What e-commerce is
E-commerce is the sale of goods and services through online channels. The definition is simple, and
behind it sit markedly different business models with different economics, metrics and technology.
| Model | Who sells to whom | Key characteristic | What to watch first |
|---|---|---|---|
| B2C | Store to end consumer | Mass traffic, short cycle | Conversion, order value, repeats |
| B2B | Seller to a business | Negotiated prices, long cycle, repeat ordering | Repeat share, volume per account |
| D2C | Manufacturer direct to consumer | Full control of brand and data | Acquisition cost, lifetime value, retention |
| Marketplace | A platform matching sellers and buyers | Platform traffic, in-listing competition | Ranking position, commission, rating |
| Subscription | Recurring charges for recurring supply | Predictable revenue | Churn, renewal, lifetime value |
One company often operates several models at once: an owned site (D2C), a marketplace presence and a
wholesale arm (B2B). The metrics cannot simply be summed — acquisition cost and margin differ by
multiples between channels.
The stack and where responsibility ends
The stack is assembled from components, each with its own zone. Confusion about the boundaries
produces either duplicate purchases or gaps in the process.
| Component | What it owns | What not to expect from it |
|---|---|---|
| Engine / CMS | Catalogue, cart, orders, content pages | Advanced personalization and ML ranking |
| Search | Query handling, synonyms, typos, facets | Selecting products with no explicit query |
| Personalization and recommendations | What to show, where | Delivering messages to inbox, push or SMS |
| Web analytics | Traffic sources, behaviour, reporting | Managing content and merchandising |
| CDP / CRM | One customer profile, segments, history | Rendering the storefront |
| Email service | Delivery, templates, unsubscribes | Building the on-site behavioural profile |
| Payments | Accepting payment, refunds, anti-fraud | Pricing logic |
| Logistics and WMS | Stock, delivery, returns | Storefront logic |
Two boundaries blur most often.
Personalization and messaging. A personalization platform decides what content to show and can
hand a ready product block to be embedded in an email. The transport itself — sending, unsubscribe
handling, domain reputation — belongs to the email service provider. They are different systems with
different service levels and different legal responsibilities.
Analytics and personalization. A personalization platform’s reports show the effect of its own
scenarios. They do not replace web analytics and do not claim to be the single source of truth on
traffic and revenue: the systems use different session models and attribution windows, so a
discrepancy is normal rather than a defect.
A separate direction is headless commerce, where the storefront is decoupled from the backend and
talks to it through APIs. That stack gives freedom on the frontend and simplifies supporting several
channels, while moving responsibility for SEO, speed and page assembly to the development team.
The funnel and the metrics
Online store economics decompose into a multiplicative chain. An improvement in any link multiplies
through the rest — which is why unglamorous conversion points are worth real money.
Traffic × CR = Orders
Orders × AOV = GMV
GMV × margin = Gross profit
Gross profit − acquisition cost = Contribution
The long-run check: LTV / CAC ≥ 3
| Metric | What it shows | Where it usually leaks |
|---|---|---|
| Traffic | Audience volume | Budget growth with no quality growth |
| Conversion rate | Share of visits with an order | The listing and the product page |
| Average order value | Basket size | No upsell, no delivery thresholds |
| Gross merchandise value | Total sales volume | Rises while margin falls — watch with profit |
| Lifetime value | Revenue across a customer’s life | No retention, revenue is all first-transaction |
| Acquisition cost | Cost of winning a customer | Computed in aggregate, not per channel |
The losses in the conversion funnel are unevenly distributed. On the
listing page the audience that did not find a relevant product in the first screens
drops out. On the product page it is those who lacked information, photography or
clarity on delivery. At checkout it is those stopped by mandatory
registration, an unexpected delivery cost or a missing payment method.
Before optimising conversion, compute the absolute losses in units and money at every step. A step
converting at 40% with a hundred thousand visits is almost always worth more than one converting at
8% with five thousand, even though the second looks worse.
How online economics differ from offline
- Traffic is bought explicitly. In physical retail a large share of the flow comes from
location, and its cost is spread across rent. Online, acquisition cost is a visible line item. - Every step is measurable. Which makes experiments possible: any storefront change can be
compared against a control group and valued in money. - The assortment is not limited by shelf space. A catalogue can hold hundreds of thousands of
items while the screen still holds a dozen. Hence the critical role of search, navigation and
selection. - Returns cost more. The shopper does not touch the product before delivery, and return rates in
some categories reach tens of percent, which belongs in the unit economics. - Retention is separate work. Offline a repeat visit can follow from a route; online every other
browser tab competes for it.
A checklist for assessing a store
- Are unit economics computed per channel, rather than averaged across all sources?
- Is the funnel decomposed by step with absolute losses, not only percentages?
- Is the catalogue current: stock, prices, attributes, product page
quality — before any storefront experiment? - Is there an experiment culture: changes verified against a control group rather than judged
month over month? - Are system boundaries clear: who selects content, who delivers messages, who owns the revenue
source of truth? - Is repeat purchase share measured — without it lifetime value is an estimate, not a fact?
- Is mobile traffic examined separately: in most projects it carries the majority of visits and
a materially lower conversion rate.