Run Amazon as a P&L, Not a Sales Channel

I joined Eitan Koter on the Commerce Untold podcast in August 2026 for about fifty minutes on something I say often enough that people are probably tired of hearing it from me: Amazon is a profit and loss statement, not a sales channel.

Most people think they know Amazon. I do not mean the professionals. I mean brand owners, marketing managers and e-commerce directors, and I say it with respect because I understand why. They know their revenue. They do not know their numbers.

What does it mean to run Amazon as a P&L?

Here is the conversation I have had more times than any other. We take over an account, revenue is climbing, and I get on a call feeling good about it. The client is not. Their bookkeeper says Amazon is not making money.

I cannot prove or disprove that. Not until the numbers are accurate.

On Amazon the money leaves in more places than most teams track. There is the referral commission simply for selling there. There are FBA fees. There is storage, which quietly becomes expensive in Q4. There are returns. There is advertising. Then add tariffs and whatever else the year has decided to introduce. Every one of those comes out of the same margin, and the brand usually sees them as separate line items rather than as one number that tells them whether the channel is worth running.

The timing matters as much as the categories. If you are waiting for your bookkeeper to reconcile last month, you are usually looking at data that is one to two months behind. I am making decisions weekly and daily. By the time a two month old report tells you that you lost money, you have lost it twice and you are now chasing it.

You do not need it reconciled to the dollar. You need it close enough to act on this week. For me it has never been about perfection, it is about getting as close to accurate as we can and then building a path forward from there. That principle sits underneath everything in our Amazon brand management work.

What happens when you stop chasing the metric you have always chased?

This is the part of the episode I would point at if you only have five minutes.

The brand does about $10M a year across Amazon, direct to consumer, Meta, Google and email. Mostly e-commerce, not retail. Profitable. Also completely plateaued for about a year and a half. I was brought in to grow it, which is a different job from maintaining it.

The account was roughly a fifty fifty split between branded search and terms aimed at new customers. It had been built the way most good accounts were built over the last few years: tighten everything to exact match, negate aggressively, do not waste a cent on generic searches. That approach works. It also produces exactly the plateau they were sitting in, because an account optimized that hard has nowhere left to go.

What most agencies do at that point is protect the numbers. Maintain ROAS, maintain ACoS, maintain margin. That is management, not growth, and over-managing is precisely how an account flattens out.

So I did the thing the dashboard would have argued against. I backed off branded search and I opened the account up, adding broad and a wider selection of terms.

The reason is that search itself has changed. Between Rufus and AI-assisted shopping, people are not typing three-word fragments any more. They are asking full questions the way they would in Google, things like what is the best machine for 2026. An account fine-tuned to exact match cannot be found by a question it was never told to expect.

The numbers, at the same spend

Metric	Before	After	Direction
Impressions	baseline	1.2M more	Up
Cost per click	about $3.85 to $4.00	about $0.70	Down
Cost per acquisition	about $40	about $11	Down
Conversion rate	higher	lower	Worse
ACoS	higher	lower	Better
ROAS	higher	lower	Worse

Read that table honestly. Three of those moved in the wrong direction. Conversion rate fell, because we were showing to far more people and a wider audience always converts at a lower rate. ROAS fell too.

And the cost to acquire a customer dropped by roughly three quarters, at the same budget.

If your reporting is built to defend conversion rate and ROAS, this test looks like a failure and you turn it off in week two. The only reason we could run it at all is that the numbers underneath were accurate enough to show what was really happening. If you want the grounding for reading these reports properly, we covered it in how to read your Amazon PPC reports, and the structural side sits in our Amazon PPC work.

People get stuck on the one metric they have been improving and forget to look anywhere else. Deciding which metric matters right now, rather than which one mattered three years ago, is most of the job.

Why is a small brand with great margins better than a bigger one without?

Two real examples from the same portfolio.

One brand does a little under $7,000 a month on Amazon. Profit is close to $3,000 of it. That is a slow grower with excellent margins.

Another does $35,000 a month at about a 5% margin.

The second one looks five times better in any revenue conversation and makes less money. It is also far more fragile, because a fee change, a tariff, or one bad quarter of storage fees can push a 5% margin under water.

This is the era we are in. After 2020 far fewer of the brands I talk to are chasing top line at any cost. The ones with venture or private equity behind them still do, and those are usually not the brands I work with. Everyone else has moved to margin, and they will accept a lower year over year number to protect it.

My own version of that test is simple. If a brand is making three times what they pay me, in profit, they are not going anywhere. If I am not paying for myself twice or three times over, they will eventually start looking around, and they should.

Why did I build the software instead of buying it?

I have paid for a great deal of software over the years. Advertising automation in one tool, inventory in another, profitability in a third, and none of them talk to each other. You might have profitability for Amazon but not for Walmart, or not for TikTok Shop, or not for your warehouse. As an agency my software costs got high, and the last couple of years have been about running lean enough to stay healthy.

What I wanted did not exist: one system that thinks about advertising, inventory and profitability together instead of in silos. No brand I talk to has this, and that includes very large ones. Most are in a spreadsheet or waiting on a reconciliation.

The design question I kept asking was, if I had all the time in the world, what would I build? The honest answer is that I would pull every report from every platform, put them in one place, and look at all of it. Nobody has that time, so most teams do that exercise once a quarter, or once a year if they are lucky.

A concrete example of what it does: it auto-tags products. In a normal account you might have ten products that have been running for a year and two you are launching this quarter. Those two should not be judged by the same rules. Low stock is a different situation from growth mode, which is different again from clearing inventory. One set of rules applied across all of them will give you confident, wrong answers.

The goal was never to sell it. It was to make my account managers better informed, so that an advertising change is not made on conversion rate and ROAS alone. Before a bid goes up, the question is what the margin on that product is and how much stock is left. There is no sense doubling down on ads for something that is about to sell out.

Worth being straight about how it got built. Before AI I was leaning on offshore developers part time, whenever I could afford it or find the time to hand work over. Now it is me, working on it obsessively when I can, with Claude Code doing the coding. That is the real change AI made for me. It is also exactly the caution I raised on a panel a week later, which we wrote up in what five operators actually run: an agent is only as good as the context you give it.

How should a brand pick which channel to launch on?

It starts with the category, because the category decides how much firepower you need.

We usually launch on Amazon first, even for a brand that wants to be direct to consumer, because traffic will find its way to Amazon whether you plan for it or not and you would rather it find you than a competitor.

After that it depends entirely on saturation. Take a pet brand coming from Ireland into the United States, which is a real recent example. Established internationally, strong infrastructure, and entering a category that is both enormous and crowded. Amazon alone will not do it. That one needs Meta traffic, Google traffic, a site that actually informs people who have never heard of the brand, and eventually TikTok and affiliates. Not all at once, but on the roadmap.

Other categories are unsexy and barely contested, and there are still plenty of them. In those, a heavy Meta or TikTok strategy is not what decides the outcome.

One channel we deliberately do not start with is Chewy, even though it performs well. Pricing is hard to adjust there, and they look at your other channels to decide how much to order. Going in early with the rest of your presence unsettled works against you. The multi-channel thinking behind all of this sits in our multi-marketplace work.

What actually builds a brand right now?

Brand owners ask me a version of this constantly, usually phrased as a complaint that everything is expensive now. Ten years ago you could put $10,000 into Facebook and be fairly confident of what came back. That is over.

If someone still wants to operate as though it were ten years ago, we are not a good fit, and I would rather find that out early than halfway through a contract.

What I tell the rest is that viral moments are good for brands but they do not build them. What builds a brand in a tight market is consistency, incremental growth, and knowing why you got the sales you got. Knowing your margins. Knowing which keywords actually drive revenue. Building real relationships with a few influencers instead of buying reach.

And treating the whole thing as a test. E-commerce is a testing environment by definition. It is always changing, which is what makes it good and also what makes it exhausting. Standard operating procedures are useful right up to the point where they stop being true, and then you have to be willing to throw the playbook out and rebuild it. I have tried something every day for a year and had it work on day 366.

Frequently asked questions

What does it mean to run Amazon as a P&L instead of a sales channel?

It means judging Amazon on profit rather than revenue. Referral commissions, FBA fees, storage, returns, advertising and tariffs all come out of the same margin. Until those are accurate you cannot tell whether the channel makes money, and a bookkeeper reconciling two months later is describing money you have already lost.

Why did cost per acquisition fall while ROAS and conversion rate got worse?

Because they measure different things. Opening the account to broader terms showed the ads to far more people, so a lower share of them converted and ROAS fell. The absolute cost of winning a customer still dropped from about $40 to about $11 at the same spend, because clicks got much cheaper and volume rose.

Should Amazon sellers move away from exact match keywords?

Not entirely, but an account tuned only to exact match cannot be found by a shopper asking a full question. With Rufus and AI-assisted shopping, people search in sentences. Leaving room for broader terms is how you get discovered by queries you never thought to add.

Is higher revenue on Amazon better than higher margin?

Usually not. A product doing under $7,000 a month with close to $3,000 of profit is a better asset than one doing $35,000 a month at a 5% margin, because the thin margin cannot absorb a fee change, a tariff or a bad quarter of storage fees.

How should a brand judge whether an Amazon agency is worth the fee?

Measure the profit the relationship produces against what you pay. The bar Marknology holds itself to is three times the fee in profit. Below roughly twice the fee, the brand is right to start looking elsewhere.

What to take from this

1. Get your Amazon numbers accurate before you set a strategy. Not reconciled to the dollar, just accurate enough that you can act on them this week.
2. Write down which metric you are actually optimizing right now, and why. If the answer is "the one we have always watched", that is the plateau talking.
3. Check whether your account is tuned so tightly it cannot be found by a question. Conversational search is where the growth moved.
4. Compare your SKUs by profit, not revenue. The $7,000 product may be beating the $35,000 one.
5. Judge your partners on multiples of their fee in profit. Three times is the bar I hold myself to.

Thanks to Eitan Koter and the Commerce Untold team for the conversation. If you are looking at your own numbers and cannot tell which SKUs are carrying the account, that is the exact problem worth solving first.