Measuring landing pages: metric cards, six measurement layers, a worked 8% rate and PECR consent limits

Landing page measurement starts with a metric contract, not a dashboard tile. Product-specific definitions of a form start or a conversion differ between platforms, so copy the definition from your own tool onto the card. Define the event, eligible population, denominator, period, Europe/London timezone, source, exclusions, missing data and owner before reading a result. The key metrics for each page then stay comparable from one report to the next.

What to take away

Define the event, population, denominator, period, exclusions and owner before reading any result.
Use one metric card per reported value, since two systems can label different actions as conversion.
Keep availability, form completion, qualified leads, transactions, attribution and incrementality in separate layers.
Treat consent and privacy rules as part of data quality, because they decide which visits you can count.
Leave decision thresholds blank until the owner approves them.

Write the decision before choosing a measure

A report should answer a named decision. Examples include whether a released page works as specified, whether a form route needs correction, or whether there is enough evidence to continue a controlled comparison such as A/B testing.

Record the decision owner, the options, the evidence window, the earliest decision date and the action attached to each threshold. Leave the thresholds blank until that owner approves them.

Design any evaluation before release. The Magenta Book is HM Treasury guidance on what to consider when designing an evaluation, and its page shows an update on 15 May 2026. It is written for central government, so borrow the discipline, not the authority.

Use one metric card for every reported value

Each metric card holds the same fields, so two people reading it reach the same number:

Name of the metric and the decision it supports.
Event counted, and the system that records it.
Eligible population, numerator and denominator.
Period, timezone, source and query version.
Exclusions, missing, blocked and corrected records.
Known bias, uncertainty and comparability warning.
Accountable owner, blank threshold and review trigger.

The card below is a worked example with illustrative figures.

Field / Entry (illustrative)

Metric and decision: Form completion rate; does the enquiry form need correction
Event and system: Accepted submission recorded by the form handler
Eligible population: Form starts in the period, after test records are removed
Numerator and denominator: 40 accepted submissions and 500 eligible form starts
Result: 8%
Period and timezone: One named calendar month, Europe/London
Exclusions: Test submissions and duplicates inside the declared window
Known limit: Starts are counted only where consent allows measurement
Owner and threshold: Named owner; threshold blank until approved

Two systems can display conversion while counting different actions, people, windows or dates. Freeze the definition with each report. If the query changes, version the series, or recompute the earlier period on the new basis and explain the change.

Keep the measurement layers apart

The table shows what each layer counts and what it cannot show. Do not merge layers in one tile.

Layer	What it counts	What it cannot show
Availability	A defined endpoint or journey responding	That a visitor could submit the form
Form completion	Accepted submissions over eligible form starts	Lead quality
Qualified leads	Records meeting the buyer's own status rule	Revenue
Transactions	Source-of-record events such as settled orders	Which page caused them
Attribution	Credit assigned under a stated model	Causal effect
Incrementality	Difference against a credible comparison	Anything without a counterfactual

Form completion needs a declared denominator. In this illustrative example, 40 accepted submissions from 500 eligible form starts is a completion rate of 8%. Using page sessions as the denominator would give a different figure for the same page, so name the base on the card.

A qualified lead is a buyer-defined status. Document the eligibility fields, the reviewer or rule owner, the evidence time and how duplicates are handled. A missing review is not a failed lead unless the card says so.

Transactions come from the finance or order system, not from a landing page display. Revenue recognition, VAT and refunds need qualified accounting review. Attribution assigns credit under a model, so label the product, conversion action, model and lookback window, and keep a row for conversions that stay unattributed. See measurement mistakes for traps that merge these layers.

Treat privacy rules as part of data quality

The ICO is the regulator of data protection and other information rights legislation. Under regulation 6 of PECR, a person must not store information, or gain access to information stored, in a user's terminal equipment.

Check each script and pixel on the page against that rule.

The ICO guidance on the exceptions sets out five exceptions to the prohibition on storing or accessing information on people's devices under PECR. They are communication, strictly necessary, statistical purposes, appearance and emergency assistance. Outside them, consent is needed.

The guidance was finalised on 29 April 2026. The statistical purposes exception means you do not need consent where the sole purpose is to collect statistical information about how the service is used, with a view to improving it. You must give clear information about the purpose and a simple, free way to object.

It is essentially an analytics exception, and a narrow one. It covers aggregate figures such as how many people access your service, what they access and for how long. It is not for identifying, tracking or monitoring people, and it does not apply to online advertising.

If your service uses storage or access technologies for online advertising, you must get consent. That includes ad measurement and performance. Linking a particular user across sites and devices is not strictly necessary to provide your service, so it also needs consent.

A dashboard should therefore show the eligible population, any blocked or refused measurement you can observe, and every change in configuration. Do not fill gaps with an invented multiplier. Describe the coverage gap, and avoid comparing periods with different collection rules.

Map sources before building the dashboard

Create a source register covering the landing page platform, web analytics, consent records, advertising platforms, customer system, order system, finance ledger and incident log. For each source, name the owner, event definition, identifier, timezone, retention, known delay and extraction version.

Then document the joins. A source that exists is not necessarily complete or suitable. Keep raw extracts protected, and store the query version with each run. If someone edits a figure, the correction log should record the old value, new value, reason and approver.

Build a dashboard that exposes uncertainty

Put data health above outcome tiles. Show the extraction time, latest complete event time, source status, unmatched-record count, missing-field count, correction notice and definition version. A reader should see whether a series is comparable before seeing whether it rose or fell. The reporting dashboard guide gives a layout.

For each result, display the value or a blank, the denominator, period, timezone, source, exclusions and an uncertainty note. Show modelled and observed values in separate fields, and never add them unless their units and populations match.

Set decision rules without manufacturing certainty

Write each rule as an incomplete contract until the buyer approves it. Illustration: if the owner approves a 10% review trigger, the 8% completion result above opens a form review, provided the data-quality gates pass.

The general rule has this shape: if a defined metric crosses a blank threshold for a blank period, and data-quality gates pass, a named owner takes a specified action. Add a minimum evidence condition, an exception path and a review date.

A form fault may need immediate action without any claim about commercial performance. A movement in attributed conversions may justify investigation, but not a budget change, if the denominator, consent rate, model or query changed. Check operations and delivery rules before acting on a fault.

Do not keep testing until a preferred answer appears. Fix the primary outcome, the comparison and the stopping rule before the test starts. Check the lead capture forms route for consent wording before you change a form.

Before you act

Write the decision and its owner before choosing a measure.
Complete a metric card for every reported value.
Freeze and version each metric definition with its report.
Map every source and join before building the dashboard.
Inventory cookies, pixels and scripts, and check consent for each.
Record corrections with old value, new value, reason and approver.

Common questions

Why is a form submission not a qualified lead?

A qualified lead is a buyer-defined status with its own eligibility rule, reviewer and evidence time. A submission only meets the validation and deduplication rules. A missing review is not a failed lead unless the metric card says so.

Does analytics measurement always need consent?

No, but the exception is narrow. Aggregate statistics about how your service is used, for improving it, can fall within the statistical purposes exception. Online advertising purposes, including ad measurement, need consent.

When can a result be called causal?

Only with a credible comparison. Record the assignment method, the populations and the uncertainty. If no credible comparison exists, report observation or attribution instead.