Part 7 — Scaling without breaking it

Scaling is where good accounts die, because the move that doubles revenue is rarely "double the budget." Auction dynamics are non-linear: the next $10k of spend buys different (worse) auctions than the last $10k, and every scaling decision is really a bet that marginal returns hold. The discipline is to make the bet in small, measured steps and read the marginal — not average — results at each one.

The mechanics of a safe step

1. Move budgets roughly 20% per learning cycle, one variable at a time. Larger jumps reset bidding’s confidence and blur cause and effect.
2. Track marginal CAC by spend tier — what the LAST increment paid per customer. Averages hide the crossing point where growth turns unprofitable.
3. Read impression-share headroom before stepping: if you already own most of the auctions you are eligible for, more budget buys inflation, not customers. Expansion (new queries, new campaigns) is the move, not intensification.
4. Hold targets steady while stepping budget. Loosening tROAS and raising budget simultaneously is two experiments wearing one change history entry.

The expansion sequence

When intensification is exhausted, expand in order of intent: adjacent non-brand queries first, then category head terms, then PMax reach on proven winners, then genuinely new demand pools (YouTube, Demand Gen) last. Each ring converts worse than the one inside it — that is expected and priced into part 1’s payback ceiling. The operational detail, including when to split campaigns as volume grows, is in scaling Google Ads without breaking it.

What changes past $100k a month

At meaningful spend, second-order effects dominate: creative fatigue cycles matter, MER by tier becomes the primary steering instrument, incrementality questions get expensive enough to answer properly, and organizational discipline (change logs, testing tranches, quarterly economics rebasing) beats tactical cleverness. What changes past $100k a month maps that territory. If you have worked through all seven parts, you already have the foundations it assumes.