Why consent changed measurement
For most of the history of web analytics, a tag observed every visitor it could reach. Identifiers were set in cookies by default, every page view and event was recorded, and the main loss was technical: blocked scripts, failed loads, bots. Privacy regulation in many jurisdictions, and the consent bannersSoftware that collects, stores and passes on visitors' consent choices for cookies and tracking. it produced, inverted the default for a large share of visitors. Where consent is required, a tag that sets or reads identifiers must wait for permission, and a meaningful share of visitors do not grant it.
Three consequences follow for anyone planning measurement. First, the population in analytics is no longer the population of visitors; it is the population of visitors who consented, which differs by market, device and audience, and a conversion rate computed on it is a different number from the one the business used to report. Second, the gap is unevenly distributed: markets with stricter rules or more assertive banners lose more, so cross-market comparisons carry a consent effect that has nothing to do with marketing. Third, the tools responded with modelling, so reports now mix observed and estimated figures, and a plan that does not distinguish them cannot say what its numbers mean.
Browser changes compound the regulatory ones — restrictions on third-party cookiesCookies set by a domain other than the site visited, used to recognise a browser across sites. and shortened cookie lifetimes affect visitors who did consent — but the design principle is the same: measurement is now built on a partial view, and the plan has to be honest about which part.
What is and is not collected without consent
What a tag does when consent is declined depends on how the site is configured, and the configuration is a decision the plan should record. The documented consent types for Google tags separate storage for analytics from storage for advertising, and separate the use of data for advertising from personalisation, so a visitor can grant one purpose and refuse another. Under each combination, the behaviour differs:
| Visitor's choice | Analytics observation | Advertising observation | What the reports contain |
|---|---|---|---|
| All consent granted | Full: identifiers set, sessions stitched, events with parameters | Full: click identifiers stored, conversions matched, audiences built | Observed data only for this visitor |
| Analytics granted, advertising declined | Full analytics events | No advertising cookies; conversions not matched to ad clicks; no audience membership | Site behaviour observed; ad platform conversion missing or modelled |
| All declined, basic consent mode | Nothing: tags do not fire until consent | Nothing | This visitor is absent; aggregate modelling may estimate the gap |
| All declined, advanced consent mode | Limited cookieless signals without identifiers | Limited cookieless signals | Aggregate modelling uses the signals; no individual-level data |
| No banner, consent not required in that jurisdiction | Full | Full | Observed data, subject to browser restrictions |
The point of the table is not the mechanics — those are the consent mode guide's job — but what each row means for a number. A conversion rate for a market in the third row is computed on consented sessions only and will usually read higher than the true rate, because consented visitors tend to be more engaged. A channel comparison across the second and first rows understates channels whose visitors decline advertising consent more often. Audience lists for remarketing are built only from the rows where advertising storage was granted, so their size says as much about the banner as about demand.
Non-Google tools behave according to their own documentation, and a plan with several vendors needs the same table per vendor. Where behaviour is undocumented, the plan should say so rather than assume.
Consent mode and modelling
Consent modeA Google tag feature that adjusts what tags collect according to the visitor's consent choices. is Google's documented mechanism for telling its tags which consent the visitor gave. The site sets default consent states before any tag loads, the banner updates them when the visitor chooses, and each tag adjusts its behaviour accordingly. Tag Manager (GTMTag management system.) has built-in support for reading and enforcing these states, and most consent platforms integrate with it.
Modelling is what Google does with the gap. The documentation describes two kinds. Behavioural modelling in GA4Google's event-based analytics platform. estimates the sessions, users and events of visitors who declined analytics storage, using the consented visitors as a reference, and requires the advanced form of consent mode plus eligibility conditions the documentation sets out; when active, the reports blend observed and modelled data, and the interface indicates it. Conversion modelling in the advertising platforms estimates conversions that could not be matched to clicks because advertising storage was declined, so that campaign reports and bidding include an estimate of the unobservable share.
Three things to understand before relying on either. Modelled figures are estimates with uncertainty the reports do not always show; they are only produced where the documented conditions are met, so some properties will see none; and the model cannot estimate what it has never observed, so a market where almost everyone declines gives the model little to work from. A plan that reports modelled numbers labels them as such, records which reports include them, and keeps a view of observed-only data for the decisions where the distinction matters. Treat modelling as a reasonable filling of a known gap, not as the gap having closed.
First-party data and server-side approaches
Two families of technique reduce dependence on browser-side observation. Neither removes the need for consent; both change what can be done with it.
First-party dataData a business collects directly from its own customers, visitors and users. is information the business holds about a person because of a direct relationship: an account, a purchase, a newsletter subscription. Enhanced conversions, documented by Google Ads, send hashed first-party identifiers such as an email address with a conversion so that the platform can match it to a signed-in user's ad interaction where consent allows; the enhanced conversions guide covers the set-up. Meta's Conversions API (Conversions APIServer-side event sharing with Meta.) sends events from the business's server rather than the browser, with first-party identifiers for matching; the Conversions API guide covers it. Both improve matching for visitors who consented and whose browsers would otherwise have lost the link; neither is a way to match visitors who refused advertising use of their data, and the platforms' documentation says consent signals must be passed with the events.
Server-side taggingRunning tags on a server the business controls instead of in the visitor's browser. moves the tag container from the browser to a server the business controls. The browser sends one request to that server, and the server decides which vendors receive what. The documented benefits are control over data sent to third parties, reduced browser load and resilience against some blocking; the server-side tagging basics guide explains the architecture. For consent, the server must receive the visitor's consent state and honour it in exactly the way the browser container would; a server container that forwards everything regardless is a compliance problem relocated, not solved.
The honest framing for a plan: first-party and server-side approaches recover measurement lost to technical causes among consenting visitors, and give the business control and durability. They do not recover measurement from visitors who said no, and a plan that implies otherwise will mislead the people reading it.
Designing measurement under consent
The design question is not "how do we get the numbers back" but "which decisions need which numbers, and what is the most honest source for each". The table sets out the common cases.
| Decision | Preferred basis | Why | Note in the plan |
|---|---|---|---|
| Revenue and orders reported to the business | Order system or CRM | Not dependent on browser consent; complete by definition | Analytics revenue is a diagnostic, not the figure |
| Lead counts and lead quality | CRM, joined to channel where consent and identifiers allow | Complete count; channel view is partial | State the share of leads with a known channel |
| On-site conversion rate for optimisation | Consented sessions, observed only | Comparable over time within one consent regime | Flag banner or rule changes as breaks in the series |
| Channel performance comparison | Analytics with modelling where eligible, read alongside platform figures | Modelling reduces the uneven-loss bias between channels | Label modelled rows; expect disagreement between sources |
| Platform bidding and optimisation | Platform conversions including modelled and first-party matched | What the platform's algorithms actually use | Not the business's reported figure |
| Budget allocation between channels | Incrementality tests or mix modelling, with attribution as a secondary view | Does not depend on individual-level tracking | Schedule tests; see the attribution hub |
| Audience building and remarketing | Consented visitors only | The only lawful population where consent is required | Sizes reflect consent rates; compare across markets with care |
| Cross-market comparison | Metrics normalised for consent rate, or order-system figures | Consent rates differ by market and banner | Report consent rate beside every market metric |
The pattern across the rows is that outcomes move to systems the business controls, behavioural metrics are reported on an explicit population, and decisions about spend lean on methods that do not need to observe every individual. The decision table belongs in the measurement plan, next to the source-of-truth table, and the measurement planning hub shows where it sits in the wider sequence. Some industries narrow the table further: the healthcare provider marketing guide explains why clinics measure booked appointments rather than condition page visits, since data about health is special category data.
A retailer redesigns reporting after a banner change
A European homeware retailer moved to a consent banner with an equally prominent reject option. In the first full month, GA4 users fell by 38% and reported purchases by 41%, while the order system showed orders down 2%. The marketing lead's first instinct was that tracking had broken. The analyst's plan review showed the opposite: the tags were now behaving correctly for the 36% of visitors who declined. The redesign moved revenue reporting to the order system, labelled the analytics conversion rate as "consented sessions" with a note of the 64% consent rate, enabled the advanced consent mode so that behavioural modelling could run once the documented conditions were met, and set the channel comparison to read modelled analytics beside platform figures. The quarterly budget review added a geo-split holdout test for paid social, because the attribution view for that channel had become the least complete of the 5 channels. No tag changed; the plan did.
Documentation and accountability
Measurement under consent is defensible only if it is written down. The plan should hold, in one place: the consent types the site collects and the vendor tags each governs; the consent mode form in use and the date it was set; the behaviour of each tag under each consent state, in a table like the one above; which reports include modelled data; the consent rate by market, reported beside the metrics it affects; and the dates of every banner, rule or configuration change, because each is a break in every time series. The consent banner QA checklist turns the launch of a banner into a gate that includes recording its wording and vendor list on the day it goes live.
Accountability has two sides. Inside the business, a named owner for the consent configuration, who is consulted before any tag or banner change and who reviews the configuration with whoever advises on privacy on a schedule. Outside it, the ability to show — to a regulator, an auditor or a customer — what the site does with a visitor's choice. The documentation that serves the second purpose is the same documentation that keeps the measurement plan honest, which is a reason to maintain it well rather than resent it. Beyond the site's tags, the tutorial on mapping consent across the marketing stack extends the same documentation to the CRM, email, text message and customer list tools, one row per purpose.
Rules vary by jurisdiction and change over time; this guide describes measurement design, not the law. Involve legal or privacy counsel in the design and in every change, and record their advice with the decision it informed. The same purpose and consent questions follow customer data into AI tools, and the routine for minimising data before prompting covers preparing an export so that only what the task needs leaves the system.
When an audit finds tracking that runs before visitors have answered the banner, the troubleshooting page on tags firing before consent traces each early cookie or request to its cause in the container or the page. To keep this documentation current as tools, forms and banners change, the privacy and consent operations playbook lists the changes that reopen the consent map and who checks each one before it goes live. Each row of the consent mapOne table with a row per purpose the stack uses personal data for: where permission is captured, which system holds it and how a withdrawal reaches each tool. keeps its last test result, so those checks leave a record purpose by purpose.
Consent governs what tags may observe; it does not make it acceptable to send a visitor's email address or name in a page address, a search term or an event, and the guide to keeping personal data out of analytics shows how those values slip in and how to sweep for them.
Common mistakes
- Reading a consent change as a tracking bug. A fall in users after a banner change may be the tags behaving correctly; check the consent configuration first.
- Mixing observed and modelled figures without labels. Every report that includes modelling says so, or nobody can tell what moved.
- Comparing markets with different consent rates as if alike. Report the consent rate beside the metric.
- Treating server-side or first-party matching as a way round refusal. They recover technical loss among consenting visitors; they do not observe people who declined.
- Reporting analytics revenue as the business figure. The order system is complete; analytics is a diagnostic view of a consented subset.
- No record of banner and configuration changes. Each is a break in the series; undated breaks become mysteries.
- Designing without privacy advice. The rules differ by place and change; a measurement design that was never checked against them is a liability as well as a plan.
