Scope: what this audit covers and what it does not
The scorecard covers a web analytics implementation end to end: the written plan of what should be measured, the tag manager container that deploys tags, how consent choices govern those tags, the events and key events in the GA4Google's event-based analytics platform. property, ecommerce or lead data, how campaign traffic is labelled, the links to ad platforms and other tools, and whether the resulting numbers agree with the business's own records. It applies equally to ecommerce and lead generation sites.
It does not re-teach configuration. Building a property is covered by the Google Analytics set-up checklist; proving that a tag fires correctly is covered by the tracking verification checklist. It does not judge marketing performance or the choice of AttributionAssigning conversion credit to touchpoints. model for budget decisions, which belong in reporting work, and it does not offer legal advice. Consent is scored as a technical behaviour: do tags respect what the visitor chose? Whether the organisation's approach meets the law, including the GDPR where it applies, is for its privacy lead to judge. Whether the tools behind the integrations should stay in the stack at all is a separate question, which the tutorial on running a marketing stack audit answers tool by tool.
Before you score: access and evidence
You need editor access to the analytics property and read access to the tag manager container, plus a person who can see the business's own figures for orders or qualified leads. Ask for the measurement plan or tracking plan; if none exists, that is already a finding. Agree a fixed reconciliation window, such as a recent complete month, and use it at every rescore.
Before rating, export the container's tags, triggers and variables, the list of events and which are marked as key events, and the property's data settings. Use the tag manager's preview mode and the browser's network panel to walk through the main journeys with consent accepted and then declined. Pull the traffic acquisition report to see how much traffic arrives unlabelled or with inconsistent UTMURL tags identifying traffic source. values. Write down the business totals for the window next to the analytics totals.
While the exports are open, search page addresses, search terms and custom dimensions for email addresses, names and phone numbers; any hit is a finding, and the guide to keeping personal data out of analytics sets out the patterns to search for and the fixes.
Criteria and scoring
Rate each row missing, partial or solid. The plan, consent and key event rows are gates.
| Criterion | What good looks like | Score guidance |
|---|---|---|
| Measurement plan (gate) | A current document links business goals to events, parameters and owners | Missing: no plan / Partial: outdated or incomplete / Solid: current and used |
| Container organisation | Tags, triggers and variables named to a convention; no dead or duplicate tags | Missing: unmanaged sprawl / Partial: partly tidy / Solid: clean and documented |
| Container governance | Publishing restricted, versions described, changes reviewed | Missing: anyone publishes / Partial: some control / Solid: reviewed releases |
| Consent behaviour (gate) | Tags wait for or adapt to the visitor's choice, as configured | Missing: tags fire regardless / Partial: some tags bypass consent / Solid: all tags respect choices |
| Event design | Events and parameters follow the plan and a naming convention | Missing: ad hoc events / Partial: inconsistent names / Solid: planned and consistent |
| Key events (gate) | Only real business outcomes marked as key events; each fires once per outcome | Missing: wrong or duplicated / Partial: correct but includes micro-actions / Solid: clean set |
| Ecommerce or lead data | Purchase or lead events carry value, currency, identifiers and items where relevant | Missing: no value data / Partial: incomplete parameters / Solid: complete and accurate |
| Cross-domain and payment journeys | Sessions survive moves to checkout, booking or payment domains | Missing: payment provider shows as referrer / Partial: some breaks / Solid: continuous |
| Campaign tagging | Every paid and email link tagged to a shared convention | Missing: largely untagged / Partial: inconsistent values / Solid: consistent convention |
| Channel grouping | Default or custom groupings put traffic in sensible buckets | Missing: large unassigned share / Partial: some misfiled / Solid: clean grouping |
| Platform integrations | Ad platforms, search and CRM linked where the plan needs them | Missing: none linked / Partial: some links / Solid: all needed links working |
| Internal and spam filtering | Internal traffic and known junk excluded | Missing: no filtering / Partial: some / Solid: defined and active |
| Data settings | Retention, reporting identity and user-data settings chosen deliberately | Missing: defaults never reviewed / Partial: some reviewed / Solid: documented choices |
| Reconciliation | Key event totals agree with the back office within an agreed tolerance | Missing: never checked or far apart / Partial: checked occasionally / Solid: regular and close |
Two auditors should reach the same score from the same evidence, so write observable lines into the plan before scoring. A key event that fires more than once for the same outcome counts as duplicated, and the key events row scores Missing however small the excess. Rows such as reconciliation and channel grouping need lines of your own, and the example below shows the form they can take.
Illustrative scoring lines (set your own)
Reconciliation scores Solid when totals sit within the agreed tolerance (the worked example below uses 3%) and are checked monthly, Partial when they are checked less often or sit outside tolerance but within 10%, and Missing when they are never checked or are more than 10% apart. Channel grouping scores Solid when under 5% of sessions sit in an unassigned bucket, Partial between 5% and 15%, and Missing above 15%.
For consent and data settings, score against the behaviour the Google Analytics Help centre documents and the consent mode guide explains, not against what a banner vendor promises. Key events are scored using the definitions in the key events and conversions guide.
How to score and prioritise
Score the three gates first. A missing plan means there is nothing to score the events against, so write a short one with the stakeholders before continuing; the tracking plan guide gives the structure. A missing consent score is both a data problem and a potential compliance exposure, so it goes to the top of any list and to the privacy lead the same day. A missing key events score means every channel report is wrong in a way nobody can see.
Then rank the remaining rows by how many decisions they distort. Reconciliation gaps and cross-domainConfiguring analytics so a visit that moves between a business's own domains stays one session. breaks usually distort revenue by channel and rank high. Campaign tagging gaps distort channel credit, which matters most where budgets are moved on the reports. Those gaps show up first in the channel groupingRules that assign each visit or conversion to a channel, such as organic search or email, from its source, medium and campaign values., as traffic left in an unassigned bucket. Container tidiness and naming rank lower, unless the GTMTag management system. container is so tangled that every other fix is risky to make.
Reading the result: what to do with each band
Red means a gate scored missing. Tell report owners which figures are unreliable, pause decisions that depend on them, and fix the gate first. Re-run the tracking verification checklist after each fix and rescore once a complete reconciliation window has passed.
Amber means the gates hold but several rows are partial or missing. Schedule fixes in releases ordered by how many decisions each gap distorts, and add a short note to dashboards that names any known gap until it is closed.
Green means the set-up can be trusted. Keep reconciliation on a monthly rhythm, rescore after any site release that touches templates or checkout, and update the plan when goals change.
The academy exercise on combining two paid programmes after a cloth nappy maker buys a rival practises acting on a red band in the business just bought, whose duplicated purchase event keeps its analytics out of the baseline until the fix passes a rescore, with its back office read instead.
From audit to action plan
Each partial or missing row becomes a ticket with the finding, the expected behaviour from the plan, the evidence (a screenshot or network trace), the owner and how verification will be done. Bundle container changes into versioned releases so each can be rolled back. Share the band and the list of figures currently unsafe to use with everyone who reads the reports. Where the tickets cluster in an inherited property's own settings, such as key events, custom definitions, interface rules, audiences and filters, the tutorial on cleaning up an inherited GA4 property works through them in an order that moves whatever reads each one before it changes.
A furniture shop's reconciliation gap
An online furniture shop finds its analytics reports 1,180 purchases for a month in which the back office recorded 940 orders. The audit traces the gap to two causes: the purchase event fires again whenever the confirmation page reloads, and a payment provider's domain breaks sessions, so a share of orders are credited to the payment site as a referral. Key events score missing, because the purchase event fires more than once per order, and cross-domain journeys score missing. Reconciliation scores missing too, with analytics about 25% above the back office against the 10% line. Consent scores partial because a chat widget's tag fires before choice. The key events gate is missing, so the band is red, whatever the other rows say. Fixes ship in two container releases; the following month analytics records 905 purchases against 921 orders, 1.7% apart and within the 3% tolerance the team agreed.
Common mistakes
- Scoring without a plan. Events cannot be right or wrong without a document that says what they should be. Write the plan first.
- Testing only with consent accepted. Most consent faults show up on the decline path. Walk both.
- Trusting totals that look plausible. Plausible is not reconciled. Compare against the business's own records.
- Treating legal compliance as a score. The audit checks technical behaviour; legal judgement belongs to the privacy lead.
- Tidying the container before fixing the gates. Renaming tags feels productive while revenue is being double-counted. Gates first.
- Rescoring on a different window. A month with a sale and a quiet month reconcile differently. Keep the agreed window so the rescore compares like with like.
- Leaving dashboards unlabelled. Readers act on what they see. Until a known gap is closed, say so on the report itself.
