Two stores with identical revenue can be completely different businesses. One acquires a customer, sells to them four times, and spends nothing to do it again. The other buys every order it gets. Revenue does not distinguish them. Customer economics does.

First, a customer has to be one person

Most customer analytics quietly break before they start, because the thing being counted is not a customer. It is a phone string, or an email, or a checkout. The same shopper who once typed their number with a country code and once without becomes two customers, repeat rate falls, and the business concludes it has a retention problem it does not have.

ORVX resolves a canonical customer per store, using the storefront's own customer identifier first — the platform has already decided which orders belong to one person — and falling back to normalised phone and email. That identity is what every figure below is counted on.

The definitions, stated once

  • Delivered customer — a canonical customer with at least one delivered order.
  • Repeat customer — at least two distinct delivered orders, over their whole history with the store. Two line items in one checkout is one order, so a multi-product basket never counts as repeat purchasing.
  • Repeat rate — repeat customers ÷ delivered customers, lifetime by default. A period-scoped version is published separately and labelled, because "repeated within August" and "has ever repeated" are different questions.
  • Revenue per customer — realized revenue ÷ delivered customers, in the store's own currency.
  • Lifetime revenue — what a given customer has actually spent, across all their delivered orders.
  • Estimated CAC — ad spend ÷ delivered customers acquired.

Measured, not forecast — and it says so

"LTV" usually means a prediction: a model projecting what a customer will be worth over some horizon. ORVX does not do that, and does not pretend to. What it reports is lifetime revenue that has already happened.

That is a deliberate limit rather than a missing feature. A predicted LTV inherits every assumption in its model — retention curve, horizon, discount rate — and those assumptions are invisible by the time the number reaches a budget meeting. A measured figure is smaller, duller and true, and it is the right input for deciding what you can afford to pay for a customer today.

Concentration, and when it is suppressed

The share of revenue held by your top customers is one of the more useful risk figures a store has, and one of the easiest to render meaninglessly. ORVX suppresses it below a minimum population, because "the top customer is 100% of revenue" in a store with one customer is arithmetic, not insight.

Where this changes a decision

  • What you can pay to acquire. Estimated CAC against profit per delivered order — not against revenue per order — is the comparison that says whether growth is affordable.
  • Whether retention is real. A repeat rate computed on canonical identity is often materially higher than one computed on phone strings, and the difference has usually been read as a retention failure.
  • Concentration risk. Discovering that a fifth of revenue rests on a handful of buyers is a different conversation from an average order value chart.

What is never sent anywhere

Customer records stay in your account. They are visible only to you, every query is scoped to the account that owns the data, and no screen or export accepts another account's identifier.

The AI layer receives customer information aggregated to counts — how many were identified, how many repeated. No name, phone number or email address reaches a language model, and there is no code path by which one could.