Your ROAS looks great.
Your bank account is flat.

A founder called me on a Tuesday. His Meta dashboard showed a 7x. Every dollar in, seven back, by the platform’s math. He had been scaling on that number for months, adding budget every week because the dashboard kept rewarding him for it.

The problem was his bank account. Flat. Revenue had barely moved in a quarter while spend had nearly doubled. The dashboard was never built to tell him the one thing he needed to know: was any of this spend actually making the business bigger.

The ad account is not the business. It reports numbers about itself, and those numbers are a proxy for performance, not the truth of it. When the account looks strong and the business does not feel it, the cause is almost never the ads. It is the measurement, the margin, or the cash.

Chasing revenue and trusting that margin will widen later is how founders end up three years on with a much bigger, much more fragile company making the same profit they always did. This playbook exists to stop that. Every chapter feeds one final calculation, and the last chapter hands you the number to set in the platform.

Everything starts here, read as a waterfall. Money lands at the top and gets cut at each ledge on the way down. The single most useful habit in this guide: when a number near the bottom is wrong, the cause is always something above it.

Your accounting software shows you net and hides everything above it. March nets $96K, April nets $110K, and you go hunting in ad spend and headcount for what changed. Pull the gross-to-net breakdown and the story flips: gross was $120K both months. Refunds were $20K in March and $10K in April. The lever was above the line your software shows.

Refunds, discounts, and under-collected shipping are profit levers sitting in that blind spot. Cut refunds a few points, tighten discount logic, collect a little more shipping, and the bottom line lifts without selling one extra unit.

Overhead should not move when revenue moves. Cost of delivery and marketing are variable; they rise with revenue, and that is fine. Fixed overhead is supposed to stay flat as you grow, which is exactly how profit widens as a percentage when you scale. The brands that stall are the ones where it climbs faster than revenue because the founder kept reinvesting into headcount and tools. In ecommerce the growth lever is marketing, not overhead.

Ask a brand their margin and they will hand you the on-paper margin — price minus factory cost — while calling it the real one. The two are not close, and the gap decides whether your ads make money.

Take a $100 t-shirt. The factory charges $30, so the brand says 70% margin. That is the on-paper margin. It ignores everything else it costs to deliver the shirt. Load all of it on:

That is a deliberately stark example. Run a kinder one: a $60 supplement at $14 cost, $7 shipping, $3 fulfilment, $2 fees, 5% returns lands near 53%, not the 77% the invoice implies. Every product has its own real number, and you cannot set a budget without it.

Picture measurement as a pyramid. The bottom is fast and unreliable. The top is slow and trustworthy. Most teams live at the bottom and make business-level decisions there, which is the whole mistake in one sentence.

Why ROAS sits at the bottom. ROAS leans entirely on attribution, and attribution is correlation dressed as causation. Summer pushes up both ice cream sales and drownings; neither causes the other. A click before a purchase looks like the cause of that purchase the same way, when usually it is the last thing that happened before a sale that was already coming.

So ROAS over-credits the bottom of the funnel, and across platforms it compounds: lifting TikTok spend can raise the reported ROAS on Meta and Google while TikTok’s own number stays flat, because TikTok plants the demand the others harvest. Read those numbers literally and you would cut the channel doing the planting.

aMER, not MER. MER is total revenue over total ad spend, and it misleads, because it folds in returning-customer revenue the ads had little to do with. aMER is new-customer revenue over ad spend, and it is the honest version. A blended MER of 2.0 can sit on top of an aMER of 1.4 that is not profitable at all.

NCAC, not CPA. The same separation, on the cost side. New-customer acquisition cost is total spend over genuinely new customers, from your order data. Platform CPA blends returning buyers in and flatters everyone. The pair to run the business on: aMER for efficiency, NCAC for price.

CAC is total ad spend over genuinely new customers. Not platform CPA, which runs on attribution and flatters you. And CAC means nothing until you pair it with the gross profit on the first order. A $120 CAC is spectacular against an $1,800 first-order profit and a disaster against a $40 one. Same number, opposite verdicts.

LTGP to CAC, the efficiency metric that survives scrutiny. Lifetime gross profit divided by CAC. Time-box it at 90 and 180 days, because unbounded LTV rises forever and justifies any spend. And use gross profit, not revenue: you can only spend the profit.

A quick anchor: with $100 of first-order gross profit, a 2.0 means a $50 CAC. Halve your real first-order gross profit and that is the CAC at the floor of the good zone.

Your break-even floor is 1 divided by your real margin. At 40% margin, that is 2.5x. Any campaign under it is losing money before overhead. One number tells your whole team where the floor is.

Now watch a discount move that floor. This is where margin disappears without anyone noticing. Costs fixed in dollars do not move when the price drops; only the price does.

A single 30% coupon pushed the required efficiency from 2.5x to 7x, almost three times harder. In CAC terms the break-even CAC fell from $40 to $10, and a $10 CAC is nearly impossible on paid. That discount cannot work for cold acquisition.

This does not mean never discount. It means run the math first, and prefer the levers that protect margin:

• • Bundling. A second unit does not double shipping and fulfilment, so margin improves while order value rises.
• • Gift with purchase. A $4-cost gift presented as a $50 value, in exchange for a second unit. Order value nearly doubles, margin percentage barely flinches.
• • Flat discounts on stuck stock only. Clearing dead inventory at break-even is fine; it was not selling anyway, and stock turned back into cash funds the next order.

Everything above collapses into one calculation. You know your real margin (chapter 3), your overhead share (chapter 2), and the net profit you want to keep. The share of revenue left for ads, and the ROAS that share demands, fall straight out:

share for ads      = real margin % − overhead % − net target %
target ROAS        = 1 ÷ share for ads
ad budget per sale = price × share for ads

Hit 5.9x and, after ad spend, overhead, returns and every other cost, you keep 12% net per order. Beat it and the extra drops straight to profit. The floor beneath it: at 44% margin the break-even floor is 2.27x. The space between 2.27x and 5.9x is your cushion.

Where it goes

Three adjustments before you type it in

Your Meta attribution setting. Your target is measured against real new-customer revenue; Meta reports an attributed number, and looser settings make Meta report more sales for the same ads. On 7-day click with no view-through the two usually sit close, which is why that window is the standard. If your account runs view-through or long windows, measure the gap (platform-reported new-customer revenue over actual, trailing 30 days) and multiply your target by it.

The prospecting split. Your target is a blend across all spend, but remarketing mostly harvests demand that was already coming. Split it: if r is your remarketing share of spend, then in the skeptical case where remarketing adds nothing truly incremental, prospecting alone must clear target ÷ (1 − r). At 20% remarketing, a 5.9x blend means prospecting carries 7.4x. Credit remarketing only what an incrementality test earns it, and keep its share capped.

The repeat relaxation. If your measured LTGP to CAC sits at 2.0 or better, the first order does not have to carry the whole target: your ad budget per sale rises to your 90-day gross profit per customer divided by 2, and the first-order ROAS requirement relaxes accordingly. Modelled retention relaxes nothing; only measured cohort data counts.

1. 01Calculate your real margin. Load every variable cost onto one order. Most teams find it sits far below the factory math, and every number in this playbook depends on it.
2. 02Compute your target ROAS and ad budget per sale, per product. Use the chapter 7 formula. Set it as your Meta cost cap. Share the break-even floor with whoever runs your ads.
3. 03Move reporting to aMER and NCAC. New-customer revenue over ad spend, and spend over genuinely new customers, on 7-day click with no view-through. Stop running the business off blended dashboards.
4. 04Separate new from returning everywhere. Prospecting and remarketing, new and returning revenue, aMER and MER. Different levers drive each; mixing them flatters the ads and hides the truth.
5. 05Re-run the math before any discount. If the offer pushes your required efficiency past anything the account has ever achieved, it is not an acquisition offer.

Every figure in the examples is invented to show the math. None are benchmarks or claims about your business. Run the logic on your own numbers.