Media buying 101
Buy ads
without guessing
Nine steps, always in this order. Nothing is judged without a ceiling, nothing is scaled without permission, and every rule names the number that triggered it.
Nine steps,
always in this order
- 1Know your ceilingWhat a lead or order may cost
- 2StructureFour lanes, no leaks
- 3SettingsSet once, then hands off
- 4MetricsWhat each number tells you
- 5Kill rulesWhen to stop an ad
- 6Scale rulesDuplicate, then +20%
- 7Creative cadenceNew ads every week
- 8NamingOne convention
- 9RhythmMachine daily, human twice a week
Result = a lead
A first contact that becomes a booked call, a held call and a customer. Valued in what a customer is worth.
Result = a purchase
A new customer as your store counts it, not the pixel. Valued in what each order contributes after costs.
Both applies to every business. Lead gen and E-commerce mark where the rules split.
Know your ceiling
Work out once a month the most a lead or order may cost. Every rule after this compares against that number.
Lead gen From customer value to max cost per lead
Max CAC = customer lifetime value ÷ the LTV:CAC ratio you require (3, 6 or 9)
Max CPL = Max CAC × booking rate × show rate × close rate
Example: a customer is worth €3,000 and you require 3:1, so a customer may cost €1,000. If 50% book, 70% show and 30% close, a lead may cost €1,000 × 0.105 = €105. Blended CPL is total spend ÷ total leads, never the average of channels.
E-commerce From order value to break-even ROAS
Contribution = order value − product cost − payment fees − net shipping − returns
Break-even ROAS = order value ÷ contribution = 100 ÷ 65 = 1.54
Keep returning customers under 15–20% of conversions. Above that, the cost per purchase you see is no longer the cost of a new customer.
Both Permission to scale
Scale nothing. Fix the economics or tracking first.
Testing is fine. No graduating, no scaling.
Graduating and scaling are open.
On screen: green at or under the ceiling, amber under 1.5×, red at 1.5× or more. Each channel gets its own funnel, all with the same Max CAC.
Structure
Four lanes, each its own campaign, so budget can't leak from one job into another.
Graduate = duplicate the winner. The original keeps running.
E-commerce Build up by monthly budget
| Monthly budget | Prospecting | Scaling | Retargeting | Retention |
|---|---|---|---|---|
| Under €3k | 90–95%, all broad | — | — | 5–10%, customer list |
| €3k–10k | Broad | Top creatives, duplicated | — | Customer list |
| €10k–30k | Broad + one interest per ad set | Yes | — | Customer list |
| €30k+ | More prospecting | Yes | Add-to-carts and visitors | 180-day buyers + email list |
This ladder is practice, not published research. 20–30% of broad already reaches warm audiences (Jon Loomer), so separate retargeting is optional below €30k.
Settings: set them once
Meta learns from stable conditions. Every significant edit restarts that learning.
Resets learning
- Targeting
- Creative, or adding an ad
- Optimisation goal
- Bid strategy
- A pause of 7+ days
- A big budget jump (€100 → €1,000)
Safe
- Small budget steps (€100 → €101)
- Letting it run
Lead gen Leads objective, optimised for conversion leads (booked calls) where you have the volume. E-commerce Sales objective, campaign budget on for prospecting, daily budget at least one order value, highest volume bid.
Both Maximise conversions. Dynamic creative off. One interest per ad set. Don't touch budget or schedule after launch.
Both Judge on 7-day click. 1-day view only as a second column. Budget follows 7-day click results.
Meta needs about 50 results per ad set per week. Lead gen Below that, "learning limited" is normal, not a fault. E-commerce Daily budget around 7× your target cost per purchase.
Metrics: what each number tells you
Only cost per result decides a kill or a scale. Everything else tells you where to look.
Cost per result
Lead gen: CPL. E-commerce: cost per purchase and ROAS.
Link CTR
Hook rate
Frequency + CTR
Frequency rises and CTR falls: rotate creative. High frequency alone is fine.
Cost per landing page view
Over 2× the account median: the page loses people before the form or cart.
CPM
A diagnosis only. Read it with CTR.
Read CPM together with CTR
Meta's auction price is bid × estimated action rate × quality. Switching off on high CPM or CPC demonstrably turns off top performers (Barry Hott).
Kill rules
Per ad set and per ad, top to bottom. The first rule that matches wins. Only active ads with a ceiling and delivery yesterday are judged.
- Link CTR above 1% in 7 days, but 0 resultsDon't kill: fix the page or form
- 3,000+ impressions, CTR under 0.5% and landing page views cost over 2× the account medianStop: early kill
- Spent 2× (tight budget) or 3× (funded) the ceiling with 0 results, after 5–7 days and a Friday–SaturdayStop: no results
- 7 days in a row at 1.5× the ceiling or more per resultStop: too expensive
- 5+ results in 7 days at 70% of the ceiling or lessGraduate to scaling
- Cost per result above the ceiling over 7 daysWatch
- None of the aboveOK: let it run
What never happens
- Both Stopping the last active ad in an ad set or campaign. Stopping a graduated original or anything in retargeting: at most "watch". Reactivating automatically.
- E-commerce Killing on add-to-cart. The unit is the purchase. A full ceiling spent with no add-to-cart means: watch, and look at the product page.
Expect about 70% of new creatives to be switched off within two weeks, and 10–20% to hit the target (Curtis Howland). That is normal. Common Thread Collective found per-ad stop-losses have no measurable effect on account results; we keep them to cap spend when signal is thin, and make them deliberately slow.
Scale rules
Horizontal: duplicate winners into scaling. Vertical: raise budget in small steps, behind seven gates.
Graduate (horizontal)
Pausing a working ad to "move" it costs results. Meta's own bar is about 50 results a week; 5 in 7 days is a compromise for low-signal accounts. Ask first: could this ad spend €10,000?
Raise budget (vertical), in scaling only
- 15+ results in 7 daysat or below the ceiling
- 2Nothing on stopand nothing paused today
- 380%+ of the weekly budget spentnot per day: Meta may spend 175% on one day
- 424 hours since the last stepone step per day
- 5Budget as you set ita human changed it? Wait 7 days
- 6Inside the monthly ceilingotherwise raise the ceiling first
- 7E-commerce: ROAS on targetand your store data agrees
- +20%at most +50% in total, inside the ceiling
Why +20% and not more: jumps above 30–40% a day raise cost per result 35–60% for 48–72 hours, and doubling raises it 80–120% for 5–7 days (RocketShip HQ). Expect ROAS to drop 20–30% when spend doubles.
E-commerce +5–15% on days that return more, via Meta's budget scheduling. Never by switching ads off on weak days.
E-commerce An alternative to budget steps once a campaign does 50+ purchases a week. Lead gen Not yet: volume is below the learning threshold.
Lead gen Alert after €20 spent without a lead, then every €10. E-commerce Revenue over total ad spend below break-even for two weeks: cut, starting with creative and funnel.
Creative cadence
15–20% of budget goes to testing. About 70% of new ads build on winners, 30% try new angles. Winners run long: statics over 100 days are normal.
70% iterations on winners
30% new angles
| Monthly budget | New creatives a week average → top 25% | Concepts a month |
|---|---|---|
| Under €10k | 2.8 → 4.8 | 8–12 |
| €10k–50k | 4.1 → 8.1 | 15–25 |
| €50k–200k | 6.7 → 16 | 25–40+ |
Lead gen 3+ new ads a week, in one batch. E-commerce 4+ a week; expect fatigue after 10–14 days.
4–9% of launches become winners. Finance and business services sit at the low end.
From reviews, Reddit and support mail: desires, objections, failed alternatives, benefits, pain points. Find the 3–5 main objections.
Recognition, data, confrontation, pain point, question, statement.
Video: founder story, dream state, the "ugly" unscripted ad. Static: one big line on the image, one point. Vertical Reels-first runs 30–50% cheaper CPM than Feed.
Naming and UTMs
Freeze names when you publish. Link spend to results with the platform ID (utm_id, campaign_id), never with the name.
Campaign {yyyy-mm}-{niche}-{objective}-{audience}, ad set Broad_Pack_W{week}. One fixed UTM string on every ad.
Campaign {goal}_{offer}_{market}_{yyyy-mm}, ad set {audience}_{placement}, ad {format}_{concept}_{version}.
Rhythm and automation
A machine checks every day. A human reviews twice a week, and is the only one who changes a ceiling, creates a scaling campaign, reactivates anything or touches retargeting.
Lead gen Low booking rate: lead quality or the form. Low show rate: follow-up and reminders. Low close rate: sales, fit or offer.
E-commerce Returning orders count for revenue, not for acquisition.
Both Before acting, compare your CRM or store with the ad platforms. Meta, Google and your own data should roughly agree.
Start with your ceiling
Every rule on this page needs one number first. The calculator works it out from your funnel in a few minutes.
Calculate your ceilingSources
- Meta Business Help Center: learning phase, significant edits, daily budgets, budget scheduling, creative fatigue, the auction
- Meta for Developers: conversion leads integration (Dec 2025)
- Common Thread Collective: Meta methodology (Jun 2026), cost controls study of $200M across 253 accounts (Aug 2026)
- Motion: creative benchmarks 2026, 578k creatives (Apr 2026)
- Foxwell Digital: creative testing frameworks (Mar 2026)
- RocketShip HQ: budget scaling rules (Sep 2026)
- Jon Loomer: 19 rules, budget increases, CPM and performance, creative fatigue
- Barry Hott: CPMs and CPCs don't matter
- Ben Heath, Curtis Howland: when to kill an ad; $150M spent on Meta (2026)
Practice, not confirmed by a reputable source: a 3× CPL default for lead gen, frequency cut-offs, add-to-cart kill rules and the €3k/10k/30k structure ladder.