Optimizing budget: cut, test, scale — Shift

Optimizing isn't "looking", it's deciding

A running campaign produces numbers every day. Most entrepreneurs contemplate them without acting, or react to the slightest variation. Optimizing means turning that data into repeatable decisions: what do I cut, what do I keep, what do I amplify? And above all, on which threshold, set in advance, do I decide. Without a written rule, you steer on mood — and mood is expensive.

:::key[Key takeaway] Profitability doesn't come from the perfect campaign on the first try, but from the discipline of cutting losers fast and patiently reinvesting in winners. :::

The two numbers that drive everything: target CPA and ROAS

Before optimizing, you need to know what a good result is for you. Two benchmarks:

Target CPA: how much you can pay to acquire a customer and stay profitable. It's derived from your margin and the value of a customer over time.
ROAS: how many euros of revenue each euro of ad spend brings in. A ROAS of 3 means €3 earned for €1 spent — whether your margin supports it is another matter.

These two numbers can't be guessed: they're calculated from your offer. A "€25 per customer" campaign is excellent for someone who earns €400, catastrophic for someone who earns €30.

:::warning[Trap: targeting a ROAS without knowing your margin] A ROAS of 4 can be loss-making if your margin is thin and your overhead high. Calculate your acceptable CPA from real margin, not from a number seen online. :::

The discipline of testing: isolate one variable

You improve a campaign like you run an experiment: by changing one thing at a time. Testing a new audience, a new creative and a new budget at once means making it impossible to know which produced the result. You test creatives against each other at constant audience, then audiences at constant creative. Platforms offer built-in A/B tests (split tests) that split traffic cleanly — use them rather than comparing two campaigns launched at different times.

:::tip[Tip: let the test reach significance] A test stopped too early "because one version is leading" is often wrong: at small volumes, chance mimics performance. Wait for a sufficient number of conversions per variant before declaring a winner. :::

Cutting losers: the rule that saves budgets

Most ads and audiences won't work — that's normal and expected. The skill isn't avoiding losers, but cutting them fast, on a defined threshold: "if after €X spent and exiting the learning phase, the CPA exceeds my threshold, I cut". Writing this rule before launching avoids the double trap: cutting a winner out of impatience, or feeding a loser out of hope.

graph TD
    A[Campaign out of learning] --> B{CPA below threshold?}
    B -->|Yes| C[Keep and consider scaling]
    B -->|No| D{Creative or audience improvable?}
    D -->|Yes| E[Iterate one variable]
    D -->|No| F[Cut]

Scaling without breaking everything

A profitable campaign makes you want to triple the budget at once. That's the best way to send it back into learning and blow up the CPA. Scaling is done methodically:

Vertical scaling: increase the budget in moderate steps (often +20 to +30% every few days), to let the algorithm readjust without shock.
Horizontal scaling: duplicate what works to new audiences, new placements or a new platform, rather than betting everything on one ad set.

:::info[Benchmark: creative fatigue is real] A winning ad wears out: as the same audience sees it repeatedly, its click-through rate drops and its cost rises. Watch the display frequency and prepare new creatives before performance falls, not after. :::

Read the right numbers, ignore the rest

Ad dashboards overflow with metrics. Most are vanity metrics: impressions, reach, "likes". They flatter but don't pay the bills. You steer on the chain that leads to money: impressions → clicks (CTR) → conversions (conversion rate) → cost per customer (CPA) → return (ROAS). When a result disappoints, you walk back up this chain to find the weak link: few clicks = creative problem; many clicks but few sales = page or offer problem.

:::example[Concrete case] A shop has an excellent CTR but a CPA three times too high. The creative does its job (people click), so the problem is downstream: the landing page is slow on mobile and the ad's promo code doesn't appear there. She fixes the page, keeps the creative — the CPA falls into target without touching the budget. :::

Key takeaways

:::key[The chapter's key points]

Calculate your target CPA and viable ROAS from your real margin.
Test one variable at a time and let the test reach sufficient volume.
Write your cut rule in advance to yield to neither impatience nor hope.
Scale in steps (vertical) and by duplicating what works (horizontal), not all at once.
Anticipate creative fatigue: renew ads before the drop.
Steer on the CTR → conversion → CPA → ROAS chain, not on vanity metrics. :::