Where Should Your Next Marketing Pound Go? A Four Step Method

The two ways owners split a budget, and why both lose money

Every business owner I sit with eventually asks me the same question, usually near the end of the meeting, usually a bit sheepishly: where should the next chunk of marketing money actually go? It is the right question. It is also the one nearly everyone answers badly, in one of two ways.

The first way is habit. We spent about this much on Google last year and about this much on the socials, business was fine, so let us do roughly the same again and maybe nudge it if we are feeling brave. The trouble is that last year's split was itself mostly inherited from the year before, which was inherited from a guess somebody made when the account was young. Habit is just a very old guess wearing a suit. Nothing about the way your customers behave today is obliged to match a decision you half made three years ago. Markets move, a competitor opens the account next to yours, a platform quietly changes its auction, your best advert wears out, and the split that was sensible two years ago rots into the split that is bleeding you today, all without a single meeting ever noticing.

The second way feels much cleverer, which is what makes it dangerous. You open the reports, you find the channel with the best average return, and you reason, quite naturally, that if that channel returns the most per pound then that is where the next pound belongs. More into the winner. It sounds like discipline. It is actually the single most common way I see good businesses cap their own growth, and to explain why I have to introduce the one idea this whole article turns on.

Why the best average return is the wrong thing to chase

Here is the thing that average return hides from you. No channel returns the same amount on every pound. The first pound you put into a channel does something wonderful, because it reaches the people who were basically already looking for you. The ten thousandth pound into that same channel does much less, because by then you are chasing people who are colder, more expensive, less interested. Every channel gets tired as you feed it more. Economists call this diminishing returns, and it is not a footnote, it is the entire plot.

So when a report tells you that a channel has a wonderful average return, it is telling you about the average over all the pounds you have put in so far, most of which were the low cost early ones. It is not telling you what the next pound will do. And the next pound is the only pound you actually get to decide about. The past is spent. The question is never what a channel has returned, it is what one more pound into it will return right now, on top of everything already there. That figure has a name. It is the marginal return, and it is almost never equal to the average return, because it lives out at the tired, expensive end of the channel where you are currently operating.

This is the first thing worth stopping on. The channel with the best average return can easily be the worst place for your next pound, because you may already have fed it to the point where it is full, while some smaller channel you are ignoring is still hungry and would turn that same pound into far more. Chasing average return marches you straight into the saturated channel. Every single time.

Average and marginal are two different numbers

It helps to see the two side by side, because owners routinely say return when they mean one and then act as if they meant the other. Average return is the total contribution a channel has produced divided by the total you have spent into it. Marginal return is the extra contribution from the very next pound alone. On a channel that is tiring, the marginal number always sits below the average number, and the gap widens the more you have already spent. Here is the same idea as illustrative figures.

Spend already in a channel	Average return per pound so far	What the next pound earns
Light	1.20	0.90
Moderate	0.95	0.55
Heavy	0.70	0.30

Read across any row and the marginal number is the smaller one. Read down and both fall, but the marginal number falls faster and leads the way down. The average is a memory of every pound you ever spent in that channel. The marginal is a forecast of the only pound you are about to spend. Fund your budget out of the memory and you will always overfeed whatever channel happened to get big first.

The four step method

So we throw out both habits and we replace them with a method. It is the resource allocation framework from the marketing analytics literature, laid out cleanly in Cutting Edge Marketing Analytics by Venkatesan, Farris and Wilcox, and it has exactly four steps. Define the objective. Map the inputs. Estimate the weights. Allocate to equalise the margins. Let me take them one at a time, because the beauty is in how ordinary each step is.

Step one, define the objective

You cannot allocate towards a goal you have not named. So name it, as a single number you want to make as large as possible. Not a vague wish like brand awareness, a number: gross profit this quarter, or new customers this year, or revenue, whichever one actually pays your wages. This number is what the analysts grandly call the objective function, and all it means is the thing every pound is ultimately trying to move. Write it at the top of the page. Everything below has to answer to it.

Most budget arguments in a business are really disagreements about the objective that nobody has said out loud. One person is optimising for revenue, another for profit, a third for their own channel's vanity chart, and they talk past each other for an hour. Pick the number first and half the argument evaporates.

A quick test for a good objective

A good objective passes three checks. It is a single number, so two channels can be compared on the same ruler. It is close to money, so growing it actually pays you, which is why profit usually beats revenue and revenue nearly always beats clicks. And it is one you can measure inside the window you are budgeting for, so you are not waiting eighteen months to learn whether you were right. If your candidate objective fails any of the three, it is a signpost, not a destination. Keep it on the wall by all means, but do not allocate to it.

Step two, map the inputs that move it

Now list the levers that actually push that number: your channels. Search, social, email, whatever you run. These are your inputs, the things you can pour money into. The job here is honesty about which ones genuinely move the objective and which are just busy. If a channel has never demonstrably shifted the number at the top of the page, it is a candidate for zero, and that is a perfectly good answer.

Write the objective as a response to these inputs. In plain words, your sales are some baseline plus a contribution from each channel that grows as you spend more, but grows more slowly the more you spend, because of the diminishing returns we just met. The tidy way to capture something that keeps rising but flattens out is the natural logarithm, so the model looks like this.

$\text{Sales} = \beta_0 + \beta_1 \ln(\text{Spend}_A) + \beta_2 \ln(\text{Spend}_B)$

Do not let the symbols put you off. $\beta_0$ is your baseline, what you would sell with no spend at all. $\text{Spend}_A$ and $\text{Spend}_B$ are what you put into channel A and channel B. The logarithm is just the mathematical shape of a thing that keeps growing but tires as it goes, which is exactly how every real channel behaves. And $\beta_1$ and $\beta_2$ are the weights, how strongly each channel moves your sales. Which brings us to the step everyone wants to skip.

The shape of a tired channel

It is worth watching the log shape do its work before we go on, because the whole method rests on it. Take a single channel with a weight of 3000 and watch its contribution and its marginal return as you pour more in.

Spend in the channel	Total contribution	Marginal return on the next pound
1000	20723	3.000
2000	22803	1.500
4000	24882	0.750
8000	26962	0.375

Look at the last column. Every time you double the spend, the marginal return halves. That is the signature of the log model, and it is a rule worth carrying in your head: under this shape, doubling a channel's budget roughly halves what its next pound earns. The contribution keeps rising, which is why the channel never looks like it is failing, but each new pound is quietly worth half of the one four thousand pounds ago. This is why no channel, however good, deserves an unlimited share.

Step three, estimate the weights from your own data

Here is where most owners expect me to say you need a data science team, and here is where I get to tell you that you mostly do not. The weights $\beta_1$ and $\beta_2$ are not things you guess or borrow from a benchmark deck. You estimate them from your own history, and the tool that does it is a regression, which is a fancy word for a line fitted through your past spend and past results. If you have a couple of years of monthly figures showing what you spent on each channel and what you sold, a regression will hand you the weights that best explain your actual numbers. A spreadsheet can run one. The point is that the weights come from your business, not from an article, not from what worked for someone else's shop.

This matters more than it sounds. Two businesses selling the same thing can have completely different weights, because their audiences, their creative, their reputations differ. Borrowing someone else's split, which is what every generic best practice guide quietly asks you to do, is borrowing someone else's weights. The whole method is built to replace opinion with your own evidence, and this is the step where that happens.

What a regression actually hands you

A regression does one honest thing: it finds the set of weights that would have best predicted your real sales from your real spend, month by month. You feed it two columns per channel, what you spent and what you sold, take the logarithm of the spend to match the shape we chose, and it returns a weight for each channel plus the baseline. You do not need to understand the arithmetic underneath any more than you need to understand a car engine to read a speedometer. What you do need is to sanity check the answer. A weight should be positive, it should be steadier than your monthly noise, and it should roughly survive dropping your best and worst months. If a channel's weight flips sign every time you add a month of data, the honest reading is that you cannot yet tell what that channel does, which is itself useful to know.

When your data is thin

Most SMEs do not have five clean years. That is fine. Start with whatever you have, fit a rough curve, and treat the weights as a first draft you improve every quarter as more months arrive. Rough weights from your own data still beat precise weights from someone else's, because at least they point in a direction that is true for you. If a channel is brand new and has no history at all, do not pretend the model knows it. Give it a small deliberate test budget, treat that spend as the price of learning its weight, and fold it into the model once it has a few months of its own. The method is not a one off calculation, it is a habit you run again every time the numbers refresh.

Step four, allocate to equalise the margins

Now the payoff, and the idea I promised you at the start. You have your objective and you have your weights. How do you split a fixed budget between the channels to make the objective as large as possible?

Not by giving the most to the channel with the biggest weight. That is the average return trap again. You allocate so that the marginal return, the extra sales from one more pound, is equal across every channel. Here is why that has to be right, and it is the sort of argument you can feel in your stomach once it lands. Suppose the last pound into search is earning you 40 pence of contribution, and the last pound into social is earning 75 pence. Then you are being silly, because you could take a pound out of search, losing only 40 pence, put it into social, gaining 75 pence, and be 35 pence better off having spent not a penny more. And you can keep doing that, shovelling pounds from the lower earning channel to the higher earning one, right up until the two marginal returns meet. The moment they are equal, no such move helps any more, and only then is your budget in the best possible place.

With the logarithmic model the marginal return of a channel is its weight divided by its spend,

$\text{Marginal return} = \frac{\beta}{\text{Spend}}$

so the optimal allocation is the one where

$\frac{\beta_1}{\text{Spend}_A} = \frac{\beta_2}{\text{Spend}_B}$

The last pound into A earns exactly what the last pound into B earns. That single condition is the whole answer to where your next marketing pound should go: wherever the marginal returns are not yet equal, that is where it goes, until they are.

The shortcut you can do on a napkin

The nice thing about the log model is that this condition has a clean solution, so you do not have to nudge pounds across by hand forever. Combine the equal margins rule with the fact that the channels have to add up to your total budget B, and the optimum falls straight out.

$\text{Spend}_A^{*} = B \cdot \frac{\beta_1}{\beta_1 + \beta_2}$

In plain words, each channel gets a slice of the budget exactly in proportion to its weight. Two channels weighted 3000 and 1500 split the money two to one, whatever the total is. And it generalises to as many channels as you run,

$\text{Spend}_i^{*} = B \cdot \frac{\beta_i}{\sum_j \beta_j}$

so the whole optimisation collapses to this: normalise your weights so they add up to the budget, and that is your split. There is even a tidy fact hiding in there. At the optimum, the shared marginal return that every channel lands on is just the total of the weights divided by the total budget,

$\text{Marginal return}^{*} = \frac{\sum_j \beta_j}{B}$

which means adding budget lowers the return every channel converges on, and cutting budget raises it. That is diminishing returns speaking about your whole account at once.

A worked example you can feel

Let me put numbers on it, illustrative ones, so you can watch total return rise as we move money. Say a regression on a shop's history hands back a weight of 3000 for search and 1500 for social, and the shop has 10000 pounds a month to split. The owner, running on habit, currently pours 8000 into search because search has always been the big one, and 2000 into social.

Let us look at the last pound in each. The marginal return of search is its weight over its spend, 3000 divided by 8000, which is 0.375. For social it is 1500 divided by 2000, which is 0.750. Stop and read those two numbers. The last pound into search is earning 0.375. The last pound into social is earning 0.750, twice as much. Search is stuffed full and social is starving, and the habit budget has it exactly backwards. Now watch what happens as we move money across, one row at a time.

Allocation	Search spend	Social spend	Marginal return search	Marginal return social	Total contribution
Habit split	8000	2000	0.375	0.750	38363
Nudge across	7500	2500	0.400	0.600	38504
Moved halfway	7000	3000	0.429	0.500	38571
Equalised, the optimum	6667	3333	0.450	0.450	38582
Overshot	6000	4000	0.500	0.375	38540

Read the two marginal columns together, top to bottom. As we take money out of the overfed search channel and feed the hungry social one, the two returns walk towards each other, 0.375 and 0.750 closing to 0.400 and 0.600, then to 0.429 and 0.500, and finally meeting at 0.450. And look at the final column: total contribution climbs the whole way up to that meeting point, 38363 to 38504 to 38571 to 38582, and then, crucially, the last row shows it falling back to 38540 when we overshoot and starve search in turn. The peak sits exactly on the row where the two marginal returns are equal. Not a penny of extra budget was spent anywhere in that table. The same 10000 pounds, split so the last pound in each channel earns the same, simply produces the most. That is the entire method working in one grid, and the overshoot row is the proof that equal margins is a genuine peak and not just a direction.

Here is the second thing worth sitting with. The optimal split, 6667 and 3333, still gives search more money than social, roughly two to one, exactly in line with its bigger weight, which is the napkin shortcut from earlier doing its job. So the answer was never to abandon your strong channel. The strong channel does deserve more. It just did not deserve as much as habit had shovelled into it, because even a strong channel gets full. The right question is never which channel is best, it is which channel is hungriest for the next pound, and those are not the same question.

The same method with three channels

Nothing changes when you add channels, the napkin shortcut just gains one more term. Suppose the shop also runs email, and the regression gives it a weight of 500 alongside search on 3000 and social on 1500. The weights now total 5000, so each channel takes its share of the 10000 pound budget in proportion.

Channel	Weight	Optimal spend	Marginal return
Search	3000	6000	0.500
Social	1500	3000	0.500
Email	500	1000	0.500
Total	5000	10000	0.500

Every channel lands on the same marginal return of 0.500, which is the total of the weights, 5000, over the total budget, 10000, exactly as the formula promised. Email, the smallest channel, is not starved and it is not shut down, it simply receives the slice its weight earns. That is the discipline: no channel is a hero and none is a villain, each gets funded to the point where its next pound is worth the same as everyone else's next pound.

A worked mini case, from habit to optimum

Let me run the whole thing once more end to end on a small illustrative business, because seeing it start to finish is worth more than any single formula. Picture a little online homeware shop with 6000 pounds a month for marketing. Out of pure habit the owner puts 5000 into paid search and 1000 into paid social, because search was the first thing that ever worked and it has held the lion's share ever since.

We pull twenty four months of spend and sales into a spreadsheet, take logs, and run the regression. It comes back with a weight of 2000 for search and 1600 for social. Notice that search still has the bigger weight, so habit was not mad, it was just lazy about how much. Now we check the margins on the current split. Search is earning 2000 divided by 5000, which is 0.40 on its last pound. Social is earning 1600 divided by 1000, which is 1.60. The last pound into social is doing four times the work of the last pound into search, and the owner has been quietly ignoring it for years.

Channel	Weight	Habit spend	Habit marginal return	Optimal spend	Optimal marginal return
Paid search	2000	5000	0.40	3333	0.60
Paid social	1600	1000	1.60	2667	0.60

The napkin shortcut gives the optimum straight away: the weights total 3600, so search takes 6000 times 2000 over 3600, which is 3333, and social takes the rest, 2667. Both channels now sit on a marginal return of 0.60. Run the contribution before and after and it rises from about 28087 to about 28845, a lift of roughly 758 pounds of contribution every month on exactly the same 6000 pound budget. That is around 9000 pounds a year the shop was leaving on the table, not because it spent too little, but because it spent it in the wrong place. No new budget, no new channel, no clever creative. Just the same money moved to where the next pound earns more.

Common mistakes that quietly undo all of this

The method is simple, but there are a handful of ways I see it go wrong in practice, and every one of them is worth naming so you can catch yourself.

Comparing average with average

The commonest slip is the one this whole article is about, and it is sneaky because it hides inside sensible language. Someone says search returns more than social, points at two average figures, and moves money towards the bigger average. But the two averages are memories, and you are spending in the future. Always drag the conversation back to the next pound. If your dashboard only shows you average return, add a column that divides each channel's weight by its current spend, and argue from that column instead.

Optimising a number that does not pay you

If you set revenue as the objective when your margins differ wildly across products, you can grow revenue and shrink profit at the same time. The model will do exactly what you asked and make you poorer. Choose the objective that is closest to the money you actually keep, and if two channels sell different product mixes, weight the objective by margin so the method is chasing profit, not turnover.

Freezing the weights

Weights drift. A platform changes, a competitor arrives, your creative ages, and last year's beta is no longer this year's beta. If you estimate the weights once and then treat them as gospel forever, you have simply invented a new habit with better manners. Estimate them again every quarter, or every time the numbers refresh, and let the split move when the evidence moves.

Ignoring the floor and the ceiling every channel has

Real channels come with constraints the plain model does not know about. Some have a minimum spend below which they barely function, some have a practical ceiling where you have simply run out of audience, and some carry a brand or coverage reason to keep a presence even when the maths would starve them. Treat the equal margins answer as the target you move towards, then apply your real world floors and caps on top. The method tells you the direction of travel, not that you should drive off a cliff to get there.

Moving everything overnight

Even when the maths is right, slamming a budget from one split to another in a single month is a good way to learn nothing and spook your results. Move in steps, watch what the marginal returns actually do as you go, and let the real world correct your weights. If shifting a thousand pounds towards social does not lift the objective the way the model predicted, that is not a failure, that is the model telling you your social weight was too high, and you update it.

How to apply this to your own numbers this week

You do not need to do all of this perfectly to beat the two lazy methods, you just need to do it at all. Here is the smallest honest version you can run in an afternoon.

First, write one number at the top of a page, the thing you are actually trying to grow, and make it the one closest to money that you can measure this quarter. Second, list every channel you spend on, and be ruthless about which have ever demonstrably moved that number. Third, pull two years of monthly spend and results into a spreadsheet, one row per month, one column of spend per channel and one column for the objective. Fourth, take the logarithm of each spend column and fit a regression to get a weight per channel, and sanity check that each weight is positive and reasonably steady. Fifth, for every channel divide its weight by its current spend to get its marginal return, and lay those numbers next to each other. Sixth, move money, in steps, out of the channels with the lowest marginal return into the ones with the highest, and stop when they have levelled off. Seventh, diarise the whole thing to run again next quarter, because the weights will have moved.

That is the entire discipline on one hand. Objective, channels, data, weights, margins, move, repeat. Everything past that is refinement, and refinement is worth much less than simply doing the seven steps at all.

What this changes on Monday morning

The reason this beats habit is obvious, habit is a stale guess. The reason it beats chasing the best average return is subtler and worth carrying with you: average return tells you where a channel has been, marginal return tells you where your next pound should go, and a budget can only ever be built out of next pounds. Once you have seen the worked table climb to its peak and fall away again on the overshoot row, you cannot unsee it, and you will never again point at a channel's average and call it a reason to spend.

If you want a hand turning your own spend history into weights and a proper allocation, that is exactly the kind of work I do, and it sits at the centre of my performance marketing work. Bring two years of numbers and we will find the pound that is in the wrong channel, and usually there is more than one.

The one line to leave with

Stop asking which channel is best. Start asking where the last pound earns the same. Your marketing budget is not optimised when your strongest channel has the most money, it is optimised when moving a single pound from any channel to any other channel would make you no better off, and the only way to reach that point is to feed the hungry channels and stop overfeeding the full ones. Define the objective, map the inputs, estimate the weights, equalise the margins. Four steps, one idea, and a budget that finally earns its keep.