How to Decide How Much Budget to Allocate to Each Channel: The Marginal Return Method

Key Takeaways

• Average return tells you what your spend has already done. Marginal return tells you what the next pound will do. Only the second is an allocation input, and almost every reporting dashboard shows only the first.
• Channels saturate. Doubling spend rarely doubles results, and the point at which returns bend is the single most valuable number in allocation — it is also the number that no platform will report to you.
• The optimal allocation is reached when the next unit of spend returns approximately the same in every channel. If one channel returns more at the margin than another, money is in the wrong place by definition.
• Cash constrains allocation independently of return. A channel with excellent returns and a fourteen-month payback can be unaffordable for a business whose cash cycle cannot carry it.
• Reserve budget for demand creation and for testing before optimising the remainder. Both lose every argument against short-term efficiency if they compete for the same pool.
• Last-click attribution systematically over-credits capture channels and under-credits creation. Allocating between channels on last-click data reliably produces a portfolio that looks efficient while growth decelerates.
• Reallocate on a stated cadence with stated triggers. Continuous reallocation chases noise; annual reallocation ignores evidence.

1. The Short Answer: How to Split Budget Across Channels

Allocate on marginal return. Size the total budget from your contribution margin and payback tolerance, set an economic gate that every channel must clear, estimate what the next unit of spend would return in each channel, then move budget from channels with lower marginal return to channels with higher marginal return until the next pound returns approximately the same everywhere.

This is the equimarginal principle, and it is not a marketing idea — it is the standard solution to allocating a limited resource across competing uses. What makes it hard in marketing is not the logic but the measurement: marginal return is not reported by any advertising platform, and estimating it requires deliberate work.

The alternative most teams use is a percentage rule inherited from somewhere — sixty per cent search, thirty per cent social, ten per cent everything else. Percentage rules have one virtue, which is that they are decisions. They have one fatal flaw, which is that they contain no information about what any channel is currently returning. A rule that would allocate identically whether a channel returned four times or one point one times its spend is not an allocation method.

• AEO Quick Answer: Size the envelope, set an economic gate, estimate marginal return per channel, move budget until marginal returns equalise.
• The principle: money is in the wrong place whenever one channel returns more at the margin than another.
• Percentage rules are decisions without information. Marginal return is the information.

2. Average Return Versus Marginal Return

This distinction is the whole guide, so it is worth being precise about it.

Average return is total revenue attributed to a channel divided by total spend in that channel. It is what every dashboard reports as [ROAS](/glossary/roas) or cost per acquisition. It describes the performance of all the spend you have already committed, blended together.

Marginal return is what the next unit of spend would produce. It is almost always lower than the average, because within any channel you buy the cheapest, most responsive demand first: your existing audience, your brand terms, your highest-intent segments. Each additional pound reaches progressively less responsive people.

The gap between the two is where allocation errors live. Consider a channel reporting an average return of 4.0. That figure is consistent with a channel where the next pound returns 3.5 — worth expanding — and equally consistent with a channel where the first eighty per cent of spend returned 4.8 and the last twenty per cent returned 0.9, meaning the next pound is destroying value. The average is identical in both cases. The correct action is opposite.

This is why scaling decisions made on average ROAS so reliably disappoint. A team doubles spend on the channel with the best average return, and the average degrades — not because the channel stopped working, but because the incremental audience was never as responsive as the base. The channel did not break; the metric was never answering the question being asked of it.

Practically, you can approximate marginal return without sophisticated modelling. Change spend in one channel by a meaningful amount — twenty to thirty per cent, held for at least one full purchase cycle — and observe the change in outcome. Marginal return is the change in result divided by the change in spend. This is crude, it is disrupted by seasonality and by concurrent changes elsewhere, and it is still far more informative than the average.

• Average return: total result divided by total spend. Describes committed spend.
• Marginal return: what the next unit of spend produces. The only allocation-relevant figure.
• Marginal is almost always below average, because cheap responsive demand is bought first.
• Two channels with identical average return can require opposite decisions.
• Crude estimate: change spend 20-30%, hold one cycle, divide change in result by change in spend.

3. Saturation: Why Channels Bend

Every acquisition channel has a response curve, and every response curve eventually bends. Understanding the shape is what lets you predict where additional budget stops being productive.

The typical shape has three regions. At low spend there is often a threshold region where returns are poor because the channel cannot produce a readable signal or reach a viable frequency — this is the minimum viable budget problem, and it is why partially funding many channels performs badly. Above that comes the efficient region, where returns are strong and roughly proportional. Then comes saturation, where each additional pound reaches progressively less responsive audiences and returns decline, sometimes sharply.

Several distinct mechanisms drive saturation, and they respond to different remedies. Audience exhaustion: you have reached the addressable audience and are now buying repeat exposure. The remedy is audience expansion or a new creative angle, not more budget. Auction dynamics: bidding more aggressively raises your own costs, and in competitive auctions raises everyone's. Creative fatigue: the same creative shown at rising frequency produces falling response — the signature is rising frequency with falling click-through, and the remedy is production capacity, not spend. Intent exhaustion in capture channels: there are only so many people searching a term this month, and no budget increase creates more of them.

The last mechanism is the most important and the least discussed. Capture channels are bounded by existing demand. This is precisely why demand creation belongs in the budget: it raises the ceiling that capture channels operate under. A business that has saturated its capture channels and has no creation investment has no way to grow paid acquisition except by paying more for the same demand.

You can locate your own saturation points empirically without formal modelling. Plot spend against result by week or month for a channel over a year, ideally with meaningful variation in spend. The bend is usually visible. If your spend has been flat all year, you have no variation to learn from — which is itself an argument for deliberately varying spend, since a constant budget teaches you nothing about your own response curve.

• Three regions: below-threshold, efficient, saturated.
• Audience exhaustion — remedy is expansion or new angle, not budget.
• Auction dynamics — bidding harder raises your own cost.
• Creative fatigue — signature is rising frequency with falling click-through; remedy is production.
• Intent exhaustion — capture channels are bounded by demand that already exists.
• Flat spend all year means no data about your own curve. Vary deliberately.

4. Step One: Size the Total Budget Envelope

Before splitting anything, decide how much there is to split. Marketing budget is not a percentage of revenue by convention; it is determined by unit economics and by cash.

Start with contribution margin per customer: revenue minus all variable costs of delivering to that customer. This is what is available to pay for acquisition and to contribute to fixed costs and profit. Businesses that plan against gross revenue rather than contribution consistently over-invest.

Then decide your target ratio of lifetime value to acquisition cost. The commonly cited 3:1 benchmark is a rule of thumb rather than a law, and it should be adjusted for your gross margin, your retention profile and the confidence interval on your [LTV](/glossary/ltv) estimate. Businesses with long, uncertain lifetimes should demand a wider margin of safety, because an LTV estimate extrapolated from twelve months of data on a five-year expected lifetime carries substantial error.

Then apply the cash constraint, which is where most theoretical allocations break. Payback period — how long until a customer has repaid their acquisition cost — determines how much acquisition your cash can carry, independently of how good the returns are. A channel with excellent lifetime returns and a fourteen-month payback consumes cash for fourteen months per customer. A business with a short cash runway cannot fund that regardless of its eventual return, and no amount of enthusiasm about lifetime value changes the bank balance. You can size this precisely with the [CAC payback calculator](/tools/cac-payback-calculator).

The envelope is therefore the smaller of two numbers: what your economics justify, and what your cash can carry. Growth-stage businesses frequently discover that cash, not return, is their binding constraint — and the correct response is often to shift the mix toward faster-payback channels even at lower lifetime return, which is an allocation decision that pure return-maximisation would never suggest.

• Build from contribution margin, not gross revenue.
• Set an LTV-to-CAC target adjusted for margin, retention confidence and estimate error.
• Apply the cash constraint: payback period determines how much acquisition cash can carry.
• The envelope is the smaller of what economics justify and what cash can fund.
• Cash-constrained businesses should weight faster payback over higher lifetime return.

5. Step Two: Set the Economic Gate

Before comparing channels to each other, set the standard every channel must clear to remain funded at all. This prevents the common failure where budget is optimally distributed across a set of channels none of which is actually viable.

The gate has three components. A maximum acceptable acquisition cost, derived from contribution margin and your target ratio. A maximum acceptable payback period, derived from your cash position. And a minimum viable spend, below which the channel cannot produce a readable signal — funding a channel below this level is not a small bet, it is a wasted one.

Apply the gate before optimisation. Channels that cannot clear it at any spend level are removed from the portfolio, not optimised within it. Channels that clear it only at low spend are capped rather than scaled.

One important refinement: the gate should differ by job. A capture channel should clear the full economic gate on its own performance, because its contribution is direct and measurable. A creation channel cannot be held to the same standard on the same measure, because its contribution is lagged and largely indirect — it appears in other channels' performance. Holding creation channels to a capture gate is the mechanism by which creation budget is systematically eliminated, and the reason is arithmetic rather than judgement: the measurement system credits creation's results to capture.

So set two gates. Capture channels: cost per acquisition, payback period, and marginal return above threshold. Creation channels: reach against target audience, cost per thousand qualified impressions, and a periodic incrementality test to establish contribution. Assess each against its own gate, and hold the overall portfolio to the blended economic requirement.

• Three gate components: maximum acquisition cost, maximum payback period, minimum viable spend.
• Apply the gate before optimising — remove non-viable channels rather than tuning them.
• Set separate gates for capture and creation; holding creation to a capture gate eliminates it by construction.
• Hold the portfolio, not each individual channel, to the blended economic requirement.

6. Step Three: Estimate Marginal Return Per Channel

This is the analytical core, and there are three practical methods with increasing rigour and cost. Use the strongest one your data volume supports.

Method one: spend variation testing. Change spend in one channel by twenty to thirty per cent, hold everything else as constant as you can, and observe the change in outcome over at least one full purchase cycle. Marginal return is change in result divided by change in spend. Cheap, available to everyone, and confounded by seasonality and concurrent changes — so run it repeatedly rather than once, and prefer decreases as well as increases, since a decrease that costs you nothing is the clearest evidence of saturation you will ever get.

Method two: geographic holdout testing. Split comparable regions into test and control, change spend in test only, and compare. This is a genuine controlled experiment and it measures incrementality directly rather than inferring it. It requires enough geographic volume for regions to be comparable, and it requires the organisational willingness to turn spend off somewhere, which is usually the real obstacle. It is the strongest evidence available to most mid-sized businesses.

Method three: marketing mix modelling. Regress outcomes against spend across channels plus external factors, estimating contribution and saturation curves. Requires substantial history and genuine spend variation — a business that has held budgets constant for two years has nothing for the model to learn from. Our technical treatment is in [marketing mix modelling with Python](/resource/blogs/marketing-mix-modeling-mmm-python).

What all three have in common is that they measure incrementality — the outcome that would not have happened otherwise. Platform-reported conversions do not measure this. A retargeting campaign showing an exceptional reported return is frequently taking credit for purchases that would have occurred anyway, and the only way to know is to withhold it and observe what changes.

A note on branded search, because it is the most common example. Bidding on your own brand terms typically reports outstanding returns, because the people clicking were already looking for you. Whether that spend is incremental depends on the competitive situation: if competitors bid on your brand and you would lose the click, it is defensive and often worth it; if you would have received the organic click anyway, much of it is not incremental. This is straightforward to test with a holdout, and the result is frequently uncomfortable.

• Method 1: spend variation — cheap, universally available, confounded; run repeatedly, and test decreases too.
• Method 2: geographic holdout — genuine experiment, strongest practical evidence, needs willingness to turn spend off.
• Method 3: marketing mix modelling — needs substantial history and real spend variation.
• All three measure incrementality; platform-reported conversions do not.
• Branded search is the standard case where reported return and incremental return diverge sharply.

7. Step Four: The Allocation Rule

With marginal returns estimated and the gate applied, allocation becomes mechanical. Move budget from the channel with the lowest marginal return to the channel with the highest, in increments, until the marginal returns are approximately equal.

That equilibrium is the optimum. The reasoning is direct: if channel A returns 3.0 at the margin and channel B returns 1.5, moving a pound from B to A gains 3.0 and loses 1.5, a net gain of 1.5. Keep moving until the two converge. When the next pound returns the same everywhere, no further movement improves the total, and you are done.

Three practical constraints modify the pure rule, and ignoring them is how theoretically optimal allocations fail in practice.

Minimum viable spend. You cannot reduce a channel below the level at which it produces a readable signal without effectively removing it. Either fund it above the threshold or remove it — a channel funded at half its viable minimum is spending money to learn nothing.

Concentration risk. A pure marginal-return optimisation will concentrate budget in whichever channel currently performs best, which maximises expected return and also maximises exposure. If one platform holds seventy per cent of your acquisition, a policy change, an auction shift or an account issue becomes an existential event rather than a bad quarter. Impose a concentration cap as a deliberate insurance cost, and know what premium you are paying for it.

Adjustment speed. Channels do not respond instantly. Paid search adjusts within days; content and SEO adjust over months; brand investment over quarters. Reallocating rapidly between channels of different response speeds produces oscillation, where you cut a slow channel before its effect has appeared, observe no loss, and conclude it was not working. Match reallocation cadence to the slowest channel in the portfolio.

The refinement worth adding as your data improves: because marginal return declines as you add spend to a channel, the optimum is a point on a curve rather than a fixed number. Moving budget into a channel lowers its marginal return; the correct allocation is where the curves intersect, not where the current best performer is.

• Move budget from lowest to highest marginal return until they converge.
• Constraint 1: respect minimum viable spend — fund properly or remove.
• Constraint 2: cap concentration deliberately, and treat the cost as insurance.
• Constraint 3: match reallocation cadence to the slowest-responding channel.
• Adding spend lowers a channel's own marginal return; the optimum is where curves intersect.

8. Step Five: Structure the Portfolio Before Optimising It

Pure optimisation of a single pool systematically eliminates two things the business needs, because both lose on short-term measured efficiency. Structure the budget into three pools first, then optimise within each.

Pool one: capture. Channels harvesting existing demand, optimised hard on marginal return and payback. This is where the marginal-return method applies most cleanly, because measurement is strongest and feedback is fast.

Pool two: creation. Channels building future demand, sized as a deliberate share and protected from comparison against capture on cost per acquisition. Assess through incrementality tests and leading indicators — branded search volume, direct traffic, category search share, and the volume and conversion rate of the capture channels it feeds. The sizing logic belongs in your [marketing mix plan](/guides/marketing-mix-plan-guide).

Pool three: testing. A ring-fenced share, commonly ten to fifteen per cent, funding experiments in new channels, new audiences, new creative approaches and new offers. Its return is information rather than revenue, and it will lose every direct comparison to an established capture channel. That is why it must be ring-fenced rather than competing.

The reason for the structure is behavioural as much as analytical. In a difficult quarter, budget is cut where the case is weakest, and the case is always weakest for the thing whose return is lagged or informational. Ring-fencing does not make creation and testing immune to cuts; it makes cutting them a visible decision that someone has to argue for, rather than a default outcome of an efficiency review.

A reasonable starting structure for a mid-sized business with established demand: the majority in capture, a meaningful protected minority in creation, and ten to fifteen per cent in testing. Earlier-stage businesses should weight capture more heavily until they have evidence that demand exists at all — creating demand for a product nobody has validated is an expensive way to learn something a smaller test would have taught you.

• Three pools: capture (optimise hard), creation (protect and test), testing (ring-fence).
• Creation is assessed on incrementality and leading indicators, never on CPA.
• Testing returns information, not revenue — it loses every direct comparison, which is why it needs ring-fencing.
• Ring-fencing makes cuts visible decisions rather than default outcomes.
• Earlier-stage businesses should weight capture until demand is validated.

9. Step Six: Reallocation Cadence and Triggers

Allocation is not an annual event, and it is also not a daily one. Both extremes fail, in opposite directions.

Reallocating continuously chases noise. Weekly performance variation in most channels is dominated by randomness, seasonality and creative cycles rather than by any change in underlying return. A team that moves budget weekly is mostly responding to variance, and will systematically cut channels after a bad week and add after a good one — which is the reverse of what the data supports.

Reallocating annually ignores evidence. A year is long enough for a channel's economics to change completely, and a budget fixed for a year cannot respond.

The workable cadence for most businesses: review monthly, reallocate quarterly, with stated triggers permitting an out-of-cycle move. Monthly review builds the evidence; quarterly reallocation acts on it once a trend is distinguishable from noise.

Define the triggers explicitly in advance, because the value of a written trigger is that it removes the argument at the moment it would otherwise happen. Useful triggers include: marginal return in a channel falling below the economic gate for two consecutive review periods; a channel reaching its concentration cap; a test channel meeting its graduation criteria; a platform or policy change materially altering a channel's economics; and a demonstrated capacity constraint downstream, which should reduce acquisition spend rather than increase it.

That last trigger is worth emphasising because it is routinely missed. If sales capacity or delivery capacity is the binding constraint, additional acquisition spend produces leads that queue, age and convert worse. The correct allocation response to a capacity constraint is to reduce spend or shift it toward higher-qualification-rate channels — not to keep buying volume the business cannot process. Diagnosing this correctly depends on funnel instrumentation, which is why [funnel stage design](/guides/funnel-stages-guide) and budget allocation are the same problem viewed from two ends.

• Review monthly, reallocate quarterly, with written out-of-cycle triggers.
• Weekly reallocation chases variance and systematically buys high and sells low.
• Triggers: gate breach over two periods, concentration cap, test graduation, platform change, downstream capacity constraint.
• A capacity constraint calls for less acquisition spend or a shift to higher-qualification channels, not more volume.

10. Allocating When You Have Little Data

Early-stage businesses have no response curves, no reliable marginal estimates and often no meaningful history. The method still applies; the inputs are weaker and the strategy should reflect that.

Sequence rather than spread. With a small budget, funding five channels at a fifth each usually produces five unreadable results. Fund one or two channels above their minimum viable level, learn, then add. Learning velocity matters more than portfolio balance at this stage.

Start with capture. If people are already searching for what you sell, capture channels give you the fastest read on whether your offer converts and at what cost. If nobody is searching, that is itself the most important finding available, and it tells you the constraint is demand creation or category education.

Buy information deliberately. Early spend is partly a research budget. Structure it to answer specific questions: which segment converts best, which message works, what acquisition cost is achievable. Write the question before the spend.

Use payback rather than lifetime value for decisions. Early LTV estimates are extrapolations from short histories and are usually optimistic. Payback period is observable within months and is far more robust as a decision input.

Set a stopping rule per channel before you start: how much you will spend before concluding, and what result would count as success. Without one, channels are kept alive by sunk cost, and a business with three zombie channels has no budget left for the one that would have worked.

As data accumulates, transition to the marginal method. The trigger is usually having enough conversion volume per channel per month that a twenty per cent spend change produces a distinguishable result — below that, you are reading noise, and disciplined sequencing beats optimisation.

• Sequence channels rather than spreading thinly — learning velocity beats balance early.
• Start with capture; if there is no demand to capture, that is your most important finding.
• Write the question each pound of early spend is buying an answer to.
• Use payback, not LTV, while lifetime estimates are extrapolations.
• Set a stopping rule per channel before starting, or sunk cost will set it for you.

11. How Allocation Differs by Business Model

B2B with long sales cycles. Attribution windows are frequently shorter than the sales cycle, which means channels influencing early consideration appear to do nothing. Weight creation more heavily than short-window measurement suggests, and lengthen your measurement windows to at least the ninetieth percentile of your sales cycle. Marginal return should be measured on pipeline created rather than on revenue closed, because revenue lags too far to steer with.

D2C and e-commerce. Contribution margin after cost of goods, shipping, payment processing and returns governs the gate — several businesses have discovered profitable-looking ROAS that was loss-making on contribution. Repeat purchase rate materially changes allocation: a channel delivering customers who repeat can justify a higher acquisition cost than its first-order return implies, and channels frequently differ substantially on this. Segment marginal return by first-purchase versus cohort value.

Subscription software. Payback period is the governing constraint, and it should be a first-class allocation input rather than a reported metric. Expansion revenue means the customer's contribution grows over time, which justifies higher acquisition cost — but only if net revenue retention actually supports it, and that must be measured rather than assumed.

Local and multi-location. Allocation is per-location, and the binding constraint differs by market. A location with high awareness and low capacity needs a different allocation from a new location with no awareness. Aggregate national allocation across heterogeneous markets is a common and expensive error.

Marketplaces. Budget splits across supply and demand acquisition, and the correct split depends on which side is currently liquidity-constrained. Over-investing in demand while supply is short produces poor conversion and wasted spend, and the metric looks like a marketing problem while being a supply problem.

Considered high-value consumer purchases. Long consideration windows with few purchases per person mean creation channels dominate and last-click measurement is close to useless. Incrementality testing is not optional here; it is the only reliable evidence available.

12. Common Mistakes, and What to Do Instead

Allocating on average ROAS. This is the central error, and it produces the standard pattern of scaling the best-looking channel until it degrades. Instead, estimate marginal return, even crudely.

Allocating on last-click attribution. Last-click systematically over-credits capture and under-credits creation, so allocation on it converges on a capture-heavy portfolio that looks efficient while growth decelerates. Instead, use incrementality testing for between-channel decisions and keep attribution for within-channel optimisation.

Never testing spend decreases. Teams test increases constantly and decreases almost never, which means saturation is discovered only by scaling into it. Instead, deliberately reduce spend in a channel periodically. A reduction that costs nothing is the cheapest possible discovery.

Funding many channels below viability. Instead, fund fewer channels above their minimum and add as budget grows.

Cutting the test budget under pressure. Instead, ring-fence it and require an explicit decision to cut.

Ignoring cash and payback. A portfolio optimised purely on return can be unfundable. Instead, apply the payback constraint as a hard gate.

Increasing spend when the constraint is downstream. Instead, check capacity and qualification rate before adding budget; if leads are queuing, more leads make the problem worse.

Holding budget constant all year. This maximises short-term stability and guarantees you learn nothing about your own response curves. Instead, vary spend deliberately so that next year's allocation has evidence behind it.

13. A Worked Example (Illustrative Model)

The figures below are an illustrative model constructed to demonstrate the method. They are not client results, and the response curves are simplified for clarity.

A business has a monthly acquisition budget of £100,000 across three capture channels. Reported average returns are: paid search 4.2, paid social 3.1, marketplace listings 5.4. On average return alone, the obvious move is to shift budget toward marketplace listings.

Spend variation testing over three months produces a different picture. Paid search: a 25% increase produced a 22% increase in conversions, implying a marginal return near 3.7 — the channel is efficient and only mildly saturated. Paid social: a 25% increase produced an 8% increase in conversions, implying a marginal return near 1.0 — heavily saturated, and the next pound is roughly breaking even. Marketplace listings: a 25% increase produced only a 4% increase, implying a marginal return near 0.9 — the high average reflects a small pool of very high-intent demand that is essentially fully harvested.

The correct action inverts the average-based conclusion. Marketplace listings, the best channel by average return, cannot absorb more budget — its demand pool is exhausted, and additional spend buys almost nothing. Paid social is saturated at current creative and audience settings. Paid search has the highest marginal return and should receive the reallocation.

Applying the rule: move budget out of paid social and marketplace listings into paid search until marginal returns converge. In this model that means a meaningful shift toward search, capped by search's own declining marginal return as it absorbs the additional spend, and capped again by the concentration limit — because moving too much into one channel creates the platform dependency discussed earlier.

Two secondary conclusions follow, and they are the more valuable ones. First, paid social's saturation is at current creative and audience settings, which is a testable constraint: new creative or a new audience may reset the curve, and that test is worth funding from the test pool before writing the channel down. Second, and more strategically, all three channels are capture channels harvesting existing demand — and two of the three are at their ceiling. No reallocation among them solves that. The business needs demand creation to raise the ceiling, which is a portfolio structure decision rather than an allocation one.

This is the common shape of a real allocation analysis. The reallocation is worth doing and is rarely the main finding. The main finding is usually a structural one that the allocation exercise surfaces.

14. Putting It Together

Deciding how much budget goes to each channel is a solved problem in principle and a measurement problem in practice. The principle is to equalise marginal returns subject to your gates and constraints. The practice is estimating marginal return well enough to act on, which requires deliberately varying spend and running genuine holdouts.

The sequence: size the envelope from contribution margin and payback tolerance; set economic gates, separately for capture and creation; estimate marginal return per channel using the strongest method your data supports; structure the portfolio into capture, creation and testing before optimising within each; then reallocate on a stated cadence with written triggers.

The single highest-value habit is testing spend decreases as well as increases. Most teams have never deliberately reduced spend in a channel to see what happens, which means they have no idea where their saturation points are, and they will find them only by scaling into them expensively.

The single most common structural finding is the one in the worked example: reallocation among capture channels cannot fix an exhausted demand pool. If your capture channels are saturated, the answer is upstream, and the decision belongs in the [marketing mix plan](/guides/marketing-mix-plan-guide) rather than in the media plan.

If you want the allocation modelled against your actual channel data before the next budget cycle, that analysis is where our [ROAS optimisation](/solutions/roas-optimization) engagements begin.