If your best campaign is already maxing out the easiest demand, doubling budget usually doesn’t double qualified pipeline. It often just buys pricier impressions and clicks.
That’s not a hot take. MarTech’s Navah Hopkins puts it plainly: when there’s “no new demand to capture,” adding spend can “drive costs instead of revenue.” The part that stings is this: the campaign can still look great in-platform while the incremental dollars are quietly getting worse.
So the real question isn’t “Is this our best campaign?” It’s “Do the next dollars perform like the last dollars?”
One move for this week: run a marginal budget test with a holdout so “scale” has to earn its keep.
The problem isn’t performance. It’s the curve.
The mental model most teams use is linear: if $10k/month works, $20k/month should work better. But paid media doesn’t behave that way for long. Once you’ve captured the highest-intent inventory in a segment, incremental spend often shifts you into lower-quality auctions, weaker intent queries, or more frequency against the same people. Diminishing returns isn’t a theory; it’s the default.
And 2026 is not a “spray and pray” year for most B2B SaaS teams. In the efficiency shift documented in 2023 budget benchmarks, total marketing spend growth slowed to 14% YoY (down from 79% the prior year), and new-business acquisition’s share of budget fell from 51% to 47% while expansion ARR’s share rose to 48%. Translation: leadership wants proof that incremental dollars beat alternative uses of cash. Not vibes.
But the context gets messier. Nearly half of advertisers planned to increase PPC budgets in 2023 (49%). So scaling isn’t “wrong.” It’s just frequently done without a marginal readout.
Why “just add budget” breaks good campaigns
Two failure modes show up over and over: measurement lies and platform volatility.
On measurement: “high ROAS” can be a mirage if the campaign is harvesting demand that would’ve converted anyway. The Marketing Budget Illusion argument is simple: more budget can magnify weak strategy, unclear attribution, or poor funnel performance rather than fix it. If the handoff is messy, if the CRM is dropping lead source, if the definition of a “conversion” drifted, then increasing spend just scales the mess.
Then there’s platform behavior. Cometly warns that scaling too fast can trigger a “relearning” phase, which can raise costs and hurt performance while systems recalibrate. Hopkins makes it operational: in Microsoft Advertising, changes above roughly 15% can introduce volatility. That’s a concrete number teams can actually use as a guardrail.
Here’s the uncomfortable part: both can be true at once. A campaign can be “the best” in your dashboard and still be the worst place to put the next $5k.
The marginal ROI test (with a holdout) that settles the argument
This is the cleanest way to answer the budget question without pretending the platform dashboard is causal.
Hypothesis (make it falsifiable): If we increase budget on Campaign A by 10–15% while holding a matched slice of traffic constant, then incremental qualified pipeline will increase at an acceptable marginal CPA because we’re currently constrained by impression share (not demand saturation).
Notice what’s embedded in that statement: a mechanism. If the mechanism is wrong—if you’re not impression-share constrained—the test should fail. Good. That’s the point.
Setup
- Audience: pick the exact segment the winner is “winning” in (same geo, same ICP, same intent tier). Don’t broaden. This is a marginal test, not a new campaign.
- Holdout: reserve 10–20% of that segment as a no-change control. In practice, this can be a geo split, a matched set of accounts, or a time-based split if geo/account splitting isn’t possible. (Directional, not definitive—but better than nothing.)
- Budget change: +10–15% only. Hopkins’ Microsoft Ads note gives a pragmatic ceiling; Cometly’s “relearning” risk is the reason not to jump 2x.
- Timeline: 14 days minimum, 21 if sales cycle is longer and lead volume is low.
- Owners: Demand Gen owns the change; Marketing Ops owns the holdout definition + tracking QA; RevOps owns pipeline stage definitions for the readout.
Launch
- Freeze creative and landing page changes during the test window (creative fatigue is real, but don’t stack variables).
- Log tracking/version changes in a simple change log. If attribution shifts mid-test, the readout becomes a debate instead of a decision.
Readout
- Primary metric: incremental qualified pipeline (or incremental SQLs) from test vs holdout, normalized per 1,000 impressions or per $1,000 spend.
- Secondary metrics: marginal CPA for qualified events; impression share (or lost IS due to budget) if available.
- Guardrails: lead-to-SQL rate stays within an agreed band; no material spike in junk leads (define “junk” operationally—e.g., disqualified reason codes).
- Stop-loss: if CPA rises by >20% and qualified pipeline doesn’t increase vs holdout by the end of week one, revert to baseline and move to reallocation instead.
Next test (if it fails): don’t keep pushing spend. Shift dollars inside the channel toward segments that change unit economics. PPC benchmark summaries point to this pattern: Smart Bidding users have reported 20% more conversions and 30% lower CPA, and first-position search CTR averages around 7.11%. That’s not a promise. It’s a clue that allocation and bidding strategy can move efficiency more reliably than a blanket budget increase.
When this is wrong (yes, sometimes you should scale)
Increase budget when you can show you’re constrained by inventory, not demand. If impression share is capped, if you’re losing auctions due to budget, if you have new geos/personas ready to launch, then the marginal curve may still be healthy.
And if the business goal has explicitly shifted from efficiency to volume, that’s fine—trade-offs are real. Just say it out loud. Expecting peak ROAS at double spend is how teams lose trust with finance and then spend the next quarter “explaining” instead of building pipeline.
The loop closes back at the original ask—“double it.” The grown-up answer isn’t yes or no. It’s: show the marginal curve. If the next dollars create incremental qualified pipeline within guardrails, scale in controlled steps. If they don’t, protect the winner by not drowning it in spend.
