Module 5 · Optimize: Read the Numbers, Make the Calls

Kill and scale, by rule

Back in Module 2, on a calm day, you wrote down your gates: the numbers at which you’d kill an ad, keep going, or add budget. This is the lesson where those gates go to work — and where you learn why the plan you wrote then outranks the way you feel now.

Kill-and-scale decisions are where beginners bleed money, and almost never for lack of knowledge. They bleed on nerve: killing winners out of impatience, feeding losers out of hope, and scaling so hard the campaign breaks. All three have the same cure — decide by the rule, not by the mood.

Killing an ad: the rule

Your Module 2 kill gate looks something like this: an ad dies when it has spent a defined amount and its cost per result is still above a defined multiple of the account’s.Both numbers are yours, set in your planner — the spend threshold gives an ad a fair audition (killing on $10 of data is a coin flip, not a decision), and the multiple keeps you from carrying passengers (an ad at double the cost of its siblings is taking budget the winners would use better).

Two things make this easier than it sounds. First, you launched several ads in Module 3 precisely so that none of them is precious — the auction votes, some ads lose, that’s the design working. Second, you kill ads, not the campaign. Turning off a losing ad while its siblings run is routine gardening. Killing the whole campaign is a different decision that belongs to the kill/fix gate you set in Module 2, made at your gate-review date with the diagnostic tree in hand — not at 11pm on a red Tuesday.

Scaling: the rule, and the trap

The scale gate you pre-committed reads something like: if cost per booked call holds under the allowable for two consecutive green weeks, budget rises 20%. Every word is doing work:

• Cost per booked call, not cost per lead — you scale on the number closest to money. A campaign producing cheap leads that don’t book calls hasn’t earned more budget; it’s earned a trip through the diagnostic tree.
• Two consecutive weeks, not one great one — a single green week can be noise. Consistency is the signal you’re paying for a fair read on. This is also the discipline that stops you tripling budget on six data points, which Module 2 warned was exactly how week-three excitement spends money.
• 20%, not double — the gentle raise is the whole trick, and here’s why. Meta’s learning phase — the calibration period from Module 4 — isn’t just a launch-week event. A sharp budget increase can knock the ad set right back into learning: the system recalibrates who to show ads to at the new spend level, and your results go erratic for days, sometimes worse than launch week, because now there’s spend behind the chaos. A modest raise lets the system adjust without resetting. Raise, hold, confirm the numbers stayed green, raise again. Compounding 20% steps grow a budget faster than one dramatic doubling that breaks the campaign and sends you back to week one.

The mirror-image mistakes

Watch for both failure modes in yourself, because most owners lean reliably toward one:

• The pessimist’s error: killing a campaign still inside its test window because week one looked scary. You knew it would — Module 2 called week one calibration noise. The kill/fix gate has a spend threshold for exactly this reason: no verdicts before the evidence is in.
• The optimist’s error: a loser limping along for months on hope and sunk cost — “it’s about to turn around.” Campaigns don’t turn around on their own; layers get fixed, or they don’t. If it’s been through the tree, taken its one-change-per-week fixes, and still can’t beat the gate, the plan already told you what to do. Do it.

Both errors are the same error: substituting this week’s feelings for last month’s plan. Meridian’s two attorneys — fictional as ever — don’t have to be brave at the week-three review. They just have to read their own planner out loud and do what it says. That’s what pre-commitment buys: cheap courage, purchased in advance.