Ecommerce Explained: Business model, transaction flow, and repeat-purchase economics

Ecommerce, or electronic commerce, is the buying and selling of goods, digital products, or services over the internet, including the money and data transfers needed to complete each transaction. It covers B2C, B2B, and C2C trade through websites, marketplaces, and mobile apps rather than only online retail.

What ecommerce includes—and what it does not#

Ecommerce is the broad umbrella for online buying and selling. Online retail is the narrower B2C subset in which a seller serves an end consumer. Digital commerce is broader in another direction: it includes the pre-purchase and post-purchase journey across digital touchpoints, while ecommerce refers to the transaction itself. Paystone and BigCommerce describe those boundaries.

M-commerce is ecommerce conducted on phones or tablets through mobile sites, apps, or SMS; it is a subset of ecommerce, not a competing category. Stripe’s comparison makes the channel distinction explicit. Ecommerce is also narrower than e-business, which includes digitized operations beyond buying and selling.

Common business-model labels include B2C, B2B, C2C, C2B, D2C, and B2B2C, alongside structural choices such as marketplaces and dropshipping. Shopify’s model overview describes B2C as the most common. The labels identify who transacts with whom; they do not by themselves tell you who owns inventory, controls fulfillment, or earns profit.

Market-size and profitability figures need the same caution. Shopify’s summary of eMarketer data places ecommerce at about 20.5% of global retail sales in 2025, projects about 21.1% in 2026 and 22.5% by 2028, and notes that the share excluding China is closer to 12.8%. The headline is therefore not a universal market benchmark. Ecommerce can be profitable, but published net-margin ranges conflict—roughly 10–20% in some sources and 10–50% in others—because niche, model, and methodology differ. Bluehost’s overview is directional evidence, not a single agreed margin standard.

Ecommerce begins with the order, not the storefront#

A useful definition of ecommerce starts at the moment an order is placed. The OECD’s 2025 definition covers goods and services ordered through purpose-built systems over computer networks. Payment and delivery do not have to happen online. Web shops, apps, structured ordering features on social platforms, and electronic data interchange can qualify; a manually typed email does not qualify under this statistical definition.

That boundary matters because ecommerce is neither a synonym for digital marketing nor a single business model. A direct-to-consumer brand, a marketplace, a wholesaler receiving EDI orders, and a subscription service can all conduct ecommerce. What changes is who controls the offer, owns inventory, serves the customer, collects the money, and absorbs failure.

Three operating choices allocate control differently#

The most important model choice is not which storefront software to buy. It is where control and risk should sit. An inventory-owning merchant gives the operator more control over price, merchandising, and customer experience, but also more exposure to working capital, demand forecasting, fulfillment, and returns. A marketplace can expand selection without buying every item, but it must attract both supply and demand while governing third-party quality. A subscription can make a recurring need easier to serve, but it is an overlay on the merchant or intermediary relationship, not an automatic retention engine.

Accounting follows the actual contract, not the label placed on the website. Under IFRS 15’s principal-versus-agent framework, the central question is whether the entity controls the specified good or service before it reaches the customer. Primary fulfillment responsibility, inventory risk, and pricing discretion are indicators, but their weight depends on the arrangement.

Amazon’s 2024 Form 10-K is a concrete hybrid example. It says Amazon generally records gross revenue for items sold from its own inventory, while recording its net share from third-party seller items as service sales. It also says Amazon is not the seller of record in its seller-program transactions and earns combinations of fees, percentages of sales, per-unit fees, or interest.

One order moves through several state machines#

A checkout confirmation is not the end of the transaction. Treat the order, payment, fulfillment, return, and cash ledgers as connected but distinct records, joined by durable identifiers.

Payment authorization and capture are separate in some flows. Stripe’s PaymentIntent lifecycle can move through states requiring a payment method, confirmation, customer action, processing, capture, or success. Asynchronous methods may remain in processing before success is known.

Order operations need their own status model. Shopify’s order-status documentation separates order, payment, fulfillment, and return statuses. A voided authorization, canceled order, refunded payment, and completed return are therefore different events, even when a customer experiences them as one purchase.

Cash arrival is another boundary. Stripe defines settlement time as the interval between a payment or funding transaction and the funds becoming available in the Stripe balance. Pending funds cannot be withdrawn or spent until they become available. A payout then needs to be reconciled to the transactions it contains.

These systems answer different questions. Payment status says whether money collection progressed. Fulfillment status says whether the promise was delivered. Return and dispute records say how the outcome changed. Settlement says when processor funds became usable. Revenue recognition, tax, and legal obligations require the applicable contracts and jurisdiction; the operational state machine alone cannot decide them.

Repeat purchase matters only after variable economics#

Revenue-based retention can make a weak ecommerce model look healthy. The useful unit is contribution: the money left from an order after costs that change with that order. A fuller customer lifetime value model then connects those cohort contributions to retention and acquisition payback.

Order contribution = net order revenue - product cost - payment cost - fulfillment cost - shipping subsidy - expected return/refund cost - other variable service cost

Define every term before comparing channels or cohorts. Net order revenue should already reflect discounts and refunds under the team’s stated reporting policy. Expected return cost should include more than the refunded price when the business also pays return shipping, inspection, repackaging, write-downs, or disposal. Fixed payroll and platform overhead belong in a later profitability view; mixing them selectively into some orders makes channel comparisons unstable.

Then group customers by first purchase and give every cohort the same observation window. Shopify’s customer cohort analysis uses the date of first order to form cohorts and can display customer retention, gross sales, net sales, or average order value across later periods.

For a fixed horizon T, calculate:

Repeat purchase rate = customers with a second completed order by T / first-time customers in the cohort
Cumulative contribution per acquired customer = total realized order contribution through T / acquired customers in the cohort
Contribution after acquisition = cumulative contribution per acquired customer - customer acquisition cost

The acquisition investment has paid back within that horizon only when cumulative contribution per acquired customer reaches the acquisition cost. A high repeat rate does not guarantee this: later orders may be small, heavily discounted, expensive to ship, or frequently returned. Conversely, a lower-frequency category can still work if each repeat order contributes enough.

Three common shortcuts distort the picture. Returning-customer share can rise simply because new-customer acquisition fell. Average order value says nothing about purchase frequency or cost to serve. Revenue retention preserves neither gross margin nor acquisition payback. Keep all three as descriptive metrics, then make the decision with cohort contribution and time.

The model decision in one sentence#