Paid Media Budget Allocation: When to Expand Channels vs. Invest More in What Works

Most teams reach the deepen-or-diversify question at the same moment: the primary channel still reports a healthy blended ROAS, growth has slowed, and someone proposes a new platform. The envelope rarely grows to accommodate the debate. Gartner's 2026 CMO Spend Survey puts marketing budgets at 7.8% of company revenue, up from 7.7% in 2025, so a new channel bet is usually funded by moving money away from something that already works. This guide sets out a paid media budget allocation method for exactly that decision. It starts with the total envelope, gives a marginal-return test for the core channel, turns the 70-20-10 split into a promotion ladder with written gates, and shows how overlap tests, incrementality and media mix modeling should feed each reallocation. The percentages come last, because they should fall out of the evidence rather than lead it.

Most allocation arguments are measurement arguments in disguise. When a team debates adding TikTok against pushing Meta harder, the number that would settle it is usually missing: what the last increment of spend cost per incremental customer. Platform dashboards report averages, and averages are kind to saturated channels. GPI's position is that the split should follow the decision. Decide what you need to know first (is the marginal customer still profitable, does this channel create demand or harvest it), choose the cheapest measurement that answers that question, and let the percentages emerge from the result. A partner who proposes a split before proposing the test has the sequence backwards.

TL;DR: deepen or diversify?

• Keep funding an existing channel only while its marginal CAC, the cost of the next customer rather than the average customer, stays inside your target. Blended ROAS hides saturation because early, cheap conversions subsidise the expensive later ones.
• Open a new channel when marginal cost on the core channel is climbing toward target after you have refreshed creative and widened targeting, or when one platform carries so much new-customer volume that finance would refuse the same concentration in any other line item.
• Treat 70-20-10 as a governance default, not a law. Write down the conditions under which a test channel graduates to core budget and a core channel gets demoted, before the test budget is released.
• Do not cut a channel on platform attribution alone. Run a sequential pause, geo holdout or conversion lift study first; overlap and cannibalization distort last-click numbers in both directions.
• Budget creative production, measurement and compliance review inside the paid media envelope. Go Fish Digital cites a Motion 2026 ecommerce benchmark attributing 70% of Meta campaign performance variance to creative, which makes creative a media cost in practice.
• Review on a fixed cadence: monthly for pacing, quarterly for the split. Change allocation when a pre-agreed trigger fires, not when a dashboard has a bad week. Measured describes the triangulated measurement stack those triggers should draw on.

Set the total envelope before you argue about splits

Why the size question comes first

Start by writing down the total paid media envelope for the year, because every later argument is really an argument about a share of it. The total is more stable than the split, and it is where finance has the strongest view. Gartner's 2026 CMO Spend Survey reports marketing budgets at 7.8% of company revenue, up from 7.7% in 2025. That is effectively flat. The practical consequence for paid media budget allocation is that a new channel is funded by taking money from an existing one, not from growth in the envelope, and the diversification case has to be argued as a trade rather than an addition.

What enterprise benchmarks can and cannot tell you

The Gartner figure comes from 401 marketing leaders at companies with revenue above $1 billion. It describes how large, established organisations budget in a flat year. A growth-stage company deciding how much to spend on acquisition should record that population next to the number and treat it as context, not a target. Percentage-of-revenue benchmarks say nothing about your gross margin, payback tolerance or the capital available for acquisition, which are the inputs that actually set the envelope. Broader budgeting guides such as Harvard Business School Online's and Improvado's cover top-down budgeting steps, but the same caveat applies to any benchmark they quote: check whose spend it describes.

The CMO Survey's November 2024 edition offers a second, historical reading. It reported digital marketing spend growth rising from 8.9% to 11.1%, with marketers forecasting a further 12.7% increase over the following twelve months. That is sentiment from late 2024, self-reported by US marketers, and it should not be read as a 2026 condition. It is useful mainly as a reminder that forecast optimism and actual envelopes diverge.

Acquisition versus retention is an allocation decision too

The same November 2024 survey found marketers spending 19.6% more on acquiring customers than retaining them. The deepen-or-diversify debate is almost always conducted inside the acquisition bucket, which means a prior allocation choice has already been made without discussion. Surface it to the CFO explicitly.

A practical way to write the envelope: a base figure committed for the year, plus a named contingency tranche (sized by your own risk tolerance) that can be released to a scaling or diversification decision without reopening the whole budget. The tranche is what later sections of this guide allocate.

The marginal-return test: when more of the same stops working

Average ROAS versus marginal CAC

Marginal CAC is the incremental cost of the incremental customers produced by the last spend increase. If you raised a channel's monthly budget from 100 to 120 units and acquired 20 more customers than the prior month after adjusting for seasonality, marginal CAC is 20 units of spend divided by 20 customers, whatever the blended figure says. Blended ROAS averages the cheap early conversions with the expensive later ones, so it can stay above target while the newest tranche of spend loses money. This is the central mechanism behind the deepen-or-diversify decision, and it is why GPI's earlier guide on scaling paid media when increasing spend stops increasing growth treats diminishing returns as a measurement problem before a channel problem.

A hypothetical, built purely to show the arithmetic and not drawn from any account, campaign or benchmark: suppose a channel spends 500 a month at a blended 4x ROAS, returning 2,000. Split out the last 20% of spend, the final 100, and suppose it returned 180, or 1.8x. The blended number still looks excellent. The marginal view says the last 100 is barely covering cost of goods, and the next 100 will likely do worse. The figures are chosen for clarity, not measured; the point of the model is that blended ROAS and marginal CAC can point in opposite directions on the same channel in the same month. The question shifts from whether the channel is good to where the next 100 should go.

Five saturation signals that show up before the blended number moves

• Prospecting frequency creeping upward at constant reach.
• Rising CPM while reach stays flat, meaning you are paying more to show the same people the same ads.
• Shrinking share of new-to-brand customers in the channel's conversions.
• Conversion rate decaying as daily budgets rise.
• Growing audience overlap between campaigns that were meant to reach different segments.

Vertical scaling: what usually restores efficiency before you leave the channel

Exhaust the in-channel levers first, because they are cheaper than a new channel and the existing channel has already absorbed its setup costs. Creative is the largest lever on social. Go Fish Digital, writing in August 2026, relays a Motion 2026 ecommerce benchmark finding that creative accounts for 70% of campaign performance variance on Meta, more than targeting, bidding and budget allocation combined. That is vendor benchmark data reported by an agency, drawn from ecommerce accounts on one platform; it should not be assumed to transfer to search, where query intent does much of the work creative does on social. Second, widen targeting. Straight North's commentary is that broad targeting has become effective when paired with strong creative and Meta's optimization, a shift from the narrow-audience era. Third, revisit bid strategy and landing pages, which are frequently the actual cause of conversion rate decay at higher budgets.

Horizontal scaling: the case for a second channel

Two triggers justify a new channel. The first is marginal CAC on the core channel sitting within a tolerance of target that you set in advance, measured after a creative refresh, not before. The second is concentration risk: one platform carrying most new-customer volume exposes the business to a policy change, an auction shift or an account issue that finance would not accept in any other supplier relationship. Cross-platform guides such as Growth Engines' frame the second channel as portfolio management for that reason.

The answer is rarely binary. The realistic move is to shift the increment, the contingency tranche or the next planned increase, into a test, while the core channel keeps its base funding. The same discipline applies when you evaluate a partner: an agency that can explain how it sizes and reads a second-channel test is easier to trust than one that leads with a target split, which is one of the questions in GPI's guide to choosing a paid media agency.

A diagnostic table for the deepen-or-diversify call

The creative and targeting rows draw on the Motion benchmark relayed by Go Fish Digital and the broad-targeting commentary from Straight North; the other rows are operational diagnostics rather than published findings.

Turning 70-20-10 into a promotion ladder

What the rule is actually for

The 70-20-10 split puts roughly 70% of paid media budget into proven channels, 20% into scaling bets and 10% into experiments. It is a governance default. It prevents over-concentration in one channel and prevents the opposite failure of scattering small tests that never conclude. None of the captured sources derive it from a study of optimal allocation, and Go Fish Digital presents it as a structuring device for ecommerce budgets rather than a measured optimum. Its weakness in practice is that most guides treat the three buckets as static. Without rules for movement between them, test channels sit in the 10% bucket for years and core channels stay funded past saturation.

Promotion gates: how a test channel earns core budget

Write the gates before the test spends anything. In order:

1. Minimum duration: the test runs through at least one full purchase cycle for your product, so early or late converters are counted.
2. Minimum spend floor: set from the channel's expected CPA, so the test can generate enough conversions to read. A test that cannot produce a usable conversion count at its budget is a decision not to test.
3. Marginal CAC tolerance: the test channel's marginal CAC, as defined in the previous section, lands within a tolerance of the core channel's marginal CAC that your team named in advance.
4. Incrementality read: the channel's contribution is confirmed by a holdout or geo test, not by platform attribution alone. Measured's triangulated framework positions incrementality testing as the causal channel-level check, with platform attribution reserved for tactical optimisation; a promotion decision belongs at the causal layer.

Demotion gates: how a core channel loses it

A core channel is demoted to the 20% bucket, or below, when either condition holds: marginal CAC sits above target for two consecutive reviews after a creative refresh, or the measured incremental contribution comes in materially below the platform-reported figure. Demotion does not mean exit. It means the channel's next increment is no longer automatic.

The incrementality row reflects Measured's layering of causal testing above platform attribution (Measured). All thresholds in the table are placeholders for values your team sets; none of the captured sources establish an industry-standard cut-off.

Sizing the 20% so a test can actually reach significance

Test budgets fail silently. Ten percent of a small envelope, split across three platforms, produces a handful of conversions per channel and a verdict nobody trusts, so the channels stay in limbo. The honest alternative is fewer, larger tests: one channel at a time, funded from the combined 20% and 10% if necessary, with the spend floor from gate two respected. A single conclusive test is worth more than three inconclusive ones, and it frees the next quarter's test budget for the next candidate.

Review cadence and what is allowed to change between reviews

Monthly: pacing within each bucket, bid strategy and creative rotation. Quarterly: the split itself, including promotions and demotions against the gates above. Annually: the envelope. Between reviews, the split does not move. An out-of-cycle change is permitted only when a named trigger fires, such as a platform policy change that removes a targeting capability, a measured incrementality result that contradicts the current allocation, or a supply shock that moves CPMs beyond a pre-set band. Write those triggers into a one-page charter with sign-off owners; that document, not the split, is the deliverable. If you want a reference point for what documented evidence looks like from the buyer's side, GPI's directory methodology sets out how it checks a partner's claims against their supporting material, and the same standard is a reasonable bar for your own charter.

Funnel split and the brand-versus-performance balance

What the actual split looks like

The CMO Survey's November 2024 edition reported that marketers allocate 68.8% of budget to short-term performance and 31.2% to long-term brand building, while stating that their ideal split would be 50/50. This is self-reported US data from late 2024, so it describes a historical sentiment rather than a 2026 measurement, but the gap between actual and ideal is the useful part. Marketers know they are skewed and cannot correct it. Search Engine Journal treats the upper-funnel share as an open question rather than a fixed percentage, which is the right posture.

Why the short-term skew is rational and still risky

The skew persists because performance spend is attributable inside the quarter and upper-funnel spend is not. Incentives follow what measures, and quarterly reviews reward the channels that show up in attribution. That is rational behaviour at the individual level and a portfolio risk at the company level: demand that performance channels harvest has to be created somewhere, and a plan that only funds harvesting slowly exhausts its own supply.

How to fund upper funnel without abandoning performance discipline

Give upper-funnel spend its own line, its own metrics and its own review cadence, so it is never judged against conversion channels on last-click. Suitable reads include brand lift studies, branded search volume trends, new-to-brand share across the account and MMM contribution where the data history supports a model. This matters for the deepen-or-diversify choice too: a diversification move is often into an upper-funnel channel (video, audio, connected TV), and evaluating it on conversion-channel metrics guarantees a false negative.

The skew that makes the demand-creation row vulnerable is documented in the CMO Survey, November 2024; the channel groupings and time-to-read estimates are practitioner framing rather than measured figures.

The downturn question

BCG's 2023 analysis found that companies which cut brand marketing during downturns experienced market share loss, reduced sales and higher long-term costs of rebuilding the brand afterwards. That is macro-level historical analysis; outcomes depend on category elasticity, competitive response and how much of a company's demand is habitual. It does not mean upper funnel should be protected at any price. It means the cut should be modelled as a deferred cost, not a saving, and presented to finance in those terms.

Measuring overlap and cannibalization before you cut

Where overlap hides in attribution

Three mechanisms inflate channel numbers. Branded search absorbs demand created by other channels, so its ROAS looks exceptional while it mostly captures people who had already decided. Retargeting claims conversions from users who would have converted anyway, because the audience is defined by intent it did not create. Two prospecting channels reaching the same users each take credit for the same conversion, so both look better than they are. Signal loss makes this worse: Marketing Dive reports IAB findings that privacy legislation and signal loss are shifting budgets, and modelled conversions fill gaps with assumptions that can compound the overlap.

Three tests that isolate it

In rising order of cost and rigour:

1. Sequential pause. Pause the channel for a pre-agreed period against a baseline defined in advance, with a seasonality control such as the same period last year or a comparable unaffected market. Cheap, but vulnerable to anything else that changed during the pause.
2. Geo holdout. Match markets, withhold the channel in the holdout set, and compare outcomes. Isolates the channel's causal contribution reasonably well; requires enough geographic volume to matter.
3. Platform-supported conversion lift. Where the platform offers randomised holdout at the user level, it isolates that platform's contribution cleanly, though it relies on the platform's own conversion measurement and cannot see cross-channel effects.

Measured positions incrementality testing as the causal channel-level validation layer, distinct from platform attribution; the durations above are guidance on design logic, not published benchmarks.

Reading the result

The gap between platform-attributed conversions and incrementally measured conversions is your cannibalization estimate. Report it as a range with the test's uncertainty, never as a point value, because a single geo test has sampling error and a sequential pause has confounding. If the incremental read is far below attribution, the channel is harvesting. If it is close, the channel is creating or capturing demand that would otherwise be lost.

Overlap cuts the other way too. A channel with weak last-click numbers may be creating demand that branded search or retargeting later harvests. The same tests protect against the wrong cut as well as the wrong keep, which is why they belong before any exit decision.

When cutting is still the right call

Cut when incremental contribution stays below marginal cost after a test of adequate length. Do not cut because attribution share fell, because a dashboard changed, or because a new channel needs the money. The last case is the most common and the least defensible; if the diversification case is real, it should survive being funded from the contingency tranche while the test on the incumbent runs.

Incrementality and MMM as the reallocation engine

Three layers, three questions

Measured describes a triangulated framework in which incrementality testing validates channel-level causality, media mix modeling guides portfolio-level allocation, and platform attribution drives tactical optimisation. Each layer answers a different question. Incrementality: does this channel work at all, causally? MMM: how should the portfolio be weighted across channels and over time? Attribution: what should change this week? Each also has a different latency and cost, from near-real-time attribution to MMM refreshes that depend on months of history.

Matching the tool to the decision size

The decision's size should pick the tool. A daily bid change does not need a holdout. A quarterly shift of, say, 15% of the envelope between channels does, because the cost of being wrong is large and attribution alone cannot see overlap. An annual channel entry or exit should be informed by MMM where the data allows. MMM needs spend and outcome history across enough periods, genuine variation in spend (a channel that has always been funded at the same level teaches the model nothing) and external variables such as pricing, seasonality and distribution. Smaller advertisers usually lack that history and should lean on tests rather than models.

The layer definitions follow Measured's framework; the latency and cost characterisations are working estimates, not measured figures.

These layers map onto the promotion ladder from earlier: the spend-floor and duration gates are checked against attribution data, the promotion gate itself requires an incrementality read, and the annual envelope review is where MMM contributes if it exists.

What AI-driven tooling changes and what it does not

Gartner's 2026 survey finds CMOs allocating 15.3% of marketing budgets to AI initiatives, while 70% call AI leadership a critical goal and only 30% report mature readiness. Those figures describe planned spend and self-assessed readiness; they say nothing about whether automated budget tools improve outcomes. Search Engine Journal's discussion of AI-driven PPC budget rebalancing frames the same shift. The allocation implication is modest: automated pacing and bidding can execute a decision rule faster, but the rule, and the incrementality check that validates it, still need a human owner.

Budget lines that are not media but belong in the paid media plan

Creative production and refresh

If creative explains 70% of Meta performance variance in the Motion 2026 ecommerce benchmark that Go Fish Digital cites, then a media plan that funds impressions but not the assets shown in them is underfunding its largest lever. Size the creative production line to the media it supports, and add a refresh cadence for channels showing fatigue. Go Fish Digital treats creative as the line ecommerce brands most often get wrong. No captured source establishes a standard percentage for creative, so set it from your own production costs and refresh frequency.

Measurement and testing costs

Holdout tests and MMM have real costs: foregone revenue in holdout cells, tooling licences and analyst time. If those are not budgeted, the promotion ladder cannot function, because every gate that requires an incrementality read will be skipped for lack of resources. Treat measurement as a fixed line inside the envelope rather than a discretionary add-on.

Compliance, legal review and localization

Regulated categories carry costs that budgeting guides routinely omit. Financial services, health and B2B with procurement review add legal review time and creative adaptation to every new channel, which lengthens test duration and raises the minimum viable test budget. WOLF Financial's guide addresses paid media allocation in financial services specifically, a reminder that category rules reshape the envelope. Localization adds the same friction for multi-market advertisers.

AI tooling is a fourth line. Gartner's 15.3% figure shows the line exists at enterprise level; the question for paid media is whether it displaces working media or is funded separately.

The practical rule: when comparing a new channel with deepening an existing one, state the total cost of entry (media plus creative, compliance and measurement). The existing channel has already paid most of those costs, which is a real advantage the marginal CAC comparison alone does not capture.

What GPI looks for in a partner's allocation discipline

Questions to ask before handing over the envelope

Three questions expose whether a prospective paid media partner, or an in-house team, has an allocation process rather than a set of percentages. First: how do you distinguish marginal from average return on a channel? Second: how does a test channel graduate to core budget, and what demotes a core channel? Third: which measurement layer do you use for which size of decision? A partner who can describe something like Measured's triangulated stack in their own operating terms has thought about the problem. Vague answers on any of the three suggest the partner runs on percentages rather than a process. GPI's guide to choosing a paid media agency covers the wider selection process; these three questions are the allocation-specific subset.

Evidence a credible agency can show

Ask for the last reallocation the partner made and the evidence behind it, including a case where the incremental read disagreed with platform attribution. A partner who has never disagreed with a platform dashboard has not tested it. GPI's methodology weights documented, verifiable outcomes over self-reported ROAS, which maps directly onto these questions: an agency that reports blended ROAS without marginal or incremental context is presenting the number least likely to survive scrutiny. GPI does not run campaigns and makes no claim about any listed agency's results.

Profiles such as Admiral Media, Arda Media and Avalaunch Media can be read against the three questions as an exercise in checking claims against evidence; they are places to look, not endorsements.

Allocation is a decision process, not a percentage. The partner's job is to make that process legible to your finance team, with gates, cadence and measurement written down before the money moves.

Browse the paid media and performance marketing agencies listed on Growth Partner Index and read their profiles against the three allocation questions above.

FAQ

How should a company with a small paid media envelope run the 20% and 10% buckets when a 10% test cannot reach a meaningful conversion count?

Combine the two buckets and test one channel at a time. Set the spend floor from expected CPA first; if even the combined budget cannot produce a readable conversion count within one purchase cycle, defer the test rather than run an inconclusive one.

What is the minimum evidence for demoting a core channel, and how do you separate seasonal decline from genuine saturation?

Marginal CAC above target for two consecutive reviews after a creative refresh, with the comparison made against the same period in the prior year or a seasonality-adjusted baseline. Saturation shows in frequency and CPM at constant reach; seasonal decline usually does not.

How do you present a diversification test to a CFO who benchmarks everything against the core channel's blended ROAS?

Show the core channel's marginal ROAS on its last increment next to the blended figure, then frame the test as an alternative use of that increment, funded from the contingency tranche, with pre-agreed gates and a fixed end date.

If a business runs a single paid channel profitably, what concentration risk should trigger a second channel even when marginal CAC is fine?

Ask finance what supplier concentration it would tolerate elsewhere. If one platform's policy change or account action could remove most new-customer volume, a second channel is insurance, and its test should be judged on that basis as well as on CAC.

How much of the paid media budget should be reserved for creative production and measurement, given the Motion 2026 benchmark and the cost of holdout tests?

No captured source sets a percentage. The benchmark relayed by Go Fish Digital argues creative deserves a line sized to the media it supports; size measurement from the number of gate decisions you expect to make in the year.

Can MMM replace incrementality testing for quarterly reallocation, or are both needed?

Both, in Measured's framing: MMM weights the portfolio, tests validate individual channels causally. A quarterly shift usually needs the test; MMM informs the annual shape.

How often should the 70-20-10 split itself be revisited versus the pacing within each bucket?

Pacing monthly, the split quarterly, the envelope annually. Between reviews the split moves only on a named trigger.

Does the Gartner 7.8% of revenue figure apply to a growth-stage company deciding its paid media envelope?

As context only. The 7.8% figure describes 401 enterprises above $1 billion in revenue; a growth-stage envelope should be set from margin, payback tolerance and available capital.