Measurement
Part of Measuring one customer data journey without pretending correlation is cause
Screening customer data platform benchmark claims before you trust a number
Screen CDP benchmark claims through a reproducible protocol for population, identity rules, denominators, uncertainty and funding, with no invented England result.
No directly comparable authoritative England or UK benchmark was found for the defined customer data platform journey in sources reviewed on 6 September 2026. The defensible result is "not established", not a number borrowed from broad software research or a supplier customer average.
This protocol screens candidate studies for one journey: authorised account events, documented identity decisions, correction and suppression states, and an approved audience export. It does not assume that CRM, warehouses, identity resolution, preferences, tags, analytics, activation, clean rooms and managed integration measure the same job.
What to take away
- No directly comparable authoritative England or UK benchmark was found for the defined customer data platform journey.
- The defensible result is not established, not a number borrowed from broad software research or a supplier customer average.
- Matching a metric name is insufficient because official statistics quality requires fitness for purpose, including relevance, accuracy, timeliness and comparability.
- Broad counts of businesses, software use or advertising activity do not measure a CDP journey.
- When no candidate passes, establish a reproducible internal baseline with frozen protocol and query version.
Write the benchmark question precisely
Name the buyer decision and metric before searching. Specify the event, eligible population, numerator, denominator or unit, geography, reporting period, Europe/London timezone, identity rule, late and missing treatment, exclusions and correction state. Leave the target blank.
Specify the benchmark question
- Name buyer decision and metric
- Specify event and eligible population
- Define numerator, denominator or unit
- State geography and reporting period
- Set Europe/London timezone
- State identity rule and exclusions
- Leave the target blank
The ONS defines official-statistics quality through fitness for purpose, including relevance, accuracy, timeliness and comparability on its quality page. This CDP research is not an official statistic, but those dimensions explain why matching a metric name is insufficient.
Admit sources through a fixed screen
Record the following for every candidate source:
Admission screen for candidate sources
- Journey and CDP function
- Geography and field period
- Population and sampling frame
- Numerator, denominator and unit
- Identity and deduplication
- Missing and corrected data
- Model or query version
- Weighting and uncertainty
The UK Statistics Authority's Code of Practice for Statistics, edition 3.0 applies to official statistics and calls for suitable sources, rigorous methods and openness about quality. Use it as a screening discipline, not as a badge for commercial research.
Reject seductive but incompatible numbers
Broad business, software or advertising counts do not measure a CDP journey. A supplier aggregate may omit customers that stopped using the product, apply a proprietary identity model, combine industries or use a global population.
Publication bias may also favour unusually positive cases. Mark these sources 'contextual only' or 'reject' them rather than adjusting them into an England result.
Do not assume missing events are random. Refused device tracking, unlinked identities, corrections and suppression can change who appears in a dataset. The ICO's data minimisation guidance also requires the buyer to consider what personal information is necessary for the stated purpose. More fields do not automatically produce a better benchmark.
Design a buyer study instead
When no candidate passes, establish a reproducible internal baseline. Freeze the protocol and query version, use authorised records, report all eligible states, preserve identity uncertainty and publish corrections. Repeat the same definition across buyer-chosen periods only after checking that sources, rules and populations remain comparable.
If the buyer wants a causal performance claim, a descriptive benchmark is the wrong design. HM Treasury's updated Magenta Book separates monitoring from impact evaluation and discusses counterfactual approaches. A qualified evaluator should determine whether an experiment or quasi-experiment is possible.
The evidence record for this article therefore remains blank: no admitted benchmark source, value or threshold. The statistician should rerun the search and admission screen before publication, document exclusions, and keep "not established" unless comparable evidence actually passes.
Before you act
- Name the buyer decision and metric before searching.
- Specify event, population, numerator, denominator, geography, period and identity rule.
- Record admission fields for every candidate source.
- Reject incompatible numbers or mark them contextual only.
- Do not assume missing events are random.
- Freeze protocol and query version for internal baseline.
Common questions
Why is a supplier customer average not a valid benchmark for an England CDP journey?
A supplier aggregate may omit customers that stopped using the product, apply a proprietary identity model, combine industries or use a global population. Publication bias may also favour unusually positive cases. Such sources should be marked contextual only or rejected rather than adjusted into an England result.
What should a buyer do when no candidate benchmark passes the admission screen?
Establish a reproducible internal baseline. Freeze the protocol and query version, use authorised records, report all eligible states, preserve identity uncertainty and publish corrections. Repeat the same definition across buyer-chosen periods only after checking that sources, rules and populations remain comparable.
When is a descriptive benchmark the wrong design for a buyer?
If the buyer wants a causal performance claim, a descriptive benchmark is the wrong design. HM Treasury's Magenta Book separates monitoring from impact evaluation and discusses counterfactual approaches. A qualified evaluator should determine whether an experiment or quasi-experiment is possible.
