How do you allocate a marketing budget across channels?
Marketing budget allocation is the process of dividing spend across channels by marginal return, what the next euro earns, rather than by the average return a channel produced last quarter. The question is not which channel performed best, it is where the next euro earns more than it costs. Rank channels by that, then move money from the saturated ones to the under-funded ones.
Almost nobody does it, because the reports most teams allocate from cannot answer the question. They show average return by channel, measured on last click. Both halves of that quietly bias the split.
Why does last-click reporting give you the wrong split?
Last-click attribution hands the whole conversion to the final interaction before purchase. It is simple, it is auditable, and it is wrong in a specific direction: it overpays the channels that harvest existing demand and underpays the ones that create it.
Roivenue client data shows over 50% of revenue is misattributed under last-click models, and the shape of the error is consistent. Branded search, retargeting, coupon affiliates and email look excellent, because they are usually the last thing a ready buyer touches. Prospecting social, display and video look weak, because they rarely are.
Reallocate on that report and you get a loop that is hard to see from inside. Money moves to demand capture, fewer journeys start, and two quarters later the capture channels have less demand to capture. Their volume falls and the drop gets filed under “the market softened”, when the budget decision caused it.
Platform dashboards fail differently. Each platform reports the conversions it believes it influenced, on its own window and its own view-through rules, so the totals across Meta, Google, TikTok and your affiliate network sum to more than the revenue finance booked. Allocating from numbers that do not reconcile to booked revenue is the most common structural mistake in channel planning. The mechanism is in how multi-touch attribution works, the model trade-offs in this guide to attribution models.
Where does the next euro go?
Every channel gets worse as you spend more into it. You buy the cheapest, most relevant inventory first, then the audience widens, frequency climbs and the auction gets more expensive as you bid into less qualified impressions. That is diminishing returns, and it is why average return and marginal return can point in opposite directions.
| Illustrative example | Channel A | Channel B |
|---|---|---|
| Monthly spend | €100,000 | €12,000 |
| Average return on spend | 4.0 | 3.0 |
| Return on the next €10,000 | 1.4 | 2.9 |
Channel A wins every dashboard in the building. Channel B is where the next euro belongs. A team allocating on average return moves money the wrong way and can produce a report that justifies it.
You do not need a saturation model to spot this. Two signals sit in data most teams already hold: does return fall as spend rises, and does extra spend buy new reach or just more frequency against the same people.
Impressions matter here and most setups do not capture them. A click-only view cannot see saturation forming, because saturation shows up in delivery before it shows up in clicks. Hence the case for tracking upper-funnel activity properly.
Is reallocating the same decision as spending more?
No, and collapsing the two is why budget arguments stall.
Reallocation holds the total constant and changes the split. It needs only a relative claim: channel B returns more per euro than channel A at current spend levels. You can be wrong about the absolute numbers and still right about the direction, and cash risk is close to zero.
Increasing total spend needs an absolute claim: this money will generate profit that would not have arrived anyway. Attribution cannot answer that, because it describes journeys that happened rather than what would have happened without the ad. That needs a holdout, a geo test or a media mix model.
So reallocate first, then use the credibility to argue for more. Every result below came from reallocation.
What does the allocation process actually look like?
- Rebuild the baseline on complete data. One view containing spend, impressions and clicks from every platform, plus revenue reconciled to what finance booked. This is a plumbing job, and it decides whether the other five steps are worth doing. Roivenue connects to 70+ platforms for exactly this step.
- Fix the denominator. Return on media spend is not return on marketing. Include agency fees, production, tooling and discounts funded from the marketing line. Channels differ enormously in non-media cost, and the ranking changes once you count it.
- Separate the jobs before comparing the numbers. Demand creation and demand capture do not compete for the same objective, so their returns are not comparable. Compare prospecting to prospecting, and judge capture channels on whether they capture everything available, not on how high their ROAS is.
- Set a floor and a ceiling per channel. The floor is the minimum spend at which a platform can exit its learning phase and return a readable result. Below it you are paying for noise. The ceiling is where marginal return crosses your threshold.
- Move money in tranches, and keep a control. Shift a slice rather than making one large reallocation, and hold at least one channel deliberately flat. Without something unchanged you cannot separate your decision from the season.
- Read the result on the conversion clock, not the calendar. Costs land on the day you spend, revenue lands whenever the customer converts, so read a change only after the conversion tail has run. Then let evidence trigger the next move: the signal to rebalance is a measured change in return, not an elapsed interval.
A performance manager needs that data to arrive without manual assembly, or the process quietly reverts to last year’s split plus 10%.
How do teams set the split, and which method holds up?
| Method | How it works | When it is defensible | Main weakness |
|---|---|---|---|
| Percentage of revenue | Fixed share of revenue, split roughly as last year | Stable business, stable mix, planning needs a number | Says nothing about channels. Encodes last year’s mistakes and makes budget follow revenue rather than cause it |
| Competitive parity | Match or index against competitor spend in the category | Brand building, where relative presence drives share | Their split reflects their economics, not yours |
| Objective and task | Bottom-up from targets: conversions needed, CPA per channel, spend implied | New markets or launches with no history | Uses today’s CPA as if it held at any volume, the diminishing returns assumption that fails |
| Marginal return | Rank channels by return on the next euro, move money until marginal returns level out | Established mix with usable data | Needs complete cross-channel data. Estimated from history, so it assumes conditions repeat |
The honest fifth option is “last year plus or minus 10%”, which is what most annual plans are. It survives committee, and it is wrong by however much the market moved.
What does reallocation look like when it works?
Four client results, all from changing the split rather than raising the total.
The Luxury Closet improved return on marketing by 17% in 10 weeks with no additional budget, purely by reallocating across existing channels.
UPC Slovakia ran the same logic in reverse and cut marketing spend by 34% while keeping performance strong. Finding spend that is not doing anything is the same analysis as finding spend that is.
Eppi reached 124% higher marketing return and 25% more conversions after moving to a model that valued the full journey rather than the closing click.
Orange shows the upper-funnel repricing effect most clearly. Once ad impressions entered the model rather than clicks alone, channels that had looked marginal turned out to be starting the journeys other channels closed, and revenue doubled.
Measurement changed first in all four. The budget moved second, and the total stayed flat or fell.
Where do budget allocation models fall short?
Anyone selling a clean answer here is overstating. The real constraints:
Attribution is not counterfactual. A marginal return estimated from past data tells you what happened at that spend level under those conditions, not what would have happened with no spend at all. Incrementality needs experiments or a mix model, and the two lenses will sometimes disagree. That disagreement is information, not a failure.
Untrackable demand pushes budget towards trackable channels. Word of mouth, offline conversation, brand memory and most out-of-home leave no touchpoint, so any model built from touchpoints under-allocates to whatever generates them.
Reallocation is not perfectly reversible, and the curves move. Cut a channel and you lose its platform learning, its audience momentum and sometimes its rate card, so restoring last month’s split does not restore last month’s performance. Meanwhile seasonality, competitor bidding, creative fatigue and algorithm changes keep shifting the relationship between spend and return: a saturation read from March is stale by November.
The optimal split is often not the achievable one. Contracted media, agency retainers, sponsorship commitments and internal politics constrain the range. Optimise inside the range you control and say so, rather than presenting an optimum nobody can execute.
What should you evaluate in whatever you use to make the call?
- Does it capture impressions, or only clicks? Click-only data cannot price upper-funnel media or see saturation forming.
- Does spend arrive automatically from every platform? If someone assembles a spreadsheet by hand, the process will not survive a busy quarter.
- Does revenue reconcile with what finance booked? A model that disagrees with booked revenue is a reporting project, not a decision tool.
- Can it show return as a function of spend, by channel? The slope is the decision. A ranking is only the headline.
- Can you explain the method to a CFO? A CMO defending a 15% shift needs a mechanism, not a black box.
- Does the vendor sell media? A model built by a company that sells inventory carries an incentive you do not share.
Frequently asked questions
It depends on category, margin and growth stage far more than on any benchmark. Percentage of revenue is a useful constraint on the total and a useless guide to the split, which is where most of the value sits.
Enough that no single channel can sink you, few enough that each one clears its minimum viable spend. The failure mode is twelve channels funded at a level where none can exit a learning phase and none can be measured.
Do not compare their returns directly: they answer different questions on different timescales.
Set brand spend as a deliberate share defended by mix modelling or long-run experiments, then optimise the performance share by marginal return inside its own envelope.
Not with a ROAS number, because it will lose. Show the journeys: how many converting customers had an upper-funnel touchpoint earlier in the path, and what happened to journey starts the last time that spend was cut.
Only for decisions inside a single platform. Across platforms the totals overlap and over-count, because each platform claims the conversions it touched.
Cross-channel allocation needs one independent model that reconciles to actual revenue.
You need something better than last click. Multi-touch attribution is the usual answer for channel-level decisions, and for a large advertiser it will not be the only lens: mix modelling and experiments answer the questions attribution cannot.
