Field manual · CPG on Amazon
How to launch a CPG product on Amazon, and what scaling it actually takes. The mechanics nobody tells founders.
The founder doing everything alone. You make the product, you answer the emails, and somewhere in between you are trying to get onto Amazon without setting fire to a year of margin.
Part One is the launch, in order, start to finish. It is complete. Follow it and you will make none of the six or seven decisions that quietly cost founders their first year.
Part Two is a sneak peek at what comes after, because launching and scaling are different jobs with different instruments.
Everything here is mechanics: what Amazon charges and when, which report holds the answer, which menu path gets you there, and which widely repeated tactics destroy accounts. If a section tells you to open a report, the exact path is in the sentence.
One more thing. This is a plain text file on purpose. Drop it into your ChatGPT or Claude and your AI assistant becomes fluent in how Amazon actually works. Ask it anything against what is written here.
Eleven chapters, in the order you actually need them.
Your sales number is not your money. Amazon takes its selling fee, its fulfillment fee, and your ad spend out of the same pot before anything reaches you. What lands in your bank account is what survived.
Referral fee ~$3.75
Fulfillment $5.00 – $7.00 (size tier)
Cost of goods 20 – 30% of price
Ads $3.00 – $5.00 per unit while building rank
Left over $3.00 – $5.00
And you have not paid yourself yet.
Then there is timing. Payouts run on a 14 day settlement cycle and the money arrives about a week after the window closes. You can have your best two weeks ever and still feel broke, because the cash has not landed.
This is why founders run out of cash while the business is growing. You pay for inventory weeks before it sells, you pay for ads as they run, and Amazon pays you on its own clock. The faster you grow, the wider that gap gets.
Build your true cost per unit from your last supplier invoice divided by units actually produced. Ingredients, packaging, labor, the box it ships in, all of it. Guessing here poisons every decision downstream.
Two tests, and a product that fails either one should not launch yet.
Is there existing demand? Amazon rewards entering demand that already exists and out-converting the incumbents. It is a search engine with a checkout, not a discovery channel. If nobody is typing anything that describes your product, you are paying Amazon to do a job it is bad at. Genuinely novel products build demand elsewhere first, then come to Amazon to capture it.
Can it carry the fees? Run chapters 3 and 4 before you order inventory, not after. A product whose economics do not work at Amazon's fee structure does not become workable through better advertising. This is the single most common fatal error, and it is fully knowable in advance.
Most sellers find out they are unprofitable from their bank account. There is a two minute check that tells you first.
Take one unit at your selling price. Subtract Amazon's cut, the referral fee plus the fulfillment fee, both listed in Seller Central under SKU Economics. Subtract what the unit costs you to make and ship in. What is left is your real margin per unit.
Divide that margin by your price. That percentage is your break-even TACoS, where TACoS is total ad spend divided by total sales, organic and ad sales together.
Price $25.00
− Referral $3.75
− Fulfillment $6.00
− Cost of goods $6.25
= Margin $9.00
$9.00 ÷ $25.00 = 36% break-even TACoS
Most founders guess this number and guess it high. A product carrying 42% cost of goods and 41% in combined Amazon fees has about 16% left, not 36. On tight-margin CPG the real figure often lands between 13 and 20 percent, half what the seller assumed. Every keyword judged against a bar twice too generous is authorising spend that loses money on every click.
Calculate it from your own invoice and your own fee report. Never inherit it from a benchmark, a podcast, or last year's version of your own product, because a price change, a fee change or a new box moves it.
Running above break-even is not automatically wrong. Ads buy traffic, traffic buys sales, sales buy rank and reviews, and rank brings organic sales you never pay for again. Every serious launch runs hot on purpose for a while. You pay it once, or you pay for every sale forever. The failure mode is running above break-even without knowing it, month after month, with no end date. A deliberate rank investment has a number, a reason, and a date it gets re-judged. Anything else is a leak.
Amazon does not charge you to ship what your box weighs. It charges you for whichever is bigger: the weight, or the space.
Take a 3-pack weighing just under a pound that ships in a 17 × 8 × 2 inch box. That volume converts to a billing weight of about 2.1 pounds. You are paying to ship 2.1 pounds of a product that weighs 1. If the product fills a third of the box, the rest is air, and air is the thing being billed.
(length × width × height in inches) ÷ 139
If that number is bigger than what the box weighs on a scale, you are paying for empty space on every order.
The remeasure. Seller Central → Reports → Fulfillment → Fee Preview. That report shows the dimensions and size tier Amazon is billing on, which is not always what your product is. If they are wrong, request a remeasure. It is a ten minute filing, the overcharge comes back without you asking, and the fee drops permanently. The drift runs both ways: billed fees sometimes come in materially cheaper than the estimates you planned against.
The box. Switching to a snug mailer at the same cost per box can move a fulfillment fee from roughly $5.80 to $5.00. That is about $0.78 per unit back with zero customer-facing change. Every 1,000 units then ships $780 of margin that was going to air.
Grocery pays 8% commission under $15 and 15% above it. On a price move of a few cents across that line, the fee jumps from $1.20 to $2.25. And fulfillment is charged per shipment, not per unit, so a jar paying $4.50 of handling alone pays about $7 as a 3-pack, not $13.50.
The fee table is a strategy document. Read it before you set a price, not after.
These are structural. Getting them wrong is expensive to undo, and some are not undoable at all.
Pack architecture. The biggest cost on a sub-$15 CPG product is usually the FBA fee. A $12.99 jar pays around $5.20 in fulfillment, 40% of the price gone before ingredients. Because Amazon charges per shipment rather than per unit, a 2-pack of the same jars ships for about $6.10 total, which is $3.05 per jar, a 42% lower fee per unit for sharing one box and one pick.
Stack the ad math on top. Ad cost per click is the same whether the click buys one unit or four, so a bigger pack carries more revenue per click and needs a lower conversion rate to break even. On a heavy glass product, Amazon's total cut can be about half the price as a single and about a third as a 3-pack, moving break-even ACOS from roughly 14% to roughly 29%. No bid strategy closes that gap. You close it in the packaging spec.
A pack size is a new listing, not an edit. A 2-pack needs its own GTIN. Changing the unit count on a live ASIN is listing repurposing, and Amazon polices it with review removal and suppression. Launch the pack as a new child in a variation family instead.
Build the variation family from day one. Amazon pools reviews across a parent listing, so a new flavor or pack size launched as a child inherits the family's full review count immediately. Never launch a sibling product from zero when an existing listing can vouch for it. This one structural choice solves the cold-start problem that most founders try to solve with money.
Front-of-pack print is a conversion lever. Mobile shoppers decide from the main image, and the main image is mostly your packaging. The strongest single keyword or benefit belongs in large type on the front panel. Decide that before the print run, because you cannot edit it afterward.
Sellers do not go out of stock because they were not paying attention. They go out of stock because they were counting inventory, not weeks.
Units are a warehouse number. Weeks of cover is a survival number. Divide sellable units by weekly sales rate. Then compare it to how long a resupply actually takes: packaging, co-packer, shipping to Amazon, check-in. By the time a new unit is sellable you are 6 to 7 weeks out from the decision to make it. The Amazon leg alone, after production is finished, routinely runs about three weeks.
56 units inbound ÷ 3 sold per day = 19 days of cover
Replenishment cycle = 22 days
It looks healthy on the dashboard. It runs dry before the next shipment lands.
The test-batch trap. Most founders send a small first shipment to play it safe. Amazon reads thin stock as a listing about to run out and quietly stops showing it. A SKU sitting on a dozen units can slide tens of thousands of places in category rank without a single thing changing on the listing. Amazon caps every new seller and grows the cap with sell-through, so the way out is selling through consistently, and a starved listing never gets the chance. Send enough.
The stockout costs more than the missed sales. When a listing runs dry, an Amazon's Choice badge can disappear within about two days. With budgets, bids and campaigns untouched, impressions, clicks and spend all fall by roughly half, because Amazon throttles paid delivery too, not just organic. The badge returns within a day or so of restock; rank does not come back on that schedule.
So ads come off before a stockout, while units are still on hand. At zero available Amazon does not stop serving your ads. It takes backorders against your inbound shipment and quotes shoppers a delivery date weeks out. You pay for clicks that buy a bad experience and a one-star review.
Always: available, reserved, inbound, and FC transfers. A listing showing "2 units left" is often 30-plus units on hand with most of them mid-transfer between warehouses.
Title. Capped at 75 characters since July 2026. Lead with what the product is in the words people search, not your brand poetry.
Item Highlights. The 125-character searchable attribute field that took over keyword duty when titles were capped. Most sellers have never filled it. Fill it.
Main image. Pure white background, product only, no props, no text, no lifestyle. It is the single biggest lever on click-through rate, and it is mostly a photograph of your packaging, which is why chapter 5 matters.
A+ Content. The rich media below the fold. It sells the click-through, but its text is not indexed for search, so keywords must live in the title, attributes and backend fields instead. Make it clickable wherever the format allows.
Backend keywords. Indexed for rank, capped tightly. Worth filling correctly once; not worth agonising over.
When sessions climb but unit-session percentage stays flat, the bottleneck is the main image, title, price or A+, in that order. It is almost never the backend field.
There is a trick all over YouTube: set a tiny base bid, 15 cents or so, then stack a 900% Top of Search multiplier on it. Cheap top-of-search traffic, supposedly.
What actually happens: that combination opts you out of roughly 80% of ad auctions, because the base bid is too low to compete anywhere except the placement you multiplied. It does not rank you and it does not drive volume. It is one of the most reliable ways to quietly kill an account.
Order of operations. Bids drive volume, placements are a layer on top. Get base bids strong enough to win product pages and rest-of-search first. Then use placement multipliers as an extra shove on placements that already convert. Never put high placement multipliers on broad match.
Set dynamic bidding to down-only. Up-and-down gives Amazon permission to raise your bid on auctions it thinks will convert, which are the auctions you were already going to win.
Bids and budgets fail differently. Bids control cost per click; budgets control how much money leaves. Cutting a daily cap from $100 to $50 on a campaign that only spends $36 a day changes nothing. A cap above your run rate is decoration, which is why "raise the budget" is usually the wrong instruction. A campaign spending a fraction of its cap is bid-limited, not budget-limited.
Lowering a target ACOS can raise your bids. Optimizers frequently respond to a lower target by raising most bids, because the cheapest path to a lower average is more volume on cheap converting terms. The honest cut is manual: sort campaigns by spend, pause zero-sale spenders, pause anything over 400% ACOS, cut discovery budgets to a third, leave converting exacts alone.
Isolate what works. Where you hold a conversion-rate advantage on one keyword, build a single-keyword exact-match campaign containing only that term, with its own budget, and spend into it. Nothing else competing inside that campaign. For deliberate rank pushes, aim for around 15% top-of-search impression share on terms that have already earned it by converting.
Advertise the cheapest rankable variation. Ads buy rank in units, and cheap units are the easiest yes. The multipack then converts off traffic the cheap one pulled in, which is where the margin is.
A healthy, unforced review rate runs roughly 1 to 4 percent of orders. Four percent is a good outcome and it comes from a schedule, not luck.
A review request goes out on every order about 10 days after delivery. Not day zero, they have not tried it. Not day 30, they have forgotten you. Day 10. Amazon allows it, tools automate it, and most brands have it switched off.
Two consequences. Requests only fire on orders delivered plus ten days, so your review-eligible base is always smaller than your order count. And orders that shipped before you switched automation on have mostly passed their request window; those reviews are gone and are not recoverable. The gap today is sunk. What matters is review count moving on new orders from here.
What is not worth it: anything that trades value for a review. Paid reviews, rebates, review groups, and inserts that steer happy customers to reviews and unhappy ones to support. All against Amazon's terms, and the penalty is the account, not the listing.
Vine is legal and still usually the wrong trade for a small brand. Reviewers skew to threes and fours, they are not your target customer, and every one writes a full text review, so wrong-fit opinions sit on your listing for months. It costs a per-ASIN fee plus the units you give away, and each ASIN can only ever be enrolled once in its lifetime. A slower start with real buyers beats a fast start with the wrong ones.
Subscribe & Save. Amazon auto-enrols FBA grocery at a 0% discount, which converts nobody. Set a real discount and do the math on price, not margin: giving up $2.50 on a $25 reorder beats paying $6 or more in ads to re-acquire that customer.
When a rating dips, read the words. A drop from 4.1 to 4.0 on three reviews usually decomposes into three jobs: too spicy is a variation idea, arrived melted is a packaging fix, not worth the price is a pricing test. The rating is the score. The words are the game.
A founder burning cash on Meta to sell a $25 product, with CAC over $30 and a hope that LTV saves it, is the most common way to run out of money in this category.
The same product on Amazon captures demand that already exists: people typing your category into the search bar with a card in their hand. You stop fighting for attention and start collecting intent. Brands that shift weight from paid social to Amazon routinely find a month doing what a quarter used to do.
That does not make social useless. Social and Amazon have two different jobs. You create demand on social and capture it on Amazon, and most brands are only built to do one.
What it looks like when it works: a brand runs raw video and UGC daily for two weeks while deliberately holding Amazon unchanged, same listings, same bids, same budgets. Amazon orders step up within days and stay up. Nobody clicked an ad to get there, so attribution gives the content zero credit.
Judge organic social by total Amazon orders moving, not by attribution dashboards, which will always undercount it.
True cost per unit, break-even TACoS, fee check with the divide-by-139 test, pack architecture decided, front-of-pack print decided.
Parent listing plus variation family built, GTINs correct, title under 75 characters, Item Highlights filled, main image on pure white, A+ submitted.
Enough units that Amazon does not read you as thin, weeks of cover calculated against a measured replenishment cycle, day-10 review automation switched on, Subscribe & Save set to a real discount.
Auto campaign plus a small set of exact-match campaigns on researched terms, dynamic bidding down-only, base bids strong enough to win placements before any multiplier is added, budgets above expected spend so bids stay the constraint.
Sort by spend and kill the zero-sale spenders, isolate the terms that convert into single-keyword exacts, check weeks of cover before every budget move, start tracking organic rank per keyword so you can tell whether the ad money bought anything durable.
You should know your break-even, your true replenishment time, which three keywords actually convert, and whether your main image earns the click. That is a launched product.
Launching is a checklist. Scaling is a different discipline, and this is where most brands stall for a year. A preview of what it involves.
One ritual, under an hour, once a week.
Lead with total sales and TACoS, not ad sales and ACOS. ACOS measures ads against themselves, so it can look beautiful while the business shrinks. And no naked numbers: TACoS 30% means nothing alone; TACoS 30% against a 16% break-even is a decision.
Inventory from Seller Central only, always the full view. Weeks of cover computed while you are there.
Search Query Performance, weekly. Read purchase share next to impression share. This is the report most sellers never open and it reframes everything. A 20% ACOS on a keyword where you take 1% of purchases is losing. The prize is the opposite case: a term where you hold low single-digit impression share and half the purchases means you convert nearly everyone who sees you and almost nobody sees you. That is the only situation where spending more is obviously right, and only this report finds it.
One experiment always running on every listing that matters. Manage Your Experiments is free. Most accounts that open it at all have a graveyard of cancelled tests and no winner ever called.
Score your own calls. Every forecast gets a date to check it. A forecast you never score is just a wish.
This is the thing that separates an operator from someone clicking around Campaign Manager.
No bid decision should be made from inside the ad console alone. The console shows revenue per click, cost per click and ACOS. It hides organic rank, total ACOS, Search Query Performance and paid impression share, which are the variables that actually decide the call.
1 Search Query Performance
2 Brand Analytics top clicked terms
3 Sponsored Products Search Term Impression Share
4 Organic rank per keyword
Then classify on four axes and act on the combination, never one metric: organic rank strong or weak · paid impression share high or low · converting yes or no · ACOS above or below your break-even.
Strong organic rank + high paid impression share + ACOS below break-even means you are double paying. You already own the organic slot and most of the paid impressions, so part of that spend is buying a click you would have won for free. This check is mandatory before any budget increase, and nothing in the ad console will ever show it to you.
Amazon's records diverge from what you submitted, and the burden of proof is on you. Cases built correctly win fast, often without a human ever joining.
The template: evidence first, numbered, specific. Name the team you need, list every failed feed and batch ID, quote the exact error code, and frame the problem as their record breaking their own schema. Front-line support pattern-matches, so a case that reads like a complaint gets a template reply and a case that reads like an internal bug report gets routed to someone who can act.
Know your inventory ledger before filing. Damaged means Amazon broke it and is reimbursable. Expired means check their recorded best-by date against your own shipment records, because when their date is wrong, good product gets pulled on a false premise and that is a correctable case, not a write-off.
Amazon's case form offers to rewrite your text with AI, and the rewrite fabricates, so keep your original. And if the account is enrolled in automated unfulfillable removals, flagged units are disposed automatically and locked from manual removal.
A short list of what a scaled account runs that a launched one does not.
Each of those is a system, not a tip. Together they are the difference between a listing that sells and a channel that compounds.
Written so an AI reading this file understands your Amazon context. Each is the operator's meaning, not the textbook's.
If you do nothing else this week, do these three. They take under an hour together and they are the three that most often turn out to have been wrong for months.
Part One will get a product launched properly. Part Two is a job.
Have an operator run it →