Marketing Attribution Setup: Implementation and QA Checklist

A useful attribution setup starts with consistent transactions, campaign tagging and reporting rules. Choosing an advanced model before fixing those foundations can give you precise-looking answers to the wrong question.

This guide is an implementation checklist for marketing teams. For definitions and a worked comparison of six models, use marketing attribution models explained.

1. Write down the decision and outcome

Define what the report should support: understanding acquisition, comparing campaigns, inspecting customer journeys or informing a budget test. These are different jobs.

Choose a primary outcome with a clear definition. For ecommerce, document whether purchase revenue includes discounts, tax, shipping, refunds and cancellations. For lead generation, distinguish form submissions, qualified leads and closed sales. Do not optimize a sales decision against a report that only measures form fills.

Record the owner, source system, refresh schedule, currency and reporting timezone. A transaction record is the reference for what sold; attribution is a method for assigning credit for that sale.

2. Make event collection consistent

Walk through a test transaction and check that it appears once in the transaction system and once in each intended measurement destination. Use a consistent transaction identifier where supported. Confirm that reloads, payment redirects and browser/server event duplication do not create additional purchases.

Then test refunds, consent choices, cross-domain checkout, mobile purchases and missing campaign parameters. Respect the user's choices throughout the flow. A server-side integration changes event delivery; it does not make every transaction attributable to a known user or channel.

Keep a small QA log:

3. Standardize campaign names and channel rules

Use a shared tagging convention for manually tagged campaigns. Keep source and medium values consistent in spelling and case, document campaign identifiers and avoid putting personal data in URLs. For products with auto-tagging, confirm how it interacts with your analytics implementation.

Separate branded search, non-brand search, prospecting and retargeting when the business decision requires that detail. A blended paid-search number can hide very different audiences and levels of existing demand.

Do not add campaign tags to internal site links to measure navigation. That can interfere with acquisition interpretation. Use an appropriate internal event instead.

4. Distinguish acquisition from conversion credit

A first-user channel tells you how the recorded user was acquired. A session channel describes a visit. Key-event attribution allocates credit for an outcome. A returning user can therefore appear under one first-user channel and another session channel without either report being wrong.

In GA4, confirm the dimension scope before comparing reports. Google's attribution settings documentation explains how the reporting model applies to event-scoped traffic dimensions, rather than first-user and session dimensions.

Choose a lookback window that is plausible for your buying cycle and document it. A window is an eligibility rule, not proof that an interaction influenced the purchase.

5. Reconcile totals before comparing channels

Use the same time period, currency and revenue definition. The following worksheet is an illustrative QA example, not an expected match rate:

The $8,000 analytics gap needs investigation; do not automatically label it tracking loss. The $115,000 sum of platform claims is not deduplicated business revenue. Different platforms can credit the same order.

Keep a reconciliation log with the amount, likely explanation, evidence, owner and next check. Correct identifiable collection errors before changing an attribution model to make the totals look better.

6. Compare models without claiming causality

A model comparison is a sensitivity analysis. If branded search loses credit when you move from last-click to another model, you have learned that allocation depends on the chosen method. You have not established that spending less on branded search will preserve sales.

Use the comparison to choose a question worth testing. For example: does retargeting create enough additional contribution to cover its cost? A randomized holdout or a carefully designed geo experiment can address that question more directly than changing attribution weights.

7. Calibrate with a matched experiment

Suppose a test estimates $5,600 incremental revenue while the platform attributes $33,600 for the same audience, period and spend. The calibration factor is 16.7%. The iROAS calculation guide shows the full arithmetic and the distinction between relative lift and this factor.

Use the free iROAS calculator to compare the ratios and margin threshold. Record the experiment's uncertainty alongside the factor. Apply it elsewhere only when the conditions are sufficiently comparable; do not treat a single campaign test as a permanent channel-wide constant.

For strategic planning, compare experimental evidence with media mix modeling. Agreement is helpful, but two methods can share assumptions or data errors. Investigate large differences rather than averaging them automatically.

A practical review rhythm

Check transaction totals and event health frequently enough to catch errors before they affect decisions. Review channel definitions when campaign structures change. Revisit calibration when audiences, offers, budgets or attribution settings materially change.

Keep the reporting definition stable during an evaluation. Annotate implementation changes and compare like-for-like periods. A higher reported ROAS after a tracking change is not necessarily a better campaign.

For a transaction-level example, use the Shopify and GA4 revenue reconciliation walkthrough. It separates missing purchases, duplicate collection and revenue-definition differences before investigating channel allocation.