Ecommerce profitability beyond ROAS: measuring margin, discounts, fulfilment, returns and repeat purchase

An ecommerce dashboard can show rising revenue and a strong return on ad spend while the commercial result remains uncertain. The campaign may be attracting heavily discounted orders, low-margin products, expensive fulfilment, frequent returns or customers who never buy again. None of those conditions makes ROAS useless. They show that ROAS answers a narrower question than leaders often ask it to answer.

ROAS normally compares an advertising conversion value with media cost. It can help assess and automate media efficiency when values are accurate and appropriately defined. It does not, by itself, establish gross margin, contribution, cash flow or profit. A more useful operating view connects ad-platform signals to trustworthy transactions, order economics and customer behaviour, then reconciles the result with finance-owned definitions.

The short answer

Build the smallest dependable measurement stack. First, validate purchase, value, currency, transaction identifier and refund events. Second, add finance-approved product cost, discount and variable order-cost signals. Third, distinguish new and returning customers and observe repeat purchase in cohorts. Finally, reconcile the marketing view with commerce, operations and finance at an agreed cadence. Use ROAS for the advertising decision it supports, but do not label attributed revenue as profit. If foundational data is unreliable, repair it before feeding more complex values into automated bidding.

The smallest useful view begins with trustworthy transactions, not a perfect enterprise model.

ROAS answers a narrower question than profitability

A platform ROAS can be written conceptually as attributed conversion value divided by advertising cost. Both parts have boundaries. Advertising cost may exclude agency services, production and internal time. Conversion value may be gross order revenue recorded before later refunds, and attribution depends on the platform’s model, window, identity evidence and consent conditions. Two platforms can each claim influence over the same order. The figure is therefore a platform decision signal, not a complete business ledger.

Keep the metric, but label it honestly. Use it to compare campaign or product-set performance within a controlled context, guide media allocation and evaluate whether value-based bidding is receiving a sensible signal. Pair it with commercial measures outside the ad platform. If leaders want to know whether growth is healthy, the report must show which costs and adjustments are included, which are missing, who owns the definition and how far the result has been reconciled.

Move through a metric ladder

• Platform ROAS: fast media feedback, limited by attribution and revenue quality.
• Net revenue: removes cancellations, discounts and refunds where dependable.
• Contribution proxy: adds approved product, fulfilment and return assumptions.
• New-customer and cohort views: separate acquisition from repeat behaviour.
• Finance-reconciled profit: remains owned and defined by finance.

Attribution-based ROAS, blended marketing efficiency and causal incrementality answer different questions. A strong attributed result does not prove that every order was caused by the reported spend.

Start with reliable revenue and refund events

Advanced profitability analysis built on duplicated purchases or missing refunds creates false confidence. Validate the transaction foundation first. Each purchase should carry the correct value and currency with a stable transaction identifier. Test duplicate prevention, tax and shipping treatment, payment success timing, cross-domain journeys and consent behaviour. Confirm that the same transaction can be traced through the commerce platform, analytics, advertising destinations and finance records without assuming that every system will report identical totals.

Refunds deserve equal attention. Define whether full and partial refunds are sent back to analytics and advertising systems, when they appear and which identifiers connect them to the original transaction. Returns can happen after an advertising reporting window has closed, so historical campaign figures may not fully restate. Record that limitation. Use sampled order reconciliation and controlled test purchases rather than trusting the presence of a purchase event in a tag manager preview.

Add product and order-margin signals

Ask finance which decision-ready margin definition can be used. Product cost may come from standard, average or actual cost and may change over time. Discounts can be product-level, order-level, promotional, loyalty-based or funded by another party. Taxes, duties and shipping revenue may be included in one dashboard and excluded from another. There is no universal formula suitable for every retailer. Publish the definition beside the measure and preserve the finance system as the authority for financial reporting.

A practical marketing proxy might subtract approved product cost and selected variable order costs from net order revenue. That can distinguish a high-revenue, low-contribution product from a lower-revenue, stronger-contribution product. Call it a proxy unless it reconciles to the finance definition. Do not silently include estimates as facts. Where exact cost data cannot be sent to media platforms, a product group, margin band or another governed signal may be a safer first step.

Account for fulfilment and returns

Order economics continue after checkout. Pick-and-pack, packaging, payment charges, shipping subsidies, oversize handling, split shipments and customer-service contacts can vary materially by product and location. A free-shipping promotion can lift conversion while creating a different cost profile. Define which costs are sufficiently reliable and variable to inform the decision. Avoid allocating every fixed overhead to an individual advertising click merely to make the model look complete. Transaction evidence should reconcile with ecommerce checkout optimisation.

Keep tax, cash timing and working-capital questions visible but separate from the marketing proxy. A prepaid order, delayed supplier payment, marketplace remittance or buy-now-pay-later arrangement can affect cash without changing the ad platform’s reported value. Inventory commitments can also make a high-volume promotion commercially difficult. These issues should be interpreted by finance and operations using their authoritative records. The marketing dashboard can flag relevant order groups, but it should not convert those flags into accounting or treasury conclusions. Record the finance owner and review cadence so a material cash constraint reaches campaign decisions before media or promotions are expanded.

Returns and exchanges require their own view. Measure by product, reason, campaign, customer cohort and time where the data and privacy position allow. A high return rate may reflect fit, quality, description, fulfilment or customer expectation rather than advertising alone. Include restocking loss, return freight and service effort only under an agreed definition. The aim is to locate commercial patterns and improve decisions, not to manufacture one precise profit number from incomplete operational data.

Finance must approve the included costs, timing and treatment used by the organisation.

Separate customer acquisition from order acquisition

A new customer’s first order and an existing customer’s routine reorder are not equivalent commercial outcomes. Define customer status consistently. Decide which source resolves identity when a guest checkout uses a new email, a household shares details, an order is placed through another channel or records are merged. The ad platform’s new-customer label may use different evidence from the commerce or CRM system. Compare definitions before using the field in reporting or bidding.

Then separate the questions. What does it cost to acquire a genuinely new eligible customer? What does it cost to stimulate another order from an existing customer? Which categories attract customers who return? Which promotions bring high first-order value but weak subsequent behaviour? These questions help marketing balance acquisition, retention and margin without assuming that every reported purchase deserves the same value.

Use cohorts to observe repeat purchase

Group customers by a meaningful starting point, such as first purchase month, first product category, acquisition campaign or offer, then observe behaviour over comparable periods. Track the proportion that makes another purchase, time to repeat, net revenue, returns and contribution proxy where available. Cohorts can reveal that two acquisition sources with similar first-order ROAS create different later patterns. They do not prove that the source caused every later purchase. Cohort design must respect first-party data and consent.

Allow cohorts to mature before comparing them. A recently acquired group has had less opportunity to repeat than an older group. Account for seasonality, stock availability, product replacement cycles, promotions and channel changes. Avoid turning an early historical pattern into a guaranteed lifetime-value forecast. Use ranges, disclose the observation window and update the model as behaviour changes. Lifecycle marketing can be evaluated against this evidence, but strategy and execution remain separately scoped paid work.

Choose optimisation values carefully

Google Ads value-based bidding uses the conversion values supplied to optimise for value, with or without a target ROAS depending on the strategy. If values represent gross revenue, the system seeks that declared value, not an unreported finance outcome. Supplying a better commercial proxy can improve alignment, but only when the data is timely, stable, sufficiently complete and understood. Poor values can automate the wrong objective faster. Emote’s search advertising work can use approved value signals where appropriate.

Begin conservatively. Validate the measurement, compare the proxy with finance outcomes and document exceptions before changing bidding. Conversion value rules can adjust values for defined characteristics, but rules are not a substitute for clean source data or product economics. Monitor the effect after any change and avoid simultaneous campaign, feed, attribution and value-model changes that make diagnosis difficult. Current Google Ads capabilities and eligibility must be rechecked immediately before implementation.

Define ownership and reconciliation

Assign each field to a source and owner. Marketing owns campaign configuration and media interpretation. Analytics owns event design and data-quality evidence. Ecommerce owns catalogue, transaction and promotion context. Operations owns fulfilment and return processes. Finance owns cost, margin and financial reconciliation definitions. Customer or CRM owners define identity and status. One working group can govern the model, but it should not blur these responsibilities.

Set a reconciliation cadence and tolerance. Daily advertising decisions may use a faster proxy, while monthly commercial reporting uses matured refunds and finance data. Explain why totals differ rather than forcing every source to match. Preserve version history when definitions change. A campaign should not appear to improve merely because shipping treatment, customer classification or attribution settings changed. Detailed analytics design, data engineering and dashboard implementation require a paid scope.

Build the smallest useful profitability view

• Stage 1: trustworthy purchase, value, currency, transaction and refund evidence.
• Stage 2: finance-approved product cost, discount and selected variable-order inputs.
• Stage 3: reliable new-versus-returning classification and comparable customer cohorts.
• Stage 4: controlled optimisation values, reconciled reporting and documented limitations.

Do not delay every decision while waiting for a perfect data warehouse. A transparent model with fewer reliable fields is more useful than a sophisticated dashboard built from disputed definitions. Start with the commercial question, identify the minimum evidence needed to answer it, and add a layer only when it changes an action. Record what the model excludes and when that omission becomes material.

Use scenarios and decision thresholds, not one blended average

A whole-store average can hide the mechanism that needs action. Segment the view by product or margin group, customer status, fulfilment route, promotion, geography and return pattern where sample size and privacy permit. Compare a base case with plausible changes such as higher shipping cost, lower discount, more returns or slower repeat purchase. The purpose is not to predict an exact profit result. It is to identify which assumptions have enough commercial consequence to justify better data, a campaign change, a merchandising response or an operational investigation.

Set thresholds from the organisation’s own economics and risk appetite. A campaign might be allowed to learn while conversion volume is low, paused when a tracking defect makes values unsafe, or constrained when a product group falls below a finance-approved contribution proxy. Document who can change media targets, who approves value-model updates and what evidence is required. Automated bidding should not receive a new value definition merely because one reporting period looks weak; model changes need versioning, controlled release and an evaluation window.

Keep a decision log beside the dashboard. Record the observed signal, data limitations, action, owner, expected mechanism and review date. This separates measurement from retrospective storytelling. If profitability improves after a pricing, feed and campaign change occurred together, the team can acknowledge that the combined intervention worked without assigning unsupported credit to one element. Over time, the log also reveals where decisions repeatedly stall because a cost field, customer definition or refund process is not dependable. That evidence can justify the next paid measurement or platform scope.

Separate measurement quality from commercial performance

A falling reported result can mean the business performed worse, or that measurement became more complete. Adding refunds, correcting duplicated orders or replacing gross revenue with a contribution proxy may reduce the number while improving the truthfulness of the decision. Maintain a data-quality view beside the performance view. Include event coverage, unmatched transactions, refund completeness, cost-data freshness, identity confidence and reconciliation difference. This prevents teams being penalised for making the model more honest and helps leaders distinguish a commercial problem from a tracking change.

Use change windows carefully. When a new value model is introduced, preserve the previous reporting series, label the effective date and avoid direct comparison until a suitable overlap or reconciliation exists. If historical data cannot be restated, begin a new baseline. Inform media operators before values change so bidding behaviour can be monitored and budgets controlled. A finance-approved definition can still be unsuitable for real-time optimisation if it arrives too late or at insufficient volume, so reporting and bidding may need related but distinct governed values.

Marketing can use commercial signals without silently redefining finance measures.

Frequently asked questions

Is a high ROAS always good?

No. Interpret it against margin, volume, attribution, returns, fulfilment, customer status and the business objective. A high ROAS can still describe low-scale or commercially weak orders.

What ROAS should an ecommerce business target?

There is no universal target. It depends on margin, operating costs, customer value, cash constraints, growth objectives and measurement definitions. Finance and marketing should model the organisation’s own economics.

Can profit be sent to Google Ads?

Value-based bidding can use supplied conversion values, including governed commercial proxies where technically supported. The design requires current platform validation, reliable data and finance approval. It does not turn Google Ads into the financial ledger.

Should returns be deducted from campaign revenue?

Returns should be visible in commercial analysis, but treatment depends on timing, identifiers, platform adjustment capability and finance policy. Document late-return limitations rather than claiming complete restatement.

How should repeat purchase be measured?

Use defined customer cohorts and comparable observation windows. Track repeat timing and net commercial behaviour, while allowing for seasonality, product cycles and identity gaps. Do not present historical correlation as guaranteed lifetime value.

Does Emote provide accounting advice?

No. Emote can support measurement design, analytics and marketing optimisation. The client’s finance team or qualified adviser must approve financial definitions and accounting treatment.

How Emote can help

Emote can help connect ecommerce marketing and measurement to the commercial inputs that sit behind revenue, including margin, fulfilment, returns and repeat purchase.

The smallest credible starting point is a focused diagnostic of available data, definitions and decision gaps before media or technology changes are prescribed. Full Website Discovery is used only where cross-system measurement needs formal definition.

If you want a clearer basis for judging ecommerce performance beyond ROAS, book a meeting with Emote.