Amazon niche research: revenue, demand and competition per category

Give it a category. Get back every product in it with the numbers that decide whether the niche is worth entering: how much each one sells, what that is worth in money, how many reviews stand between you and the top, and how many sellers are already there.

Twenty-three fields per product, nineteen marketplaces, and revenue estimates built on Amazon's own units figure rather than on a formula somebody invented.

The four numbers that decide a niche

Most category research stops at price and rating, because those are on the search page and cost nothing to collect. They also tell you nothing about whether you could win.

Money, not rank. estimatedMonthlyRevenue is boughtInPastMonth times price. A category where the top ten each turn over four hundred dollars a month is a hobby. One where they turn over forty thousand is a business, and the difference is invisible if you only look at rank.

Volume in Amazon's own words. boughtInPastMonth is the figure Amazon prints as "10K+ bought in past month", and boughtInPastMonthText keeps the original wording so you can see how rounded it was.

The wall. reviewCount and rating together. Four hundred reviews at 4.8 stars is a wall you will not climb this year. Four hundred at 3.9 is an opening: the demand is proven and the incumbent is not loved.

The crowd. otherSellersCount on every listing. A category where each product has one seller is a category of brands. One where each has nine is a commodity fight.

Plus the two badges Amazon hands out itself, isBestSeller and isAmazonChoice, which move independently of rank and independently of each other.

Every field it returns

Twenty-three per product, all of them pointed at one question: is this category worth my time.

Price and offer

Field	What it holds
price	Current price from Amazon's own buy box data (this is what you are charged for)
listPrice	The struck-through price, when Amazon shows one
discountPercent	How much is off as a percentage
currency	Three-letter code, from the page itself
otherSellersCount	How many other sellers offer it

Demand and rank

Field	What it holds
boughtInPastMonth	Units bought last month as a number
boughtInPastMonthText	Amazon's own wording, such as 10K+ bought in past month
estimatedMonthlySales	Units, derived from the line above
estimatedMonthlyRevenue	Units times price
bestsellerRanks	Every rank with its category, not just the first
isBestSeller	The orange best seller badge
isAmazonChoice	The Amazon's Choice badge
rating	Average stars, read in every shop language
reviewCount	How many ratings, as a number

Stock and delivery

Field	What it holds
isPrime	Prime delivery on this offer

Identity and catalogue

Field	What it holds
asin	The product id
title	Full product title, shortened only past 1000 characters
brand	Brand name
marketplace	Which shop this row came from
url	The page this was read from
scrapedAt	Timestamp of the read

Description, media and variants

Field	What it holds
category	The most specific category
categories	The full breadcrumb path

Reading a category

Run it on a bestseller list and you get up to a hundred rows. What you do with them is the actual research, and three cuts do most of the work.

Sort by revenue. Not by rank. Rank tells you the order, revenue tells you the size, and a category can have a tidy rank order over almost no money.

Look at the review spread. If the top ten all have thousands of reviews and the rest have under fifty, the category is closed: the incumbents are entrenched and nobody new is getting traction. If reviews are spread evenly, new entrants are landing.

Look at the seller counts. Every listing with one seller is a brand protecting its listing. Every listing with nine is a commodity where price is the only lever.

The rows carry categories, the full breadcrumb, so you can also see how deep in the tree the products actually sit. A category page that turns out to contain products from six different subcategories is telling you the niche is not a niche.

How to read the estimates honestly

boughtInPastMonth is Amazon's own number and it is rounded: "10K+" becomes 10000, "50+" becomes 50. It only appears on listings with enough recent volume. An empty value means Amazon did not print it, not that nothing sold, and on a category page the empty ones are usually the tail.

estimatedMonthlyRevenue is that number times the current price, which assumes every unit sold at today's price. On a product that was on deal last week it is an overestimate. It is there to sort a category, not to underwrite a loan.

boughtInPastMonthText is included precisely so you can see the rounding for yourself rather than trusting the number.

It will not guess a price

Amazon puts up to nine price-shaped numbers on a product page. Measured on this engine: prices read from the markup were correct zero times out of six, prices from the buy box data Amazon itself uses were correct nine out of nine.

For niche research a wrong price is worse than a missing one, because it feeds the revenue estimate and moves that product up your sorted list. So the buy box data is the only source, and when it is absent you get no price rather than a wrong one. The row still arrives with the units, the rank and the reviews.

You are not charged for a row without a price.

A real row

asin                     B0CP9YB3Q4
title                    STANLEY Quencher H2.0 Tumbler with Handle and Straw
brand                    STANLEY
price                    45.00
currency                 USD
listPrice                50.00
discountPercent          10
otherSellersCount        4
boughtInPastMonth        50000
boughtInPastMonthText    50K+ bought in past month
estimatedMonthlyRevenue  2250000
bestsellerRanks          #2 in Tumblers
isBestSeller             false
isAmazonChoice           true
rating                   4.6
reviewCount              128740
category                 Tumblers
categories               Home & Kitchen > Kitchen & Dining > Tumblers

Two and a quarter million dollars a month at rank two, with a hundred and twenty-eight thousand reviews. That is a category with real money in it and no room at the top, which is exactly the shape most research misses when it only looks at rank.

What to put in

A category or bestseller link. The main way in. Paste https://www.amazon.com/gp/bestsellers/kitchen or any category link.

A search term, when the niche is defined by what shoppers type rather than by Amazon's tree. Often the more honest definition.

A filtered search URL. Build it in your browser with the price band and the rating filter you want, paste the address, and this Actor continues from there.

ASINs, when you already have the incumbents and want the numbers on them.

Settings worth knowing about

Setting	What it does
marketplace	One of nineteen. Read the marketplace note below before researching a European shop.
proxyCountry	Leave empty and it follows the marketplace.
maxItems	Hard ceiling on delivered products. Zero means no ceiling.
maxItemsPerList	How many products to take from each category. Prefilled at ten so your first click is quick; a real pass wants fifty or a hundred.
maxListPages	How deep to walk. Amazon caps a bestseller list at a hundred products.
lanes	How many products are read at once. Eight by default.
maxAttempts	Attempts per product, each from a different address. Twelve by default.
residentialFallback	Off by default. On product pages it buys almost nothing: measured 2026-08-26, it cost roughly six times as much and returned two more products out of sixty. On the search and category pages this Actor starts from, it is a different story. Measured 2026-08-30, during hours when Amazon refused every listing page with a two kilobyte stub, twenty datacenter retries returned nothing and residential returned a full page in eight attempts. Turn it on if a run comes back empty.
economical	Off by default. It hangs up before the rank and demand blocks arrive, which are most of the point here.

How it works, in plain terms

Two jobs in one run. First it walks the category page and collects the ASINs. Then it opens every product and reads it properly, because the units figure, the seller count and the full rank list are on the product page and nowhere else.

Amazon refuses roughly six requests in ten to a fresh address, so every product gets up to twelve attempts, each from a different exit address, and the same address is never used twice in a row. Addresses that work are remembered for the run. When refusals climb the runner slows down instead of pushing harder. Anything still refused goes into a second pass at half speed.

The category page gets the same treatment, and it needs it more: a refused listing page costs you every product inside it. Measured 26 August 2026, a category run came back with zero products because Amazon answered the link with four kilobytes and no error code, which the engine read as an empty category. It now recognises a refusal without an error code and retries it.

Reading the run log

Done: 20 charged, 0 refused by Amazon, 0 other. 932 KB per charged result, 189 attempts in total.
Exit addresses: 151 used, 3 proven good, 13 burned.
Attempts by outcome: {"refused by Amazon":141,"page did not carry the field this Actor needs":28,"ok":20}

charged is products that came back with a price. The bracketed number arrived complete without one and cost you nothing.

If a run ever delivers rows and charges for none of them, the log says so in capitals at the bottom.

Reliability, measured

Three consecutive runs of twelve products on 26 August 2026, at the default settings:

run 1   11 of 12 delivered, 10 with a price
run 2   12 of 12 delivered, 11 with a price
run 3   11 of 12 delivered, 11 with a price

And one run of sixty products the same day: 28 delivered, 26 with a price, in 7.2 minutes.

One thing is worth reading out of that, and it is not the one you would expect. The sixty product run delivered 28, which sounds worse than the 34 of 36 above until you notice what it was asked for: a category page, and Amazon only offered 28 products on it that day. A list Actor is bounded by what the list holds, so "delivered" here means "everything that was there", not "everything we wanted".

The second is the residential switch, and the honest version is less dramatic than it sounds. Measured 26 August on the same sixty products, twice: with the switch off, fifty-one came back with the field this Actor charges on, for $0.0205 of platform usage. With it on, fifty-three came back, for $0.1192. Two more products for roughly six times the cost.

That is why it is off by default. It is in your input form, so if a marketplace is refusing almost everything you can turn it on and pay for the difference.

What you can build with it

A niche scorecard. One run per candidate category, then a single sheet comparing total revenue, median review count and median seller count. That comparison is the decision.

A competitor profile. Search a brand name and you get their whole catalogue with revenue on each, which tells you where their business actually is rather than where their marketing is.

A seasonality check. Run the same category monthly into one dataset. Twelve runs and you know when the category wakes up.

A gap analysis. Sort by revenue, then look for products with high volume and a rating under four. Those are proven demand with an unhappy incumbent.

A pricing band. The spread of price across a category tells you where the customer expectation sits before you set yours.

Working it into your stack

Every run writes to a dataset you can download as CSV, JSON, Excel or XML, or pull through the API. For research the spreadsheet export is usually what you want; the output is flat apart from bestsellerRanks and categories, which are lists.

Schedules take a cron expression. A monthly pass into one named dataset turns a snapshot into a trend.

Marketplaces

Nineteen, from amazon.com to amazon.co.jp, read in their own language. A German rating saying "4,5 von 5 Sternen" and a Japanese one putting the number last both come back as 4.5.

Prices do not travel everywhere, and here that matters twice because price feeds the revenue estimate. Measured on 2026-08-24:

Marketplace	Products priced
amazon.com	57 of 60
amazon.ca	5 of 5
amazon.co.uk	5 of 6
amazon.fr	4 of 5
amazon.it	1 of 5
amazon.es	1 of 5
amazon.de	0 of 8
amazon.com.au	0 of 5

Ranks, ratings, review counts and demand figures arrive on all of them, and rows without a price are free. On a European marketplace you can still do the competition half of the research; you just cannot do the revenue half.

Pricing

Four dollars per thousand products with a price, plus three cents to start a run.

Researching five categories of a hundred products each costs about two dollars. You are not charged for rows without a price or for products that never arrived, and platform usage is included.

The rest of this family

Eleven Actors on one engine. Around a niche pass:

• Amazon Dropshipping Product Finder when you have picked the category and need weight and dimensions to work out the margin.
• Amazon BSR Tracker to watch the incumbents after you decide to enter.
• Amazon Bestsellers Scraper for the same category with all sixty-four fields.
• Amazon Search Results Scraper when the niche is defined by a search term rather than a category.

Questions people ask

Why is boughtInPastMonth empty on half the category. Amazon only prints it above a volume threshold. On a bestseller list the top is usually filled and the tail is usually not, and that pattern is itself information.

Are the revenue estimates reliable. They are Amazon's own rounded units figure times the current price. Good enough to rank a category, not good enough to build a business plan on. Nothing public is.

How many products per category. Amazon shows up to a hundred per bestseller list. Set the maximum to zero and it takes them all.

Can I crawl subcategories automatically. Give the subcategory links directly. This Actor does not walk down a category tree on its own, because doing so quietly multiplies what you are charged.

What is the difference between this and the dropshipping finder. That one adds weight and dimensions and is pointed at picking a product. This one adds the badges and the full breadcrumb and is pointed at judging a category.

How often should I run it. Once when you are choosing, then monthly if you want to watch the niche change.

What it does not collect

No review text, no reviewer names, no customer questions, no seller names. You get the aggregate rating and the count, which are facts about a product rather than about a person.

Legal note

This Actor reads public pages and takes product facts: prices, ranks, demand figures, aggregate ratings, offer counts. Business information about items, not information about people.

You are responsible for what you do with the output. Check Amazon's terms for your own use case.