Amazon product research: the four questions that decide a product

What research is actually for

Product research has one job: to stop you spending three thousand dollars finding out something you could have known for free. Everything else — the dashboards, the score out of ten, the trend lines — is in service of that or it is decoration.

The reason this needs saying is that the tooling has drifted away from the decision. Open most research software and you get a filterable table of every product on Amazon, sortable by a dozen columns. It feels like research. What it mostly produces is a shortlist of products that clear arbitrary thresholds, with no argument about whether you specifically can win any of them.

A real decision needs four things to be true at once. Demand has to exist. You have to be able to get in front of it. The unit economics have to survive Amazon's cut. And the whole picture has to still be roughly true by the time your inventory lands. Fail any one and the other three do not rescue you.

Question one: is anyone actually buying this?

This is the easiest question and the one most likely to be answered with the wrong evidence. Search volume is not demand. It tells you how many people typed a phrase, not how many of them opened a wallet.

What counts as evidence

The most honest demand signal available to you before launch is the sales rank of the products already serving the keyword, read across the whole first page rather than at the top. One product with a strong rank means one product is selling. Ten products with respectable ranks means a market exists that can support more than one winner — which is the situation you actually want.

Review counts work as a slower version of the same signal. Reviews accumulate roughly in proportion to units sold over a product's lifetime, so a page where the top listings have thousands of reviews each is telling you about years of consistent sales, not a spike.

Reading both signals across twenty listings is the tedious part. An Amazon product demand checker does that pass for you.

What does not count

• Google search volume. People research on Google and buy on Amazon, but the two populations overlap less than you would hope, and the intent is different.
• Social virality. A product doing numbers on TikTok may have demand that evaporates in six weeks. That is a different business with different inventory risk.
• Your own enthusiasm. The most expensive research bias is having already decided.

If you want the manual version of this check, the validation checklist walks the same ground step by step.

Question two: can you realistically get in front of it?

This is the question that decides most launches, and it is the one the metrics-table approach handles worst. Competition is not a number. It is a specific set of listings you would have to outrank, each with its own weaknesses.

The useful framing is not how strong the page is on average. It is whether there is a seam. A first page of ten listings with four thousand reviews each and professional photography has no seam. A first page where six listings are strong but four are clearly neglected — stock photos, a title stuffed with keywords from 2019, reviews complaining about the same defect for two years — has one.

What a seam looks like

Reading a first page this way takes about twenty minutes per keyword by hand, which is why almost nobody does it across enough candidates to matter. It is also the specific thing SellRadar automates — the scan reads the top 20 organic listings and names which of them you could plausibly beat, rather than returning a competition score you have to interpret.

Two related reads: spotting weak listings covers the manual technique in more depth, and checking niche saturation covers the opposite failure, where the page is uniformly strong.

Question three: does anything survive the fees?

A product can have real demand and a beatable first page and still be a bad business, because Amazon takes its cut before you take yours. This is arithmetic rather than judgement, which makes it the one part of research nobody should get wrong — and the part most commonly skipped, because it requires knowing your landed cost and people want to pick the product first.

Work backwards from the price the first page has established. You do not get to choose your price in a mature category; the page has already set the band a customer will accept. Subtract the referral fee, the fulfilment fee, your landed unit cost including freight and duty, and a realistic advertising cost per unit for the first six months. What is left is the number that matters.

Amazon's own FBA revenue calculator gives you the fee side authoritatively and costs nothing, which makes it strictly better than any third-party estimate of the same numbers. The profitability walkthrough goes through a full example if you want the mechanics.

Referral rates, return rates and typical price bands all vary by category, so the same landed cost produces very different margins depending on where a product sits. We have broken down the economics of the categories sellers ask about most — home and kitchen, beauty and pet supplies among them — because the category-level picture is what tells you whether a thin margin is normal there or a warning.

Question four: will this still be true in six months?

Research is a photograph of a market that keeps moving. Your inventory will not arrive for weeks or months, and the page you validated is not the page you will compete on. This question asks how fast the photograph is aging.

Three things that age a niche quickly

1. A seasonal peak you mistook for a trend. Checking demand in November on a product that sells in November tells you very little about March. Look at whether the category has a shape across the year before treating a number as a baseline.
2. A gap that is obvious to everyone. If the seam you found is a stock photo on an otherwise strong listing, assume three other people found it this week. Easily copied advantages are priced in by the time you ship.
3. A brand that has noticed. A listing that has just been refreshed — new images, new A+ content, a revised title — is a seller who is paying attention again. That is a different opponent from the one whose numbers you read.

None of these are reasons not to launch. They are reasons to prefer an advantage that takes effort to copy — a genuinely better product, a supplier relationship, a bundle that needs tooling — over one that takes an afternoon.

Running this without spending your life on it

Done by hand, the four questions take thirty to forty minutes per candidate, and a realistic search involves dozens of candidates. That arithmetic is the entire reason research tools exist, and also why so many of them optimise for volume over judgement: filtering ten thousand products is easy to build and easy to demo, while arguing about one product is neither.

A sequence that does not waste time

1. Generate candidates however you like — browsing, supplier catalogues, a filtered database, an idea you had in a shop. The source matters far less than the filter.
2. Kill on demand first. It is the cheapest check and removes most of the list in seconds.
3. Kill on economics second, using a rough landed cost. You do not need a quote yet, just an order of magnitude.
4. Spend real time on competition only for what survives. This is the expensive judgement call and it deserves the attention the first two steps saved you.
5. Sanity-check durability before you commit money, not after the sample arrives.

If you are starting from nothing, the beginner's walkthrough covers the same ground more slowly and assumes less. If you are choosing software, the tool comparison is honest about where each option is genuinely better than us.

And if you would rather see the four questions answered for a specific keyword before you decide whether any of this is worth doing manually, that is what the free tier is for. Three products a month, no card, full analysis — the free tier explainer covers what is and is not included.

Frequently asked questions