Reconcile the reports before moving the advertising budget

Start where the money is recorded

Choose a completed reporting period and agree which orders belong in it. Decide whether you are comparing placed orders, paid orders or retained sales after refunds. Save the timezone, currency and extraction time. Without that agreement, even accurate reports may describe different business populations.

Reconcile unique order references and values to the commerce record. Repeated delivery of one purchase event is different from two channels each claiming a share of one real purchase. Fix the first as a collection problem; explain the second as a credit question.

Preserve each report's settings

Record the conversion action, customer filter, credit model and timing rules. Do not assume that “seven days” has identical meaning across products. Northbeam separates eligibility lookback from its attribution reporting window and distinguishes interaction-date from purchase-date reporting. Triple Whale's deterministic-view method has its own documented decay behavior.

The linked attribution guide explains these product distinctions and current source limitations. Your worksheet should record the actual account settings, including what cannot be verified. Changing a setting can restate history; annotate it instead of treating the movement as newly created demand.

Compare the differences that could change a decision

Start with the largest unexplained mismatch in the agreed scope. Check refunds, excluded orders, missing identifiers, date assignment and data maturity. Separate observed interactions from modeled relationships. A joined path can still miss discovery on another device or an untracked channel.

Write the evidence needed to settle each difference and assign a resolver. “The platforms disagree” is not an actionable diagnosis. “The refund integration omitted these orders from the closed-period report” is a specific repair with a completion check.

Put the channel story beside contribution

A higher credited ROAS can arise from a new allocation without changing one sale. Check product costs, variable delivery costs, returns and spend before using attributed revenue as a business return. Keep unknown costs visible; a polished report does not make them zero.

Compare new-customer groups at the same age. A mature cohort has had longer to reorder than one acquired last week. Separate realized contribution from predictions about future value and record the assumptions behind any forecast.

Decide what evidence the next budget change needs

Attribution can guide everyday campaign investigation. A material spending change may need an experiment that estimates additional demand. Broader marketing mix modeling can support planning when history, independent variation and relevant business controls justify it. Do not average incompatible ratios to manufacture agreement.

Finish with a bounded action, spending authority, review date and explicit uncertainty. A useful outcome may be holding the budget while a broken measurement link is repaired. The point is an accountable decision, not declaring one dashboard universally correct.

Sources for the product-specific window distinctions: Northbeam lookback windows and Triple Whale deterministic-view model. Official documentation reviewed 15 September 2026.

Read the reasoning behind the worksheet

Understand Meta, Google, Triple Whale and Northbeam

Reconcile the operating numbers

Check discount contribution

Work through the artifact

Looqus Editorial · Version 1 · 15 September 2026 Editable starter worksheet. All worked numbers are synthetic USD. This does not reproduce a vendor algorithm or estimate causal lift.

Use this worksheet

Choose one completed reporting period. Reconcile store records before comparing channel credit. Keep incompatible scopes separate and give unresolved differences an owner. Use approved exports; remove personal customer information from shared copies.

Business basis

Period, timezone and extraction time:

Included order states:

Gross or retained revenue:

Refund/cancellation treatment and observation horizon:

Tax/shipping treatment:

Currency and conversion policy:

Product/fulfilment/payment/return costs included:

Spend scope:

New versus returning customer definition:

Report settings register

Record actual settings. Different products use “window” differently. Do not assume a universal default.

Synthetic reconciliation

Social and search might each claim the same $160 under separate configured attribution rules. Adding the claims yields $320, not another sale. Keep the business ledger at $160 retained sales. This possible overlap is different from sending a duplicate purchase event to one destination.

Discrepancy register

Investigate order population, revenue definition, duplicates, refunds, identity, timing, model and data maturity. Do not assign the unexplained remainder to a favorite channel.

Spending decision record

Decision that depends on this comparison:

Reconciled business evidence:

Remaining uncertainty:

Whether an experiment or qualified MMM analysis is needed:

Contribution threshold and cash constraint:

Approved action and limit:

Responsible operator:

Verification evidence and review date:

Attribution credit does not establish additional demand. Improving matching can increase reported conversions without increasing store sales.