Ecommerce platforms return on investment: a framework

A sales increase after a platform change is not proof that the platform caused it. For readers assessing ecommerce platforms return on investment, a defensible calculation needs net contribution, the full economic cost of change and a credible account of what would have happened without the change. For a like-for-like ecommerce platform comparison, apply the same definitions and reporting period to the current platform and proposed change. If any of those is weak, report an observed result or attribution instead of ROI.

No merchant outcome was tested for this article. The equations below are calculation templates, not a tested return; validate their inputs and accounting and tax treatment before relying on a result.

What to take away

Define three evidence levels

Observed orders are recorded customer events during a fixed period. Specify accepted or fulfilled status, customer population, timezone, cancellations, returns and duplicates. They show activity only.

Attributed outcomes are assigned by a documented rule, such as a tagged session or matched campaign identifier. Record the lookback period, missing identifiers, cross-device limits and the share left unattributed. An attribution label does not show causation.

Causal incrementality is the difference between what happened with the intervention and a credible counterfactual. HM Treasury's updated Magenta Book is written for government evaluation, not retail benchmarking, but it usefully separates outcome monitoring from attribution and stresses early evaluation design. Its quality guidance explains why a comparable unaffected group or period matters.

Calculate contribution before return

Choose one reporting period and reconcile cash settlements to order records. Do not use gross checkout value as benefit.

For a platform comparison, define the proposed change precisely—for example, a move between Shopify, BigCommerce and WooCommerce, a checkout replacement or an integration change. Test platform-related benefits such as checkout completion, contribution per settled order, order-processing labour, integration failures and support contacts; do not credit the platform with changes the evidence cannot isolate. Use the same order population, period and definitions for the current and proposed platform.

observed net contribution = settled receipts - refunds - disputes - payment deductions - cost of goods - fulfilment - variable support cost

Refunds need the actual provider treatment. Shopify states that the original card processing fee is not reimbursed on a Shopify Payments refund in its current supplier record. Other providers and configurations need their own evidence.

Next build the denominator:

full economic change cost = subscriptions + payments setup + implementation + internal labour + migration + integrations + assurance + incidents + parallel running + exit

These lines vary by operating model. A hosted SaaS platform such as Shopify or BigCommerce usually includes core hosting in its subscription, while a WooCommerce build may have separately purchased hosting, extensions and maintenance; compare actual contracts and invoices rather than assuming either model’s total cost. Include separately billed theme or plugin licences, headless commerce API and engineering costs, and PCI DSS assurance where applicable. Populate the comparison with actual GBP amounts from contracts, invoices and processor statements, and treat VAT consistently under the merchant’s approved accounting treatment.

A subscription figure can anchor the schedule without being mistaken for ROI: Shopify's UK entry-level plan starts at £19 GBP per month billed yearly, according to its published UK plan price. Record the plan and billing basis for each candidate, then add the merchant's own payment, app, implementation and labour evidence; this price alone is not a platform's total cost or return.

Check VAT treatment against HMRC's VAT Notice 700 and the merchant's approved accounting treatment. Include staff time at the merchant's approved loaded rate, even when no extra cash left the bank.

Establish a counterfactual

The strongest feasible design depends on risk, volume and rollout. A randomised or matched comparison may be possible for some bounded journey changes. A phased introduction can sometimes provide useful comparison periods, provided seasonality, marketing, stock, pricing and customer mix are addressed. A simple before-and-after chart is vulnerable to all of those changes.

Pre-register the population, intervention, primary outcome, exclusions, comparison, minimum observation period, analysis and stop conditions. Use synthetic or authorised test data before live collection. If a credible counterfactual is unavailable, make a bounded contribution statement and leave causal return blank.

Compute and stress-test

incremental contribution = measured contribution with change - estimated counterfactual contribution

return ratio = (incremental contribution - full economic change cost) / full economic change cost

Show a range for uncertain quantities rather than a single persuasive point. Vary returns, payment mix, fulfilment cost, labour, migration rework and effect estimate. Report missing data, sample limits and external changes. Do not annualise a short observation window without a stated, reviewed assumption.

Approve continuation only when the measurement owner can reproduce the data extract, the accountant reconciles contribution, tax treatment is reviewed, and the result remains acceptable under the agreed sensitivity. Otherwise hold, repair the evidence design or stop. The calculation supports a decision under uncertainty; it cannot promise future returns.

Retain the analysis code, query version, input extracts and exclusions with the decision record. A changed order-state definition or missing refund feed invalidates comparison until the measurement owner has rerun and reconciled it.

Before you act

Common questions

What is the difference between attributed outcomes and causal incrementality?

Attributed outcomes are assigned by a documented rule, such as a tagged session or matched campaign identifier. An attribution label does not show causation. Causal incrementality is the difference between what happened with the intervention and a credible counterfactual, which requires a comparable unaffected group or period.

How should refunds be treated when calculating net contribution?

Refunds need the actual provider treatment. For example, Shopify states that the original card processing fee is not reimbursed on a Shopify Payments refund. Other providers and configurations need their own evidence. Do not use gross checkout value as benefit; reconcile cash settlements to order records.

When can a return ratio be reported as causal?

Only when a credible counterfactual is available. If not, make a bounded contribution statement and leave causal return blank. The strongest feasible design depends on risk, volume and rollout. A simple before-and-after chart is vulnerable to seasonality, marketing, stock, pricing and customer mix changes.