A channel can report a $100 CAC while its last $50,000 buys customers at $250 each. That gap is the core problem of measuring an enterprise marketing budget. Spend is split across channels that convert on different timelines. One deal may touch a dozen of them, and the revenue lands one to three quarters after the money goes out.
Blended averages paper over all of that, which is why they survive planning meetings and then quietly mislead the next allocation. A budget framework earns its keep by keeping the link between spend and incremental revenue visible long enough to act on it.
Channel allocation, attribution, ROI, and rolling governance only work as a set. Marginal CAC decides where the next dollar goes, attribution decides who gets credit for the return, and governance decides how often both are allowed to change.
Treat them as one loop and the budget adapts each quarter instead of hardening in January. That discipline matters more at scale, where the commitments are large, the planning cycles long, and the signals scattered across markets, audiences, and product lines.
Four channel types run on four different clocks
The first mistake is treating a budget as one pool. Enterprise spend lives in four channel archetypes that behave nothing alike, and a plan has to fund all four to keep growing.
- Always-on performance channels (search, paid social retargeting) are measurable, high-intent, and scale with a predictable marginal CAC until they saturate.
- Demand-creation channels (video, discovery ads, influencers) buy lower-intent attention over longer attribution windows and show up later as branded search and direct traffic.
- Relationship and lifecycle channels (email, CRM, in-product messaging) cost little at the margin and pay back mostly through retention and expansion.
- Brand and strategic initiatives (sponsorships, thought leadership) move over multiple quarters and influence assisted conversions that no click model will cleanly credit.
Within that mix, the number that should drive decisions is marginal CAC: what the next customer costs. Blended CAC, the average, only describes the past. Every channel eventually gives less for each extra dollar. So the useful question is what the next dollar here costs compared with the LTV it brings back. Plot that curve for each channel, with payback time next to it, and you can send spend where it adds the most.
A quick illustration makes the gap concrete (the numbers are illustrative). Say a channel has absorbed $100,000 and returned 1,000 customers: blended CAC reads a comfortable $100. Push another $50,000 in and, because the channel is saturating, it returns only 200 more customers. The marginal CAC on that last tranche is $50,000 / 200 = $250, or two and a half times the blended figure, and quite possibly above the LTV those customers carry. The blended number, meanwhile, still reads about $125 across the full $150,000, so it keeps waving the spend through while the next dollar has quietly stopped paying back. That is the whole argument for pricing the increment:
Blended CAC = total spend / total customers
Marginal CAC = extra spend / the extra customers it actually produced
Because those curves shift, money should move only when a test says so. Geo lift and incrementality tests show what spend actually caused: you switch a channel off in some regions and see whether sales drop. Creative and audience tests show what works inside a channel. The same experiments check whatever multi-touch model the team relies on. Without that testing layer, reallocation is just the loudest stakeholder's intuition.
No attribution model is true, so run three against each other
Every attribution model hides something. The workable approach triangulates impact from methods that each hide something different. Rules-based models (first-touch, last-touch, linear, position-based) are simple to communicate and easy to defend in a meeting, but they are fragile and actively misleading over long enterprise cycles.
Statistical models (data-driven attribution, media-mix modeling) combine more signals and hold up better. The price is lots of data and a real data-science team. Causal and experimental methods (geo experiments, audience splits, incrementality tests) are the only ones that measure true lift, which is why they anchor the others.
In practice, enterprises run all three side by side. Multi-touch attribution handles user-level signals in digital channels, media-mix modeling captures macro trends and offline spend that MTA never sees, and experimentation validates whichever prediction the other two produce. When they disagree, compare the decision scope, time window and assumptions, then check whether the experiment can support the allocation being considered. Read this way, attribution earns its cost by settling arguments about budget, creative investment, and market expansion, which another dashboard would never do.
A defensible ROI counts only what marketing caused
A defensible ROI reflects incrementality, time horizons, and unit economics, alongside the revenue booked. Five metrics are worth reporting. Marginal CAC, never blended. LTV built from ARPU, retention, expansion and churn. Payback in months. The incremental revenue actually caused by marketing. And the contribution margin left after the cost of serving the customer.
Blended CAC is the number that most often lies. The average hides a saturated channel, a worn-out audience, tired creative, rising competition and whole blocks of wasted spend. Marginal CAC shows them.
Time is the other trap. Enterprise sales cycles are long. Many deals close months after the first touch, and seasons shift the numbers, so an ROI figure has to wait for the late conversions. A month is too short a window to judge a channel whose deals close two quarters out; a quarterly decision window keeps the budget responsive without pretending that last month's number is the verdict.
Re-forecast four quarters ahead, every quarter
Governance sets how often each of those numbers is allowed to move. The core mechanism is a rolling four-quarter budget, re-forecast every quarter against updated marginal CAC, ROI projections, channel saturation, and new opportunities.
Around it runs a cadence of decisions at three altitudes. Monthly reviews watch channel performance, saturation alerts, CAC and LTV shifts, and creative-test outcomes; quarterly reviews actually reallocate budget, update attribution models, and make market-expansion calls; and annual or biannual reviews reset brand strategy and portfolio-level investment. The rule of thumb is that the frequency of the meeting should match the speed of the signal it governs.
Governance also means rehearsing the quarter you hope not to have. Simulating recession scenarios, CPC and CPM inflation, demand spikes, a new competitor, regional expansion, or slower brand payback turns uncertainty into a plan, so when acquisition cost jumps, the reallocation is already understood.
The brand-versus-activation split has its best-known evidence base in Les Binet and Peter Field's analysis of the IPA Effectiveness Databank, published as The Long and the Short of It (2013). Their often-quoted conclusion is that, on average, roughly 60% of spend on long-term brand building and 40% on short-term activation maximized effectiveness, with the balance shifting by category and, in later work, tilting further toward activation in B2B. Treat it as a starting hypothesis to test against your own sales cycle. As a default, it still beats letting last quarter's attribution report decide.
When marginal CAC misleads too
Teams that have tried to run this in practice hit the same wall. Marginal CAC is easy to define and hard to measure in an enterprise, because the customers a tranche buys arrive late. Take the same channel and suppose that only 40% of the deals the extra $50,000 eventually produces close inside the quarter. At quarter end the team sees 80 new customers instead of 200, and the measured marginal CAC reads $625, two and a half times the true $250. Cut on that number and you starve a channel that was still paying back. Media-mix models have their own lag: they typically need two to three years of weekly spend and outcome data before their channel curves stop moving with every refit.
That is a measurement problem, and it does nothing to rescue the blended number, which arrives just as late and averages the delay away. The marginal figure is simply slower to read, so the rule has to build in time. Judge a tranche only after the window in which, historically, most of a channel's conversions arrive, and until then read in-quarter marginal CAC as an upper bound that will fall as the late deals close. Where the cycle is longer than two quarters, use a geo holdout or an on/off test instead: switch spend off in a matched set of regions and read pipeline, not closed revenue, at 90 days. And when a channel's lag-adjusted marginal CAC sits above a third of LTV for two consecutive quarters, the common 3:1 rule of thumb, move the next tranche elsewhere and log the call so the next review can check it.
Buyer behavior makes the lag worse in B2B. A complex purchase usually runs through a committee, and most of its deliberation happens in internal meetings, forwarded decks and calls with peers, long before anyone fills in a form. The influence a campaign has there is real and invisible to any tracking pixel, which is one more reason the experiment, not the click path, has to settle the allocation.
Metrics that only start to make sense at enterprise scale
The gap between blended and marginal CAC has a cousin at the funnel level. A startup with one product and a short cycle can read its funnel almost click by click; an enterprise selling high-consideration products cannot, because no single touch tells the story. That is why blended funnel metrics, noise for a startup, become essential at scale: awareness to consideration to conversion, branded-search lift, direct-traffic uplift, sales-assisted conversions, lead-to-opportunity rate, and pipeline contribution. They are useful precisely because they capture the demand-creation and brand channels that click models undercount, and risky for the same reason, so they have to be read as directional.
Efficiency metrics put a price on what the funnel shows: what each sale that marketing really caused costs, what a qualified opportunity costs, whether deals close faster, and how a channel's average ROAS compares with the ROAS of its last dollar. That is the same average-versus-next-dollar comparison, applied to return instead of cost. Above both sits the portfolio view, because at this scale marketing behaves like a portfolio of bets: the split between brand and performance spend, ROI by region, channel diversification, allocated versus utilized budget, and the team's own capability maturity.
That last item is easy to skip, yet the rest depends on it. The whole system quietly assumes marketing leaders who can hold their own on unit economics, attribution triangulation, scenario modeling, statistical literacy, and experiment design, and who share one set of definitions with finance, sales, product, and data science. When marketing and finance argue from different numbers, no framework survives the first bad quarter.
Six questions for the quarterly budget review
Each one takes a number or a yes/no. A "no", or a number nobody can produce, marks where the framework is not running yet.
- For every channel that got more money last quarter, what did the last tranche cost per customer, and is that below a third of LTV?
- Has each channel's lag window closed, so the marginal figure is final?
- Which allocations rest on an experiment (geo holdout, on/off, audience split), and which on a model alone?
- Where multi-touch attribution and the media-mix model disagree by more than 20%, is a test scheduled to settle it?
- Is payback per channel under the agreed limit, say 12 months, measured on contribution margin after cost-to-serve?
- Do marketing and finance use one written set of definitions for CAC, LTV and pipeline?
Questions 1 and 2 pull against each other on long cycles, so answer the second before acting on the first. Monthly checks track these numbers; only the quarterly review moves money. Brand spend is read against branded-search and direct-traffic lift, because it will always be attributed less cleanly than it contributes. A startup with one product and a 30-day cycle can skip most of this list; at enterprise spend, a wrong quarter of allocation costs more than the whole measurement team.
Where the next dollar stops paying back
The illustration near the top carries most of the argument in two numbers. A channel that reads about $125 blended while its last tranche costs $250 per customer is a channel the January plan will keep funding, and nothing in a blended report will flag it. Budget arguments change character once the team can see marginal CAC by channel: the disagreement moves from whose channel is good to where the next dollar stops paying back.
Everything else in the framework exists to keep that view honest over time. Marginal efficiency and strategic importance set the allocation, experiments move the money, attribution assigns the credit, and the rolling four-quarter cadence decides how often any of it may change. Run that way for a year, the January plan becomes an opening position. Each quarter's evidence revises it, and with two-quarter sales cycles no other kind of plan is affordable.