Growth Marketing for Ecommerce: Product, Conversion, Margin and 90-Day Store Growth Playbook

Choose the business state the decision requires. A submitted checkout may still fail payment, cancel, return or consume exceptional support. For an early usability test, completed payment may be sufficient. For acquisition expansion, the team may require delivered, retained or contribution-positive orders.

Write eligibility from observable conditions such as market, inventory availability, device, customer status and delivery coverage. Preserve exclusions. A campaign cannot be interpreted when people were invited to buy products or use delivery options unavailable to them.

Separate diagnostic steps from the accepted outcome. Product views, basket additions and checkout starts help locate friction. They do not replace the commercial state. Report them beside the journey while keeping the expansion rule tied to mature evidence.

Name the owner for product, pricing, payment, fulfilment and media decisions. Ecommerce growth crosses several systems. One team should not change an offer to repair a warehouse constraint without the responsible owner reviewing the customer consequence.

Record the assortment and stock state available to each cohort. A treatment may appear stronger because popular products returned to inventory. Compare the actual opportunity rather than assuming catalogue conditions were stable throughout delivery.

Version price, discount, shipping promise and eligibility. Promotional changes can alter both demand and contribution. If several components move together, call the intervention a package and avoid assigning the result to a single message or layout element.

Track voucher and incentive leakage. Customers outside the intended cell may discover an offer, while eligible people may fail to receive it. Preserve issue, redemption and exclusion records so apparent response is not confused with uncontrolled distribution.

Set a correction rule for feed, price or stock errors. Pause affected exposure, protect customers and identify the valid cohort. Do not delete inconvenient orders from the evidence simply because the treatment was delivered incorrectly.

Preserve channel and treatment assignment separately from observed exposure. Use approved identity rules to connect sessions and orders. Show unmatched, duplicated and cross-device states. A perfectly attributed subset may not represent all eligible customers.

Reconcile platform totals with the order ledger at a defined time. Media systems can support delivery diagnosis; the commerce system establishes payment and fulfilment. Document timezone, currency, tax and refund treatment before calculating a ratio.

Keep new and returning customers distinguishable when the decision depends on acquisition. A returning buyer can make a channel look efficient without adding a new relationship. Define the customer rule and its lookback rather than trusting a default label.

Delay final comparison until required reversals can appear. Recent orders remain pending. Publish an early operating view only when it is labelled provisional and cannot silently authorise additional spend.

State the mechanism under review: delivery expectation, payment friction, form comprehension or product confidence. Select evidence capable of distinguishing it. A shorter checkout can increase submissions while reducing informed customer choice.

Preserve required disclosures and obtain relevant legal or policy review. Do not treat a higher completion rate as permission to obscure recurring terms, fees, availability or cancellation. Customer safeguards remain outside the optimisation trade.

Monitor errors, complaints, failed payments, duplicate orders and service contacts. These signals may mature faster than returns and can justify an early pause. Assign the response route before the treatment reaches customers.

Review accessibility across actual devices and payment paths. A desktop visual check cannot establish mobile or assistive usability. Record the tested combinations and leave unobserved paths outside the conclusion.

Define the repeat state relevant to the product: replenishment, second category purchase, subscription continuation or reactivation after a meaningful lapse. Calendar repeat alone may be unsuitable for durable or seasonal goods.

Respect permission, suppression and service context before sending. Customers with unresolved delivery, return or complaint states should not enter a generic promotion because a lifecycle timer fired. Eligibility belongs in the operating rule.

Measure incremental accepted value against a valid comparison where feasible. Report unsubscribes, complaints and discount dependence. A campaign that moves orders forward in time without changing value needs a different conclusion from genuine retention.

Limit contact changes to a bounded cell and preserve message versions. Frequency, offer and audience can interact. If all change together, the team learns about a package rather than an isolated timing effect.

Model the products, markets and delivery promises that additional demand will reach. The initial cohort may consume available stock or warehouse capacity. Expansion must use the next operating state, not the average from the easiest orders.

Set ceilings for spend, accepted orders, service workload and customer harm. Monitor the limiting constraint. A lower acquisition price is not useful when cancellations or late deliveries rise beyond the agreed boundary.

Open one dimension at a time when practical, such as market, product group or source. Preserve stable cells. If the new tranche fails, roll back its exposure without rewriting the evidence for conditions that remained sound.

Recalculate contribution with actual discount, fulfilment, payment and recovery cost. Do not publish a universal return target. The decision applies to the measured assortment, customer group, time and operating capacity.

Freeze the eligible cohort, treatment versions, delivery extract, journey definitions, order reconciliation and cost rules. Another reviewer should reproduce the accepted count and see every pending or unknown state.

List customer issues and their owners before declaring the test closed. Refunds, duplicate orders, delayed fulfilment and data correction continue after media stops. The learning budget cannot cancel the recovery obligation.

Record the authorised action as a market, product, source and capacity boundary. Include what would reopen the decision, such as stock change, new payment route, altered return policy or material source drift.

Feed reusable findings into product and measurement governance only after review. Preserve rejected causes and null results. Future teams need the conditions that failed, not a library containing only favourable commerce stories.

Record the customer-facing currency, settlement currency, conversion source and timestamp used in the review. Taxes, duties and payment fees can affect products or markets differently. Do not compare displayed revenue cells until their treatment is aligned and documented.

When exchange movement is material, separate commercial performance from translation variance. A later rate should not rewrite the original order evidence. Finance owns the accounting treatment while the growth decision retains its dated calculation and limitation.