Marketplaces vs Your Own Store: The Fee Maths, the Data You Lose, and When to Do Both

The channel decision gets argued as ideology. Own your customer. Meet them where they are. Both slogans, neither useful when you're staring at a spreadsheet trying to work out whether Amazon is making you money.

So let's do it with the published fee schedules and a model you can actually fill in. And let's be honest about the thing that never appears on any invoice, which is the part I got wrong for years.

What each channel charges, from its own page

Amazon

Amazon UK publishes it plainly. Professional selling plan at £25 per month excluding VAT, or an Individual plan at £0.75 per item sold. Referral fees vary by category, and Amazon's own summary is that most referral fees sit between 8% and 15%^[1].

The published category detail is more interesting than the range. Computers at 7%. Books at 15%. Clothing and accessories tiered from 5% to 15% by price point. Furniture tiered 10% to 15%. Everything else 15%. There's a per-item minimum referral fee of £0.25 in most categories^[1].

Then FBA, if you use it, adds fulfilment fees by size and weight band plus monthly storage, with long-term storage surcharges on slow-moving stock^[2]. Advertising is separate again and increasingly not optional.

eBay

eBay publishes final value fees by category on its public help pages, along with the fixed per-order charge and any regulatory operating fee^[3][4]. Store subscriptions change the fee structure and the free-listing allowance, so the effective rate depends on your listing volume rather than only your sales^[5].

TikTok Shop

Here's a diagnostic worth noticing. Amazon publishes a full category fee table on a page anyone can read^[1]. eBay does the same^[3]. TikTok Shop's fee documentation sits inside the seller centre behind a login^[6].

Everything you'll read publicly about TikTok Shop commission comes from agencies and calculator tools rather than the platform, and those sources broadly converge on a commission in the region of 9% in the UK. I'd treat that as directional, not as a number to model on, because the people publishing it sell TikTok Shop services. Get your own rate from the seller centre before you commit stock.

What is not in dispute is the affiliate layer. TikTok Shop's growth engine is creator commission, typically negotiated per campaign, and it stacks on top of the platform fee. That's the cost line that surprises people.

Your own store

Platform fee plus payment processing. Shopify publishes plan pricing and transaction rates^[7]. BigCommerce does the same, with GMV thresholds that auto-upgrade your plan^[8]. WooCommerce has no platform fee and a hosting bill instead^[9].

Which looks dramatically cheaper. It isn't, and the next section is why.

The comparison nobody does properly

Marketplace fees are a channel cost. Own-store costs are a channel cost plus a customer acquisition cost, and people compare the first to the first and declare victory.

Cost line	Marketplace	Own store
Platform or referral fee	8 to 15% typical on Amazon[1]	Subscription, low or zero on Woo[7]
Payment processing	Usually inside the referral fee	Separate, roughly 1.5 to 2.9% plus fixed
Fulfilment	FBA fees, or your own[2]	Yours, plus shipping subsidy
Customer acquisition	Included in the fee	Your entire paid media budget
Advertising	On-platform ads, increasingly structural	Off-platform ads, all of it
Returns handling	Platform rules, often generous to the buyer	Your policy, your cost
Customer data	You don't get it	Yours
Repeat purchase	Platform owns the relationship	You own it, if you use it

The bolded row is the whole argument. A 15% Amazon referral fee that includes demand is not obviously worse than a 2% platform fee plus a 25% blended acquisition cost. Run your actual CAC before you feel smug about owning the channel.

And run it as a blended number, not on your best-performing campaign, because the marginal cost of the next customer is what matters and it's always higher than the average.

Does this actually impact you? Four honest cases

You make something distinctive with a brand story

Own store first, marketplace second or never. Your margin can absorb acquisition cost, and the brand is the asset. Marketplaces flatten you into a search result next to eleven near-identical products.

You resell other people's products

Marketplace first, almost always. You have no brand premium to defend and the marketplace has the demand. The uncomfortable version: if you're reselling, your own store is a nice-to-have that mostly serves customers who already found you elsewhere.

You sell high-consideration, high-ticket items

Own store, with content and reviews doing the persuading. Marketplace listings are bad at explaining anything that takes more than a photo^[23].

You sell commodity consumables

Both, and the marketplace probably wins on volume while the own store wins on repeat, subscriptions and margin. This is the profile where a genuine dual-channel strategy earns its complexity.

What we found: the operational cost of a second channel

The fee comparison is the easy part. We looked at what actually breaks when a store adds a marketplace, by reading each platform's own integration documentation.

Inventory is the first failure, every time

Two channels, one stock pool. Sell the last unit on both and somebody gets a cancellation. On Amazon that damages account health metrics, which affects Buy Box eligibility, which affects sales. The penalty for overselling on a marketplace is much higher than on your own store, and the sync is usually built for the reverse assumption.

Shopify's Admin GraphQL API and inventory model can drive this, and it's the API you'd integrate against^[10]. It's also a versioned API on a quarterly cadence, so your channel integration is a permanent maintenance line^[11]. See the ERP and CRM piece for the same trap with back-office systems.

Product data does not transfer cleanly

Each marketplace wants its own taxonomy, attributes and identifiers, which is the case for a PIM if you're running more than two channels. GTINs matter here in a way they don't on your own store, and SALT's audit of 141 product pages across 29 retailers found 65% had no GTIN at all^[12]. If that's you, marketplace onboarding is a product data project before it's a sales project.

Pricing gets policed

Marketplaces watch what you charge elsewhere. Undercutting your marketplace listing on your own site is the obvious move and it has consequences on some platforms. Model your channel pricing as one decision, not four.

Returns policies diverge

The marketplace sets the rules and they're usually more generous than yours. If you sell into the EU there's also a statutory layer on top, which we covered in the EU right of withdrawal piece. Two return policies for the same product is a support problem waiting to happen.

The asset you give up, and what it's actually worth

Marketplaces do not give you the customer. You get an order, sometimes an anonymised or platform-mediated contact, and no permission to market to them.

Here's how to price that, rather than emote about it.

Take your own-store repeat purchase rate over twelve months. Take your average order value and your contribution margin. Multiply. That's the lifetime value you forgo on every marketplace order. Now add it to the marketplace fee.

For a store with genuinely poor repeat rates, and most stores' repeat rates are worse than they think, the number is small and the marketplace looks great. For a consumables brand with a 40% repeat rate, the number dwarfs the 15% referral fee.

That's the calculation, and almost nobody runs it. I didn't run it for about six years. I argued the principle instead, which is a way of avoiding arithmetic that might disagree with you.

Platform by platform: running both without losing your mind

Shopify

The deepest native channel support of the four. Sales channels connect marketplaces to your product catalogue and orders, and Markets handles cross-border pricing and duties^[13]. What you should check is which direction inventory syncs and how fast, because a fifteen-minute poll is how you oversell on a flash sale.

WooCommerce

Plugin-based marketplace connectors, quality varies enormously, and this is the platform where I'd most strongly recommend a dedicated channel management tool over a plugin. The REST API is stable enough to integrate properly^[9], but read orders through the API rather than the database, because High-Performance Order Storage moved where orders live and older direct-query integrations are broken or about to be^[14].

Magento and Adobe Commerce

Multi-source inventory gives you real allocation across locations and channels, which is the correct foundation for multi-channel selling^[15]. The REST and GraphQL surfaces are complete^[16]. Mid-market B2B and B2C hybrids on Adobe usually already have channel management and the question is whether it's any good.

BigCommerce

Decent native channel management and clean order APIs^[17]. Fewer connector options overall. Watch the GMV thresholds that push you up a plan tier^[8].

Headless and custom

You're building the channel connector yourself, which is more work and a much better result, because you control the mapping instead of inheriting somebody's plugin. Write the product-attribute mapping down as a document rather than leaving it in code. And note that headless stores are undercounted in every platform statistic you'll read about channel adoption, because the usual detection fingerprints are gone^[24].

The channel P&L, as a table you can fill in

One row per channel. Per-order figures, not annual totals, because per-order is where the decision lives.

Line	Own store	Amazon	eBay	TikTok Shop
Average order value
Cost of goods
Platform or referral fee
Payment processing		n/a	n/a	n/a
Fulfilment and shipping
Advertising, on-channel
Advertising, off-channel		n/a	n/a	n/a
Affiliate or creator commission
Returns and refunds
Contribution per order
Repeat rate, 12 months		0	0	0
Contribution including repeat

The zeros in the repeat row are deliberate and slightly unfair, since some marketplace buyers do come back to you. Treat them as zero until you can prove otherwise with data, because assuming otherwise is how the channel gets over-credited.

The tax and compliance layer, which differs by channel

Marketplace facilitator rules do some of the work

In most US states and increasingly elsewhere, the marketplace collects and remits sales tax on your behalf. That's genuinely helpful and it creates a specific trap: marketplace sales can still count toward your own economic nexus thresholds in some states, even though somebody else collected the tax on them. We worked through that in the sales tax automation piece, and it's the single most common reason a merchant discovers a registration obligation they didn't know about.

Your own store gives you the whole obligation

No facilitator. You register, you collect, you file. Which is a cost of owning the channel that never shows up in a fee comparison, and which scales with the number of jurisdictions rather than with revenue.

Cross-border adds a third layer

Selling into the EU brings import VAT, customs data and the statutory right of withdrawal, and the marketplace handles some of that for you and none of it on your own storefront. See the international selling piece for the mechanics and the right of withdrawal piece for the returns obligation specifically.

Returns are the sharpest divergence

Marketplaces set buyer-friendly return rules you don't control, and the cost lands with you. On your own store you set the policy inside the statutory floor. Two policies for the same product is a support problem and a margin problem at once, and the returns piece covers how to model the cost properly.

What AI agents change about the channel decision

This is the genuinely new bit, and it cuts against marketplaces for once.

Marketplaces exist because discovery is hard and they aggregate demand. When an assistant does the discovery, it can read structured product data from anywhere, including your own store, provided your data is legible. Google's Universal Commerce Protocol work and the Agentic Commerce Protocol maintained by OpenAI and Stripe both describe merchant product and checkout data directly^[18][19].

In principle that's the first structural argument for owning your storefront in about fifteen years.

In practice, hold it loosely. Shopify reported AI-referred orders growing roughly 13 times year on year in Q1 2026, which is fast^[20]. Peer-reviewed work by Kaiser and Schulze across 973 sites and $20 billion of revenue puts ChatGPT at under 0.2% of ecommerce traffic^[21]. Growing fast from very small.

What to do about it costs nothing and helps regardless: get GTINs on your products, get structured data valid, stop blocking crawlers you didn't mean to block. SALT found 15% of audited retailers blocked AI crawlers entirely, most likely without deciding to^[12]. Adobe found product pages score around 66% on machine readability, worst of any page type^[22].

What to do this week

• Fill in the channel P&L table. Per order, per channel, contribution not revenue.
• Get your blended CAC. Total paid media divided by new customers. Compare directly to the Amazon referral fee for your category^[1].
• Calculate your own-store 12-month repeat rate. That's what a marketplace order costs you beyond the fee.
• Check GTIN coverage across your catalogue. Missing identifiers block marketplaces and AI surfaces alike^[12].
• Find out how your inventory syncs and how often. Webhook or poll. If nobody knows, that's the finding.
• If you sell on TikTok Shop, pull your actual fee from the seller centre rather than trusting a blog figure^[6].
• Check your robots.txt for AI crawler blocks you didn't intend^[12].

The takeaway

Amazon publishes its fees and they're 8 to 15% for most categories, plus fulfilment, plus advertising that's no longer really optional^[1][2]. Your own store's fee looks smaller and comes with an acquisition bill the marketplace fee already covers.

The decision isn't ideological. It's whether your repeat rate is good enough to make owning the customer worth the cost of finding them.

Most stores' repeat rates are worse than the founder believes. Mine were. That's not a reason to abandon your own store, but it is a reason to fix the repeat rate before arguing about channel philosophy, and the email and SMS piece is where that work actually happens.

What's your 12-month repeat rate? Not your gut feel. The number.

Sources

1. Amazon, "Selling on Amazon: pricing" (UK). Professional plan £25/month excl. VAT; Individual £0.75/item; most referral fees 8 to 15%; £0.25 per-item minimum in most categories. Accessed 22 July 2026.
2. Amazon, "Fulfilment by Amazon". Accessed 22 July 2026.
3. eBay, "Selling fees" (UK). Accessed 22 July 2026.
4. eBay, "Seller fees" (US). Accessed 22 July 2026.
5. eBay, "Store selling fees" (UK). Accessed 22 July 2026.
6. TikTok Shop, Seller University (UK). Fee documentation sits inside the seller centre; no comparable public fee schedule. Accessed 22 July 2026.
7. Shopify, "Pricing". Accessed 22 July 2026.
8. BigCommerce, "Pricing". Accessed 22 July 2026.
9. WooCommerce, "REST API". Accessed 22 July 2026.
10. Shopify, "Admin GraphQL API". Accessed 22 July 2026.
11. Shopify, "API versioning". Accessed 22 July 2026.
12. SALT.agency, "Agentic Commerce Protocol PDP analysis". 141 PDPs, 29 retailers; 65% no GTIN, 15% blocking AI crawlers. Agency-published.
13. Shopify, "Shopify Markets". Accessed 22 July 2026.
14. WooCommerce, "High-Performance Order Storage". Accessed 22 July 2026.
15. Adobe, "Inventory Management introduction". Accessed 22 July 2026.
16. Adobe, "Adobe Commerce REST API". Accessed 22 July 2026.
17. BigCommerce, "Orders API". Accessed 22 July 2026.
18. Google, "Under the Hood: Universal Commerce Protocol (UCP)". Google Developers Blog.
19. Agentic Commerce Protocol, specification repository. Maintained by OpenAI and Stripe.
20. Shopify, financial reports, Q1 2026. AI-referred orders up roughly 13x YoY.
21. Maximilian Kaiser and Christian Schulze, "ChatGPT Referrals to E-Commerce Websites", Marketing Science. 973 sites, $20B revenue. Peer-reviewed.
22. Adobe, "AI traffic surge: retail sites not machine readable". Adobe Analytics, over 1 trillion visits. Vendor-published.
23. Marketplace Pulse, Amazon marketplace research. Accessed 22 July 2026.
24. HTTP Archive, "Web Almanac 2025: Ecommerce". Accessed 22 July 2026.