thousand five hundred. The profit is basically flat across a wide band of spend. This is what the analytics people call the flat maximum, and it is one of the most freeing ideas in the whole subject.

It means there is no knife edge. You do not have to find the one perfect budget to the nearest pound, because a whole range of budgets, anywhere on that plateau, produces almost exactly the same profit. Chasing the theoretical optimum to two decimal places is a waste of your evening. Getting onto the plateau and staying there is the entire job. The flat maximum is also why marginal ROI is such a forgiving tool in practice: you are not trying to hit zero exactly, you are just trying not to be a long way past it.

The flat maximum cuts the other way too, and this is the part people miss. If the top is flat, then being a little under the optimum costs you almost nothing, but being a long way over it starts to bite quickly, because past the plateau the marginal return is negative and every extra pound is now subtracting from the total. So when you are unsure, err toward the near side of the plateau, not the far side. An honest way to say it is that the penalty for timidity near the top is tiny and the penalty for greed is real, so the flat maximum quietly rewards discipline.

There is a comforting practical upshot. You do not need a perfect model of your curve to make good decisions. You need to know roughly where the plateau starts and to notice the moment you have wandered off the far edge of it. Both of those are visible in a monthly block table you can keep on one sheet.

Maximising ROAS makes you underspend

Here is the surprise that catches even experienced people, and it runs the opposite way to the one everyone expects. We spend so much time warning against overspending that we miss the mirror image.

Average ROI, our beloved ROAS, peaks at three thousand of spend, at 267 percent. So if your rule is to maximise ROAS, which is the rule baked into a thousand agency dashboards and more than one automated bidding setting, you will settle at three thousand and feel clever about it. But look at the net profit at three thousand: eight thousand pounds. And look at the net profit on the plateau: eleven thousand five hundred. Maximising your return on ad spend left three and a half thousand pounds of profit on the table every single month, because it told you to stop while the next pound was still returning 250 percent.

This is the trap in one line. Maximising average ROI does not maximise profit. It usually makes you underspend, because the ratio is best while the curve is still steep, long before you have harvested all the profit there is to harvest. Profit is a total, not a ratio, and you grow a total by spending every pound whose marginal return beats your threshold, even when doing so drags your lovely average down. Watching your ROAS dip as you scale is not always failure. Sometimes it is the sound of you finally collecting the profit you were leaving behind.

There is a version of this that shows up in the wild as a target ROAS. Somebody sets a rule that the account must hold, say, a three times return, and the platform dutifully throttles spend to keep the ratio there. It looks disciplined. It is often the single biggest cause of a profitable account being kept deliberately small, because the target sits up on the steep part of the curve and the algorithm never lets you walk down toward the plateau where the total profit actually lives. A target ROAS is a fine guardrail against nonsense at the bottom. It is a terrible ceiling to weld over the top.

A worked mini case: two channels, one budget

Numbers in isolation are only half the point. The reason marginal thinking pays is that it tells you where to move money, and you only ever move money between channels. So here is a small, deliberately tidy case with two of them.

Say you run a twelve thousand a month budget across paid search and an email newsletter programme. Today it is split six thousand to search and, because email felt like a nice to have, only one thousand to the newsletter, with the rest elsewhere. You check the last block on each.

Channel	Current spend	Marginal ROI on the last block	What the next pound is doing
Paid search	6000	0 percent	roughly breaking even, on the shoulder
Newsletter	1000	300 percent	still in the steep body of its curve

The averages look fine on both. Search is still printing a healthy blended ROAS because of all those brilliant early pounds, and the newsletter looks almost too good to be true. Average thinking says leave it alone, search is your workhorse. Marginal thinking says the exact opposite: your last pound into search is doing nothing, and your last pound into the newsletter is bringing back three. So move a block.

Take one thousand pounds off search and put it on the newsletter. On search you drop the block that was returning roughly zero, so your gross falls by about a thousand while your spend falls by a thousand, and net profit barely moves. On the newsletter that same thousand lands in the steep part of its curve and returns around three hundred percent, so gross rises by about three thousand on a thousand more spend, a net gain near two thousand pounds. Same total budget, same month, and you have found roughly two thousand pounds of profit that was sitting one column away the whole time.

At the point where you can no longer make a move like that, you have hit the condition the whole method is quietly aiming for. Across every channel, at the best possible allocation of a fixed budget, the marginal returns line up,

$\frac{dR_1}{dS_1} = \frac{dR_2}{dS_2} = \dots = \frac{dR_n}{dS_n} = 1 + h$

which just says a well spread budget is one where the next pound would do the same thing wherever you put it, and none of those pounds is below your threshold $h$. When one channel's marginal return is far above another's, you are not optimised, you are holding a free transfer you have not made yet.

The mistakes I see most often

I have watched a lot of accounts, and the same handful of errors turn up again and again. None of them are silly. Every one of them is average ROI thinking wearing a slightly different hat.

The first is judging a whole channel by its blended average. The channel is not one thing. It is a stack of pounds, the early ones wonderful and the late ones tired, and the blended figure describes none of them. Ask what the marginal block did, not what the pile did.

The second is cutting a channel the moment its average dips. Averages fall as you scale into the plateau, and that is supposed to happen. If the marginal return is still above your threshold, a falling average is a sign you are collecting profit, not losing your touch. Do not amputate a healthy channel because its blended number came off a vanity high.

The third is feeding a channel because it still looks green. A positive average can sit on top of a marginal block that is losing money, exactly as it does at seven thousand in our table. Green is not a licence to add. Only a marginal return above your threshold is.

The fourth is comparing channels on average ROI to decide where money goes. Two channels can show the same blended ROAS while one is crying out for more and the other is drowning. The averages are equal and the right move is still to shift money from one to the other, because their marginal returns are miles apart.

The fifth is worshipping precision on a flat top. People burn hours arguing whether the budget should be five thousand two hundred or five thousand five hundred while sitting squarely on the plateau where it makes no measurable difference, then set the whole thing and forget to revisit when the curve moves next quarter. The precision is fake and the neglect is real.

The sixth is moving too fast to read the response. If you triple a budget overnight you cannot tell which pound did what, and you will misread the curve badly. Marginal thinking needs deliberate steps and a little patience, which is the opposite of a heroic Monday morning reallocation.

The next pound rule

So throw away both bad habits, chasing the best average and feeding anything that still looks green, and replace them with one rule. Spend your next pound wherever its marginal return beats your threshold, and stop the moment it does not.

Formally, keep adding spend while the slope of the response curve stays above your hurdle,

$\frac{dR}{dS} - 1 > h$

where $h$ is the minimum marginal return you are willing to accept. Set $h$ to zero and you spend right up to break even, growing raw profit to its flat maximum. Set $h$ higher, say you need every marginal pound to bring back a healthy return because cash is tight or you have better uses for it, and you stop earlier, further back up the curve. Either way the decision is always about the next pound, never about the average.

In practice you run it as a loop, and you run it across channels, not just within one.

The loop never really ends, because the curves move. A channel that was on its plateau in November can have fresh room in January when a competitor drops out or a new audience appears. You are not setting budgets once a year. You are continually asking one small question, where does the next pound earn most right now, and letting the answer move your money.

How to actually run this without a modelling team

You do not need anything heavy. You need four habits.

First, measure in blocks, not in totals. Once a month, look at what you added since last time and what came back because of it. That difference, extra profit over extra spend, is your marginal ROI, and it is the only number that should drive the next budget change. The lifetime average of the channel is a vanity metric by comparison.

Second, change spend in deliberate steps and watch the response. You cannot see the slope of a curve you never move along. Nudge a channel up by a set amount, give it long enough to settle, and read the marginal return on that step. You are hand plotting your own S curve, one honest step at a time, and it is worth far more than any benchmark someone else hands you. A steady ten or twenty percent nudge is usually plenty to read the slope without blowing up your ability to attribute it.

Third, always compare across channels, never in isolation. A pound is not loyal to the channel that earned it last quarter. If paid search has flattened out and your newsletter still has a steep marginal return, the next pound belongs in the newsletter, whatever the averages say. This portfolio view of spend is exactly the kind of thing my performance marketing work is built to set up, and if you would rather not build the spreadsheet yourself you can just book a call and we will map your curves together.

Fourth, write your threshold down before you look at the numbers, not after. Your hurdle $h$ should come from your economics, what a customer is worth, what your cash costs, what else you could do with the money, and it should be decided in the cold light of day. Deciding the threshold after you have seen a tempting channel is how you talk yourself into feeding a loser, because there is always a story for why this one is special.

The one habit to leave with

If you take a single thing from this, make it a change of question. Stop asking what a channel returned on average, because the average is a monument to money you already spent and it will flatter you into both mistakes at once, stopping too early on your winners and feeding your losers too long. Start asking what the next pound will return, follow it wherever it is highest, and stop the moment it drops below your threshold. Average ROI tells you a comforting story about the past. Marginal ROI tells you the only thing you can still act on: what to do with the very next pound. Spend that one well, then ask again.

And that is the quiet power of it. You never have to solve the whole budget in one sitting or trust a model you cannot see inside. You just have to keep answering one honest little question, over and over, and let the answers walk your money to where it earns most. Do that every month and the plateau finds you, without a single heroic decision along the way.