18 min read

Before you pay for the click: What drives the cost of selling on marketplaces

Creating a product image and description that meets marketplace requirements has never been so easy or so cheap. What remains difficult, and even critical, is choosing the version that will best convince the buyer.

What drives the cost of selling on marketplaces such as Amazon and Allegro

Even years of seller experience lose out here to the perception of customers, who browse offers faster and faster, usually on a phone screen. A fraction of a second of extra attention decides which description gets read and which gets skipped. The ability to capture that attention also shapes the cost of selling on Amazon or Allegro, an ever-larger part of which is advertising fees. Under these conditions, managing product sales on marketplaces has become extremely demanding and frustrating.

How selling on marketplaces has changed

Marketplaces as the main channel to reach buyers

A marketplace is a place where thousands of independent sellers offer products side by side. Customers find them through a shared search engine. A buyer types a query and compares offers from many companies on one screen, looking at images, prices, reviews, and delivery times. For many brands, marketplaces are the main channel to reach buyers, and staying off them would mean giving up a significant share of a market made up of customers who have chosen convenient shopping and delivery.

Amazon shows this best. According to EMARKETER, around 40% of US e-commerce goes through Amazon (EMARKETER, 2026). Third-party sellers, that is, companies selling through the platform, account for 61% of units sold there. Amazon’s revenue from services for these sellers grew 16% in the second quarter of 2026, to $46.8 billion (Amazon, July 30, 2026).

The scale in Europe is just as impressive. Amazon’s European stores have an average of 181.3 million users a month (European Commission, April 29, 2026, via Cross-Border Magazine). In Poland, Allegro plays a similar role. The platform has 15.6 million active buyers, its gross merchandise value is growing 12% year on year (Allegro, Q2 2026 results), and 5.6 million people use it every day (Gemius, “E-commerce w Polsce 2026”).

The growing position of marketplaces affects the strategies of online sellers. When choosing between investing in their own stores and selling on marketplaces, they have to consider the reach their own store cannot provide. The problems are commissions and advertising costs, and the fact that the relationship with the buyer and the data about them stay with the platform.

The cost of visibility: advertising on Amazon and Allegro

The way marketplaces make money is changing. Sales on marketplaces are growing, but at the same time sellers’ spending on presenting their offer so that it has a chance to be noticed is becoming increasingly painful. In the second quarter of 2026, Amazon’s advertising revenue grew 26% to $19.8 billion, and its online store sales grew 15% (Amazon, July 30, 2026). On Allegro in Poland, advertising revenue grew 31.1%, while gross merchandise value grew 12% (Allegro, September 17, 2026). On both platforms, advertising is growing much faster than sales, as Chart 1 shows. Marketplaces are earning more and more from what sellers pay just to get buyers to see their offer.

Bar chart: in Q2 2026, advertising revenue grew year over year by 26% at Amazon and by 31.1% at Allegro, while sales grew by 15% and 12%.
Chart 1. Year-on-year growth in advertising revenue and sales, Q2 2026. Amazon: online store sales; Allegro: gross merchandise value on the platform in Poland. Sources: Amazon, Allegro.eu.

Marketplace advertising works like an auction, as it does in popular ad platforms. Sellers bid for placements in search results, and the price of a click depends on competition among them. On Amazon, most of the screen positions that convert best into sales are taken by ads, so a seller who wants to be noticed is, in a sense, forced to use this tool (Marketplace Pulse, 2023). The more companies bid on the same keywords, the more expensive each click becomes. According to data from the Ad Badger tool, the average cost per click on Amazon ads was $1.22 in 2026, $0.10 more than a year earlier (Ad Badger, August 26, 2026). As a result, advertising is no longer an add-on that a seller can drop in a weaker month. It has become a fixed line in the cost of sales, like commissions or logistics, only much harder to plan.

For sellers, the cost of marketplace advertising is now a fixed part of the bill, like commissions or logistics.

On top of that comes a new layer: AI shopping assistants, such as Rufus on Amazon, which recommend products to buyers and increasingly show ads as well. Product content today has to convince both a person and the algorithm that reads it.

Who goes on marketplaces and who stays off them

Despite the growing position of marketplaces, many companies still do not use them or treat this channel as an experiment. According to the latest Eurostat data (for 2024), 45% of EU companies selling online use marketplaces, and almost two-thirds do in Poland. Yet marketplaces bring EU companies only 1.3% of their turnover, more than five times less than their own online stores (Eurostat, via Cross-Border Magazine). Many companies and brands are present on marketplaces, but few make money there.

The reasons are well known: a lack of skills and capacity to run marketplaces in-house, the high cost of intermediaries, and high commissions. Fees and the costs of advertising, logistics, and content preparation add up to a bill that discourages companies before they even start.

Michał Grabka, founder of HeroSwipe, who has worked in online sales for more than 20 years, points to a mistake made by many companies entering marketplaces such as Allegro or Amazon. They look mainly at the number of orders and at the difference between the purchase price and the selling price. A product bought for €10 and sold for €30 looks like a good deal until you deduct VAT, the platform commission, and the costs of logistics, advertising, customer service, complaints, and overheads. After a few months, it turns out that despite thousands of orders the company is not making money, and sometimes it loses money on every sale. The biggest barrier is reaching and maintaining profitability while unplanned expenses keep growing. Many companies give up when they realize that growing sales do not mean growing profits.

Those who survive and grow on marketplaces manage sales methodically. The market is clearly concentrating: fewer than 8,000 sellers now generate half of all third-party sales on Amazon in the US, compared with the 15,000 it took in 2023 (Marketplace Pulse, February 12, 2026). For a brand with a catalog of hundreds of products, scale is an advantage, provided that decisions about each product page follow an effective, repeatable method. Without one, launching hundreds of products means hundreds of separate gut-feel decisions that take a great deal of time and energy.

What sellers are up against

Margins under pressure

The demands and stress that sellers talk about show up in their financial results. In a Marketplace Pulse survey, 38% of them describe their situation as a crisis and 23% as good. Forty-seven percent saw their margins fall year on year. They name platform fees (49%) and advertising spend (46%) as the main sources of pressure (Marketplace Pulse, Seller Index 2026).

This trend has been going on for years. As early as 2022, a typical seller was giving Amazon more than half of their revenue in commissions, fulfillment fees, and advertising, compared with around 40% five years earlier (Marketplace Pulse, 2023). In 2026, it is mainly the structure of this bill that is changing. Amazon’s fulfillment fees in the US rose by an average of $0.08 per unit (Amazon Seller Central), while in Europe they fell by an average of about €0.17 (Amazon announcement, via Brand Monkey). Meanwhile advertising spend is growing faster than sales. On Allegro, since March 2, 2026, sellers have been paying more toward deliveries in the Allegro Smart! program (Wirtualnemedia.pl, 2026). A growing share of the cost of selling therefore depends on how effectively an offer turns clicks into purchases.

Advertising is also the hardest line in the budget to plan. Experienced sellers know that platform commissions and the costs of logistics, customer service, and overheads can be estimated in advance and built into a product’s profitability calculation. Advertising costs, however, are variable and hard to predict, and they can eat up a large part of the margin, sometimes all of it.

Product images drive purchases

For 61% of shoppers, images and video are the most important element of a product page, more important than price and availability (60%) and other shoppers’ reviews (57%). One in three abandons a purchase when images are poor or missing (Salsify, 2026).

The main image (hero image), the first one a buyer sees in search results, plays a special role. On a phone screen it is the size of a postage stamp and sits next to many similar thumbnails, yet this is the image buyers use to decide whether to click at all. Research shows that we judge the visual appeal of a web page in about 50 milliseconds. People tend to confirm their first impression, so an offer that makes a poor start rarely gets a second chance (Lindgaard et al., 2006).

The image also affects what happens after the purchase. Nearly half of shoppers have returned a product that did not match how it was presented (Salsify, 2026). A good main image therefore has to grab attention and show the product accurately at the same time, because every return means extra cost and the risk of a worse review.

Scale multiplies decisions and costs

The weight of decisions about how products look on marketplaces is best shown with a practical example. A seller offering 100 products in five countries runs 500 product pages, and on each one, it has to choose the main image, additional images, title, and description in several languages. According to one agency’s price list, optimizing a product listing on Amazon US costs from about $1,000 to $3,000 (My Amazon Guy, 2026). Outsourcing it for all 500 pages at those rates would cost between $500,000 and $1.5 million for a single round.

Similar amounts appear in Europe. Agency offers we know of show that preparing standard A+ content, Amazon’s enhanced product presentation, costs at least €1,000 per product. Amazon also requires a separate, approved A+ version for each marketplace and each language (WisePIM). For a hundred products, A+ content alone costs at least around €100,000 before it is adapted to further markets. Calculations like these show the scale of spending sellers take on with no guarantee that the chosen variants will turn out to be the best.

Each European Amazon store runs a separate product page in the local language, and buyers in different countries may react differently to the same images. Then there are other marketplaces such as Allegro, each with its own requirements and its own competition. A company selling 100 products on five marketplaces in four countries manages up to 2,000 versions of its listings. Nor does optimization end with a single round. Content has to be updated and re-checked regularly, taking into account seasonality, holiday periods, changing trends, and competitors’ moves.

In practice, several people discuss a single main image: the marketplace manager, a designer, someone from marketing, and sometimes an agency and the management board. Each has their own arguments, and none has data on how buyers will react. The decision is made when the time for discussion runs out, and a few weeks later it comes back with the next product or the next market.

The second bill is paid every month, and it is measured by ACoS. Let’s assume the same company generates €1 million in monthly sales from advertising at an ACoS of 25%, so it spends €250,000 on ads. If better-chosen main images lowered ACoS by 5 percentage points, the company would save €50,000 a month, or €600,000 a year.

The cost of a mistake adds up

A weaker product image costs money for as long as it is in use. Every paid click that does not end in a purchase raises ACoS, and as ad rates rise, the loss grows month after month. Without a test, the seller does not know they are incurring it.

This cost rarely shows up in reports as a separate line. It is spread across a higher ACoS, a lower conversion rate (the share of visitors who buy), and more returns. Each of these changes looks minor on its own, but together they decide whether a product makes money. A mistake costs the most when launching a new product. The seller then pays for advertising to build the first sales and the first reviews, and a weak image lowers the return on every euro spent at exactly the moment when the product has no history yet.

From a practitioner’s experience

For more than 20 years, Michał Grabka was responsible for online sales of his own products and those of other brands, completing more than 4.5 million transactions in that time. For about three years before starting the HeroSwipe project, he built his own brand, DIBYS, with 30 products sold in 15 markets in Europe and the US, on marketplaces such as Amazon, Allegro, OTTO, Kaufland, Walmart, Cdiscount, and bol.com. In its second year, the brand passed 100,000 orders.

“In more than 20 years in e-commerce, I have learned one thing: even the most experienced marketing team cannot always predict what buyers will choose. We saw this for ourselves while building the DIBYS brand. When we started testing images with real users, it turned out that their choices often differed from ours. Simply replacing the main image allowed us, in the most spectacular cases, to lower ACoS by as much as a third. That is when I realized that instead of spending money on advertising, it is worth first testing what works on a sample of potential customers.”

— Michał Grabka, founder of HeroSwipe

Why existing methods fall short

A/B tests on the marketplace

Amazon lets sellers compare two content variants on a live product page: half of shoppers see one, and half see the other. According to Amazon, refining content based on such tests can increase sales by up to 25% (Amazon, Brand Registry). The method has clear limitations, however. It requires a brand registered in Amazon’s program and a product with high traffic, and a test should run for 8 to 10 weeks (Amazon, Manage Your Experiments). In the mode that ends a test once statistical significance is reached, results can be available after about four weeks. A new product with no traffic yet cannot be tested this way. A product with traffic shows the weaker variant to half of shoppers for the entire test, and the seller pays for those clicks. The result arrives only after the advertising budget has been spent. On top of that, each market needs a separate test.

Generative AI and simulations

AI has sped up content creation. In 2025, sellers on Amazon used it to create more than 12 million product listings (Amazon, April 30, 2026). Simple use of language models and image generators has clear weaknesses, though. AI writes correctly and quickly but reaches for the most typical phrasing, so product descriptions start to sound alike and stand out less. It can also attribute features to a product that it does not have, and every such inaccuracy risks complaints and the listing being rejected by the marketplace. Image generators create attractive visuals but do not always show the real product faithfully. A gap between the image and what the customer receives increases the risk of returns.

The most important limitation lies elsewhere, however: AI will create ten versions of an image or a description, but it will not tell you which one will work best with buyers. Simulations in which AI models play the role of buyers do not fill this gap. They are fast and cheap, but they reproduce behavior from the data they were trained on, while consumer attitudes and behavior keep changing.

Agencies and team experience

An experienced team or agency knows the market well, but its time is expensive and limited. The price lists cited above show what it costs to work through an entire catalog, and knowledge gained on one product rarely carries over to the next. Each category and each market also has its own preferences: an image that sells in Germany will not necessarily work in Poland. Experience helps rule out weak variants, but only buyers can decide between two good ones.

Sales reports

Sales and advertising reports show precisely what happened, but they are produced once the budget has been spent and the weaker variant has collected its clicks. Sellers need this knowledge earlier, before they publish content.

All of these methods share one weakness. A marketplace test and a report give an answer only after the budget has been spent, while team experience, a survey, or an AI simulation give it earlier but do not reproduce a buyer’s real choice.

The missing step: validation before publishing

Between creating content and publishing it, there is a missing step that we call validation. It means checking variants with a sample of real buyers before any of them goes live on the marketplace. This step should recreate a shopping situation rather than a survey situation and give an answer faster than a test on live listings. It should also work the same way for one product as for an entire catalog. The solution is to combine scientific knowledge of how people make decisions with technology that makes it possible to run tests at scale. Tests should be designed to capture the first impression in conditions as close as possible to a real shopping situation, because it often determines the behavior that leads to a purchase. This approach lies at the heart of the HeroSwipe project.

First impressions often decide a purchase. That is why it pays to test an offer with real buyers before we pay for the first click.

How a choice test works in HeroSwipe

The test involves real people matched to the category of the product being tested. A respondent sees two variants on the screen, for example two main images of the same product, in a layout resembling search results. They choose one with a swipe, just as they would when browsing offers, and then explain their choice in a few short questions. Many such decisions build a picture of buyer preferences. The seller learns which variant buyers chose and, from the answers to the follow-up questions, also what is worth improving in the product presentation.

Two variants of the main image of the DIBYS Around the World spice set, labeled A and B, compared in a HeroSwipe choice test.

HeroSwipe’s potential

The method developed at HeroSwipe grew out of hands-on marketplace selling. Hundreds of examples showed that, after implementing the variants chosen by testers, sales of the tested products rose and advertising costs fell. Hero A/B validates images and creative concepts with real buyers before a brand commits budget to them. Hero Content+ helps create and optimize content for an entire product catalog. The choices are made by HeroTesters, members of a panel in a mobile app. Test results will be available within as little as 24 hours, although this depends on the number and profile of HeroTesters required. The goal is to cut this time to a few hours. The test takes place before the listing is published and does not require buying advertising traffic on the marketplace. As a result, the decision about the main image is based on buyers’ choices rather than guesswork, which reduces the uncertainty sellers face today.

Sources

  1. EMARKETER, FAQ on Amazon advertising, 2026: https://www.emarketer.com/content/faq-on-amazon-advertising--retail-media-dominance--prime-video-scale--agentic-ad-tools
  2. Amazon, Q2 2026 results, July 30, 2026: https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Second-Quarter-Results/
  3. Cross-Border Magazine, Marketplace Usage Across the EU, May 8, 2026 (Eurostat data for 2024 and the European Commission’s list of very large online platforms of April 29, 2026): https://cross-border-magazine.com/marketplace-usage-across-the-eu/
  4. Allegro.eu, Report and Financial Statements Q2 2026, September 17, 2026: https://about.allegro.eu/wp-content/uploads/2026/09/Report-and-Financial-Statements-Q2-2026.pdf
  5. Gemius, PBI, IAB Polska, report “E-commerce w Polsce 2026” (E-commerce in Poland 2026): https://gemius.com/pl/news/raport-e-commerce-w-polsce-2026-jest-juz-dostepny/
  6. Marketplace Pulse, Amazon Takes a 50% Cut of Sellers’ Revenue, 2023: https://www.marketplacepulse.com/articles/amazon-takes-a-50-cut-of-sellers-revenue
  7. Ad Badger, Amazon Advertising Benchmarks 2026 (data from the tool provider), August 26, 2026: https://www.adbadger.com/blog/amazon-advertising-stats/
  8. Marketplace Pulse, Top 1.6% of Sellers Drive 50% of Amazon’s 3P GMV, February 12, 2026: https://www.marketplacepulse.com/articles/top-16-of-sellers-drive-50-of-amazons-3p-gmv
  9. Marketplace Pulse, The Marketplace Pulse Seller Index Results (2026), April 2, 2026: https://www.marketplacepulse.com/articles/the-marketplace-pulse-seller-index-results-2026
  10. Amazon Seller Central, 2026 Updates to US Referral and FBA Fees: https://sellercentral.amazon.com/seller-forums/discussions/t/f3fa3211-820b-4e2e-a023-158a9cf55f99
  11. Brand Monkey, Amazon FBA and Referral Fees Structure for Jan 2026 (Amazon announcement for Europe, December 2025): https://brandmonkey.co/amazon-fba-and-referral-fees-structure-announced-for-jan-2026/
  12. Wirtualnemedia.pl, Drożeje Allegro Smart! Ale nie dla kupujących (Allegro Smart! gets more expensive, but not for buyers), 2026: https://www.wirtualnemedia.pl/allegro-smart-dostawy-zakupow-z-allegro-nowe-ceny-od-marca-2026-roku-dla-sprzedawcow,7249050210408768a
  13. Salsify, Why Images and Videos Are Indispensable Product Page Elements (Consumer Research 2026): https://www.salsify.com/blog/images-and-videos-most-important-product-page-elements
  14. Lindgaard G., Fernandes G., Dudek C., Brown J., Attention web designers: You have 50 milliseconds to make a good first impression!, Behaviour & Information Technology 25(2), 2006: https://doi.org/10.1080/01449290500330448
  15. My Amazon Guy, Choosing Amazon Listing Optimization Services in 2026: https://myamazonguy.com/optimization/choosing-amazon-listing-optimization-services/
  16. WisePIM, Amazon A+ Content: guide (separate A+ version required for each marketplace and language): https://wisepim.com/de/guides/amazon-a-plus-content
  17. Amazon, Brand Registry: https://sell.amazon.com/brand-registry
  18. Amazon, Manage Your Experiments: https://sell.amazon.com/tools/manage-your-experiments
  19. Amazon, 2025 Small Business Empowerment Report, April 30, 2026: https://www.aboutamazon.com/news/small-business/amazon-2025-small-business-empowerment-report