Foundations
Four evidence designs for privacy first measurement in England, and what each gives up
Match one England measurement decision to an evidence design: aggregate statistics, consented observation, modelling or controlled evaluation, and its cost.
What to take away
- Choose the evidence design after naming the decision, not before.
- Purpose-limited aggregate statistics suit a provider's own service and exclude online advertising.
- Every design gives something up, so disclose what, such as the people who decline or the model's uncertainty.
- No design fits when there is no decision owner, no access to evidence or a demand for a legal guarantee.
- Commission a paid diagnostic only if you can name the report the decision owner will act on.
Name the decision before the design
A design only makes sense for a named decision, such as whether to continue a campaign. Without one, the choice of approach is guesswork. Define the work in four parts before comparing anything.
What the 2026 rule permits
- A named decision, such as whether to continue a campaign.
- An evidence design that separates necessary observations from extras.
- Controls for purpose, choice, access, retention and suppliers.
- Reporting that labels facts, calculated values, model estimates and gaps.
The designs below are not rungs on one maturity ladder. Different designs preserve different information, so the right one depends on what the decision needs, not on how advanced the method sounds.
Compare four evidence designs
Each design answers a different decision need, and each sacrifices something. The table sets the trade-off beside what the report must disclose.
Evidence designs and trade-offs
Approach
- Improve the provider's own service
- Purpose-limited aggregate statistics
- Measure consented journeys
- Observation after valid choice
- Fill gaps in platform reporting
- Consent-aware modelling
- Test whether an intervention caused change
- Controlled evaluation where feasible
Evidence given up
- Improve the provider's own service
- Individual paths after aggregation
- Measure consented journeys
- Activity of people who decline
- Fill gaps in platform reporting
- Direct observation of estimated activity
- Test whether an intervention caused change
- Some speed and simplicity
Disclose
- Improve the provider's own service
- Exception conditions and objection route
- Measure consented journeys
- Consent scope and denominator
- Fill gaps in platform reporting
- Model status, eligibility, uncertainty
- Test whether an intervention caused change
- Assignment method, spillover, sample limits
Evidence methods by decision need
Decision need
- Improve own service
- Aggregate statistics
- Measure consented journeys
- Identifier observation
- Estimate platform gaps
- Consent-aware modelling
- Test intervention effect
- Controlled evaluation
Approach
- Improve own service
- Individual paths
- Measure consented journeys
- Decliner activity
- Estimate platform gaps
- Direct observation
- Test intervention effect
- Speed, simplicity
Sacrificed
- Improve own service
- Measure consented journeys
- Estimate platform gaps
- Test intervention effect
Observation after valid choice measures consented journeys. The cost is the activity of people who decline, so the report must state the consent scope and the denominator. A consented group should never be presented as everyone.
Consent-aware modelling can fill gaps in platform reporting. Google's Consent Mode documentation says modelling may address collection gaps, and that is product material rather than regulatory approval. Report the model's status, eligibility and uncertainty, because the estimated activity was never directly observed.
A controlled evaluation tests whether an intervention caused a change, where that is feasible. It costs some speed and simplicity. Disclose the assignment method, spillover and sample limits so a reader can judge the result.
Check the rule behind the aggregate option
The aggregate option rests on a narrow rule. A provider may store or access information to produce aggregate statistics for improving its own service, subject to sole-purpose, information, objection and aggregation conditions. Online advertising sits outside that exception.
The ICO's exceptions page says the statistical purposes exception applies when the sole purpose of the storage or access is to collect information for statistical purposes about the use of the service. Treat every other purpose as outside it.
The Data (Use and Access) Act 2025 changed the underlying PEC Regulations, and the ICO's finalised storage and access guidance followed its consultation on those changes. Read the rules and compliance guide before relying on any one exception. A published factsheet is reference material, not legal advice.
Know when no design fits
Disqualify early when there is no decision owner, no access to evidence or a demand for a legal guarantee. Any one of the three means the brief is not ready, whichever design a supplier proposes.
Interview the decision owner, the practitioner who maintains collection and the privacy owner who can stop release. Three situations are worth testing:
Who buys, what disqualifies
- A service owner cannot show that each collection purpose stays in boundary.
- A marketing team needs consent gaps reported.
- An analytics lead suspects tags or retention settings have drifted.
Design selection is not the only check. Measurement vendor due diligence covers the contracting party, sub-processors and verified deletion before you commit budget.
Commission the chosen design
Three scope bands let buyers compare quotes like for like, whichever design was chosen.
- Diagnosticone decision, one data-flow drawing, one limitations note.
- Implementationa defined change to collection or reporting, with an acceptance script.
- Assurancea recurring check with response times and named evidence artefacts.
Prices were not verified, so ask for a fixed-scope figure against band one. Weigh artefacts delivered, not hours billed. A pilot needs four artefacts: a decision statement, a purpose register, an acceptance script and a limitations note.
Set review points for the route to 2027
Set review points rather than one date. The go or no-go test fits in one line: can you name the decision owner, the report they will act on and its review date? A yes justifies a paid diagnostic, and a no means the brief is not ready.
- Quarterly
- re-read platform consent and retention documentation.
- Before 2027
- require a dated change log for consent and deletion behaviour.
- When the ICO guidance changes
- re-check its exceptions page.
The next cycle's signals sit in measurement signals for 2027, which sorts each analytics measure against the exception.
Common questions
Which design suits improving my own service?
Purpose-limited aggregate statistics, subject to sole-purpose, information, objection and aggregation conditions. Online advertising falls outside that exception, so any advertising measure needs a different design.
Does Google's modelling documentation approve a design?
No. Google's Consent Mode documentation says modelling may address collection gaps, and that is product material rather than regulatory approval.
What is the smallest sensible first purchase?
One diagnostic covering a single decision, a data-flow drawing and a limitations note, priced as fixed scope.
In this guide
- Privacy first measurement market size in England: a scenarioPrivacy first measurement market size in England has no published revenue total; this is an illustrative scenario built on placeholders, not verified data.
- Five demand signals for privacy first measurement, read from regulator and platform recordsFive privacy first measurement demand signals for England, ranked from dated regulator, audience and platform records with clear research limits.
- Privacy first measurement business models compared for England buyersCompare privacy first measurement business models by buyer decision, delivery duty, recurring work and evidence risk before choosing an England offer.
- Before launching a privacy measurement offer, test the buyer and the stop conditionsUse this privacy first measurement market entry checklist to test an England offer's buyer, data flow, evidence, security, claims and stop conditions.
- Five privacy first measurement opportunity areas for England buyersLists five privacy first measurement opportunity areas in England, from purpose separation to embedded analytics governance, with evidence limits and buyer tests.
