The appeal is real: enormous existing traffic, customers who already trust the checkout, and fulfilment infrastructure you could not build. For a product with genuine demand, that reach is not replicable through your own site in year one.
The costs are structural rather than only financial. Referral fees plus fulfilment plus storage plus advertising commonly consume a large share of the sale price, and the businesses that struggle are the ones that priced for their own site and discovered the margin does not survive the platform. Run those numbers before listing rather than after.
More importantly, you do not own the customer. You will not generally receive their email address, you cannot market to them directly, and the relationship belongs to the marketplace. That is the trade: reach now against the ability to build something durable. A business that sells exclusively there has built revenue on an asset owned by somebody else, and account suspensions do happen with limited recourse.
The workable position for most new businesses is to treat it as one channel rather than the channel. Use it for the reach and the validation, price with the full cost stack included, and put deliberate effort into moving buyers toward a direct relationship where the terms allow it. Insert something in the package that gives them a reason to come to you next time, since that is one of the few permitted routes to a direct connection.
Whatever you do there, keep your own store running rather than treating the marketplace as your website. It costs little to maintain, it is where you can control the presentation and the margin, and it is the thing you still own if the marketplace account is suspended. Businesses that closed their direct channel because the marketplace was producing all the volume are the ones with no options when that changes.
Understand the fulfilment decision separately from the selling decision, because they have different economics. Shipping orders yourself preserves margin and consumes your time, while using the platform's fulfilment costs more per unit and buys faster delivery, which materially affects whether items sell. Model both against your actual product weight and size before choosing, since the storage component in particular punishes slow moving inventory.