If your paid media ROAS still looks “good” while CAC creeps up and Finance keeps side-eyeing the dashboard, the problem usually isn’t execution. It’s the ROI formula.
Platform reporting is happy to call revenue “marketing-driven.” Boards aren’t. And in 2026—when budgets are flat at 7.7% of revenue (research brief) and 35% of organizations report over 20% budget inefficiency from unreliable data (research brief)—optimism doesn’t survive the first CFO review.
If you only change one thing, change this: measure incremental net profit ROI, not revenue ROI. Avinash Kaushik lays out the logic cleanly in his piece on incremental net profit ROI, and the operator takeaway is blunt: most “ROI” numbers are missing costs and missing causality.
The scary giant inside “$1 in, $4 out” ROI
Kaushik opens with a metaphor that lands because it’s real in every ad account: the hidden cost structure—his “scary giant.” The most common ROI math, ROI = (Revenue – Media Cost) / Media Cost, is structurally biased to look better than the business reality.
Here’s the sequencing that matters. A campaign can report an ROI of 4—“$4 back for every $1 spent”—and still be a bad trade when you account for what the business actually keeps. Kaushik calls that kind of reporting “fake news” in spirit, because it creates confidence without truth (Kaushik, “Measure Marketing ROI: Incremental Net Profit ROI”).
First correction: subtract COGS. In Kaushik’s example, eyeglasses sell for $50 with $35 COGS. That single line item collapses the story from “we printed money” to “we made gross profit, maybe.” When COGS enters the equation, his example produces a much smaller ROI number (Kaushik).
Second correction: stop pretending “media cost” equals “campaign cost.” Kaushik shows a case where the total budget is $1.0M, but the initial ROI only counted $0.6M in media—leaving $0.4M of non-working costs out of the narrative. Put those costs back in, and the ROI can go negative fast (Kaushik).
That’s the first open loop closed: ROI inflation isn’t mostly fraud. It’s accounting.
The board-defensible step most teams skip: incrementality
Now the harder part. Even if the math includes COGS and full campaign cost, it still assumes all observed revenue happened because marketing happened. That’s attribution-by-assertion.
Incrementality is the fix: the portion of sales that would not have occurred without the campaign. Kaushik pushes the formula to its logical end by applying an incrementality assumption—30% in his example—and the result is sobering: an incremental net profit ROI of -0.7 (Kaushik). Same campaign. Same “return.” Completely different decision.
But the context in 2026 makes this more than a math debate. The research brief points to a credibility gap: 83% of leaders prioritize demonstrating ROI, yet only 36% can measure it accurately; 49% can’t defend their ROI methods to boards (research brief). That gap is where budgets get cut—because Marketing’s story doesn’t reconcile with Finance’s.
And there’s another layer: signal loss and cross-channel journeys make any single attribution view fragile. The research brief reflects the current consensus: teams increasingly need multi-model measurement—MMM for macro truth, MTA for journey-level directionality, and incrementality testing for causal proof (research brief). No one model earns trust on its own anymore.
So here’s the second loop closed: incremental net profit ROI isn’t “more conservative.” It’s the minimum bar for causal, profit-based decisioning.
One primary tactic: run an incrementality test that feeds incremental net profit ROI
Most teams try to “fix measurement” by buying another dashboard. The better move is simpler: run a holdout test designed to produce incremental net profit ROI as the readout, then use that number to set CAC guardrails and forecast with fewer assumptions.
The hypothesis (make it falsifiable): If we run a holdout-based incrementality test on Channel X for 4–6 weeks, then incremental net profit ROI will be lower than platform-reported ROAS because baseline conversions and non-working costs are currently being misattributed to paid media.
Step 1 — Define the ROI you’re actually going to defend. Use Kaushik’s end-state structure as the operating definition: incremental revenue minus incremental non-working costs, incremental COGS, and total campaign budget—divided by campaign budget (Kaushik). Directionally, the point is non-negotiable: profit, not revenue; lift, not credit.
Step 2 — Choose a holdout design you can execute. The research brief calls incrementality testing (geo-holdouts, audience splits, time-based tests) the “gold standard” for causal impact under attribution limits. Pick one based on constraints, not ideology. Geo-holdout is clean for localized demand; audience split works when targeting is stable; time-based can work when seasonality is low (research brief).
Step 3 — Pre-register success, guardrails, and the stop-loss. Don’t “read the tea leaves” mid-test. Lock the rules before launch so the result is defensible.
- Success (primary metric): Incremental net profit ROI (Kaushik-style) by test cell vs. control.
- Guardrails (secondary): Qualified pipeline created (definition owned by RevOps), and CAC payback proxy (stage-appropriate; research brief).
- Stop-loss threshold: If incremental net profit ROI trends below a pre-set floor for two consecutive readouts (weekly is typical), pause spend in the test cell and preserve learning. (Directional—set the floor based on gross margin and cash constraints.)
Run it this week: setup, launch, readout, next test
Setup (Day 1–2): Owner = Demand Gen + Marketing Ops; Finance partner for COGS/gross margin inputs; RevOps for pipeline definitions. Tools can be basic: ad platforms + CRM + a spreadsheet, as long as cost layers are complete. Audience: pick one paid channel where creative fatigue and auction volatility are already suspected, because that’s where attribution lies tend to be most expensive.
Budget range: Use an amount large enough to detect lift but small enough to survive being wrong. In practice, many teams start by holding out 10–20% of reachable audience or one or two comparable geos (directional guidance; the research brief doesn’t provide a universal benchmark, so treat this as an operator constraint, not a statistic).
Launch (Week 1): Freeze major creative and landing page changes during the test window unless a guardrail is breached. Otherwise, the “why” behind movement becomes unprovable.
Readout (Weekly): What to measure (and what not to over-interpret): platform ROAS is a leading indicator of delivery efficiency, not causality. The causal read is holdout lift translated into incremental profit after COGS and full costs (Kaushik; research brief).
Next test (Week 5–6): If incremental net profit ROI is negative, the fix isn’t “spend more.” It’s usually one of three levers Kaushik implicitly points to: reduce non-working costs, change tactics to increase incrementality, or fix conversion economics (Kaushik). Pick one lever per cycle.
Trade-off (say it out loud): This will reduce reported performance before it improves decision quality. It can also reduce volume in the short term, because holdouts force honesty about baseline demand.
When this is wrong: If the business is running long-horizon brand work, Kaushik notes ROI math needs a longer window and different KPIs than “six months and done.” Incremental net profit still matters, but the measurement design must match the impact horizon (Kaushik).
The cleanest way to protect a marketing budget in 2026 isn’t a prettier attribution model. It’s a profit-and-causality story that survives contact with Finance: full costs counted, baseline demand separated, and lift proven. Kaushik’s “scary giant” is only scary when it stays invisible—once it’s in the spreadsheet, it becomes something teams can manage.
