Scale, hold, or kill: a UA decision framework with real thresholds

WHAT YOU'LL WALK AWAY WITH

Four verdicts — scale, hold, watch, kill — and the True-ROAS thresholds behind each
How to derive your own thresholds from margin and payback window instead of copying anyone else's
A 30-minute weekly pass that enforces verdicts without sentiment
Guardrails for the two hardest moments: scaling a winner, and killing a favorite

01Campaigns need verdicts, not opinions

Most UA reviews end in adjectives. "Brand-US is doing well." "Generic-IN feels soft." Adjectives don't move budget — and while a campaign sits unjudged, it keeps spending. The entire job of this framework is to make sure that every live campaign carries exactly one of four verdicts at all times, and that the verdict — not the mood — decides what happens next.

A campaign without a verdict is a subscription you forgot to cancel.

The prerequisite is honest input: per-campaign True ROAS — all revenue attributable to the campaign's users (ads, IAP, subscriptions) divided by that campaign's spend. If that number still lives in six separate dashboards, start with The True ROI Playbook; this framework runs on its output.

02The four verdicts

TRUE ROAS	VERDICT	WHAT HAPPENS THIS WEEK
≥ 4x	Scale	Budget +10–20%. Note the date. Watch efficiency at the new level.
2–4x	Hold	Budget unchanged. One improvement experiment queued — creative, bid, or audience.
1–2x	Watch	A 14-day clock starts. Improve or it drops a tier. No new budget while the clock runs.
< 1x	Kill / rebuild	Pause now. Rebuild only with a specific hypothesis about what was wrong.

Two rules keep the table honest. First, a verdict changes only on the weekly pass — not mid-week because a Tuesday looked exciting. Second, "watch" is a clock, not a category. The most expensive campaigns in any account are the ones that have been "on watch" for five months.

03Set thresholds from your economics, not ours

4x is a starting point for a portfolio that wants strong margin on net-to-you revenue. It is not a law of nature. Two inputs make the thresholds yours:

THE MATH

Breakeven True ROAS = 1.0 (when revenue is already net-to-you)

Scale threshold = breakeven × safety multiple (2–4x by risk appetite)

If you track gross revenue instead, breakeven rises to cover store fees — one more reason to keep the whole ledger net-to-you, as argued in the True ROI Playbook.

The second input is your payback window — how long a cohort takes to deliver its revenue. A hyper-casual title earns most of its money in 72 hours; judge it fast. A subscription app's cohort keeps paying for months; judging it at day 3 kills winners. Match the measurement window to the revenue curve, then apply the thresholds inside that window.

HONEST CAVEAT

A subscription app with strong retention can rationally scale at 2x, because the cohort's revenue is still compounding when you measure it. A 3-day-curve casual title cannot. Same thresholds, different windows — never copy both from someone else's genre.

04Scaling without breaking the thing that works

+10–20%

PER STEP · NEVER DOUBLE A WINNER OVERNIGHT

Efficiency decays with scale — every auction-based channel serves your cheapest conversions first. Doubling a 5x campaign's budget usually buys you a 3x campaign. The discipline:

Step, then wait. +10–20%, then a full payback window at the new level before the next step.
Watch marginal, not average. If the campaign was 5x and is 4.2x after the raise, the new spend is running well below 4.2x. Two consecutive decaying steps = stop stepping.
Feed the creative. Scaled budgets burn creative faster. A scale verdict is also a creative-pipeline verdict.

05Killing without flinching

Nobody hesitates to scale. Everybody hesitates to kill — the campaign took days to build, the audience "should" work, last month it was fine. That hesitation has a price, and it is exact: a 0.8x campaign converts 20% of everything it touches into loss, every day it runs.

Three rules remove the flinch:

The clock decides. Watch-tier campaigns get 14 days. The calendar kills them, so you don't have to.
Rebuilds need a hypothesis. "Try again with new creative" is allowed only if you can name what specifically failed. Otherwise you're paying to rerun the same experiment.
Sunk cost is already spent. The build time, the past spend — none of it comes back by continuing. Only the next dollar is a decision.

06The 30-minute weekly pass

THE WEEKLY PASS · 30 MINUTES

Pull per-campaign True ROAS for the trailing window that matches your payback curve
Assign or confirm a verdict on every live campaign — no exceptions, no "special cases"
Execute: budget steps on scales, pauses on kills, clock-checks on watches
Queue exactly one improvement experiment per hold-tier campaign
Write one line per verdict change — the why, dated. Future-you audits present-you.

Same day, same time, every week. The framework's power isn't sophistication — it's that it runs even on the weeks you're tired.

07Three traps

Weekend noise. eCPMs and conversion rates swing by weekday. Compare trailing weeks to trailing weeks, never a Saturday to a Tuesday.
Seasonality panic. Q1 eCPMs sag after Q4's ad-budget flood — every year. A January "decline" that matches last January is a season, not a verdict.
Small-sample whiplash. A campaign spending $8/day doesn't produce verdict-grade data in a week. Set a minimum spend-per-window below which the only verdicts are watch or kill.

One closing note: the thresholds in this playbook assume you can see true, all-stream ROAS per campaign. That visibility is the hard part — the verdicts are the easy part. Build the ledger once, and this entire framework runs in half an hour a week.