Platform ROAS is not profit

ROAS is usually attributed sales divided by media cost. Platforms may use different click and view windows and can claim the same order, while finance uses net revenue, cost, refund, commission, and fulfilment.

Retain platform-reported ROAS beside a governed attribution view and a contribution view. The unified view exposes assumptions rather than claiming one unknowable truth.

Create stable touch and order identities

Ad facts retain platform, account, campaign, creative, event time, click ID, cost, and currency. Order lines retain stable order, customer, item, discount, refund, cost, and status.

UTM text helps classification but does not replace click and order IDs. Keep raw values and version normalization mappings.

Align timezone, currency, and watermark

Convert event time under a declared business timezone while retaining source time. Convert media and order value under approved FX policy.

Cost arrives quickly while refund and commission mature later. Show spend, order, refund, and cost completeness separately.

Treat the attribution window as policy

One, seven, or thirty days produce different outcomes, and view-through credit can amplify brand channels. Define eligible events, start, end, priority, and purchase-cycle rationale.

Version window changes and replay their impact. Never add platform-default attributed totals across channels.

Conserve credit across competing channels

An order may match search click, social view, and owned link. Last non-direct, positional, or data-driven models are choices; total assigned credit should remain one.

Persist an order-touch bridge with model, weight, and reason. Preserve uncertain or unattributed identity rather than force a risky match.

Move from sales to contribution

Subtract product, discount, commission, payment, fulfilment, refund, and media cost under an explicit fixed-cost boundary.

Report coverage when line cost is missing. A versioned estimate is preferable to silently treating missing cost as zero.

Separate attribution from incrementality

Attribution distributes observed orders; incrementality estimates what would disappear without advertising. Brand search may receive last-click credit for existing demand.

Use randomized geography or audience tests where possible, or transparent quasi-experiments with assumptions and uncertainty.

Mature cancellations and refunds by order cohort

Reverse refunds through the original order and credit weights, including partial line refunds. A cancelled order contributes no sales.

Update order-date cohorts through explicit maturity points. Deducting old refunds from today’s campaign distorts both periods.

Accept through reconciliation and edge cases

Test duplicate claims, cross-device gaps, missing click IDs, midnight, currency, partial refunds, late spend, and organic traffic. Enforce credit and amount conservation.

Reconcile platform invoice, order system, and finance net revenue, categorizing window, identity, refund, currency, tax, and watermark differences.

BI0 boundary

Pilot one channel and mature month, then publish platform, unified attribution, and contribution views before testing incrementality.

BI0 may be evaluated for integrating ad, order, and cost data. Platform connectivity, identity, attribution, and experiment orchestration require project verification; no uplift is guaranteed.

Public references

Take the next step with BI0.AI

Talk through a real business scenario and see how governed AI BI can fit your team.

Explore BI0.AI