Measurement
Part of Paid social agencies measurement: sources, attribution and impact
Why ONS, Ofcom and Google Ads sources make poor paid-social benchmarks
Assess proposed paid-social benchmarks for England with a comparability record covering geography, population, definitions, methods and uncertainty.
A report brief asks for one outside number
In this illustrative scenario, a team is preparing an England-focused paid-social report and is asked for an outside benchmark of campaign performance. Three candidates seem promising: ONS business counts, Ofcom online-use research and Google Ads definitions.
The team tests them in turn against the campaign question, then decides whether any can support the requested comparison.
The requested result concerns campaign performance, not the number of businesses, people's general internet use or the way one account records activity. Before looking for figures, the team writes that distinction on a campaign metric card, so its decision criteria are set before a tempting statistic appears.
What each source actually measures
Take a campaign metric such as cost per purchase for a UK paid-social campaign. Here is what each candidate source publishes instead.
- ONS, UK Business: Activity, Size and Location.Counts of enterprises and local units by industry and size. In the industry classification, SIC 2007 code 73.11 covers advertising agencies, so the series can tell you how many enterprises are classed there. It cannot tell you what any campaign cost per purchase.
- Ofcom, Online Nation and Adults' Media Use and Attitudes.Survey-based measures of how people use online services, such as which services they use and how much. They describe people, not the performance of an advert.
- Google Ads definitions.Descriptions of how one product counts an impression, a click or a conversion in an account. They describe how an account records events, not an England distribution of results.
So the ONS series counts firms, Ofcom describes audiences, and Google Ads defines one product's counts. None gives a paid-social cost per purchase to compare against.
Step one: establish what a valid comparison needs
The team accepts a source only if it describes a compatible campaign population, event, numerator, denominator, geography, period and method.
It also records publisher, date, access date, sample, exclusions and uncertainty. A missing field can change what a measure means, so it cannot be dismissed after a convenient number is found.
The team follows this sequence for each candidate:
- State the campaign outcome and population the report needs to compare.
- Check the source's event, numerator, denominator, geography, period and method.
- Record its publisher, date, sample, exclusions and uncertainty where available.
- Reject the candidate if its measure or population does not match the brief.
Comparability fields to capture
- Publisher, release date, reference period, access date
- Geography and assignment rule
- Eligible population, coverage frame, sample size
- Collection method, weighting, quality controls
- Event, numerator, denominator, unit, period, timezone
- Source system, query or table version
- Privacy, suppression, access limits, uncertainty
The figure serves as a checkpoint for all three candidates, not as a shortcut to acceptance. Privacy and client tracking remain separate questions. They may need attention in the work, but neither makes a source describe the campaign population or outcome named in the brief.
Step two: test the ONS candidate
The team rejects the ONS source because it describes UK businesses rather than a sample of paid-social campaigns.
The 2025 edition of ONS UK Business: Activity, Size and Location, accessed on 5 September 2026, reports business counts with regional tables, but its units and activity categories do not provide the campaign outcomes the report is meant to compare.
The edition describes enterprises and local units using a 14 March 2025 snapshot of the Inter-Departmental Business Register and UK SIC 2007 activity categories. Employment information comes from the register and related ONS sources. The landing page does not state one campaign sample size; using a count would require the relevant table notes and quality material.
Those business counts do not isolate commissioned paid-social services or provide impressions, interactions, accepted outcomes, attributed conversions, cost or causal impact. A later 2026 edition has since been published, but the point does not change. Statistical privacy and suppression conditions do not authorise client tracking. The team rules out a performance comparison, while keeping business demography separate from budget evidence.
Step three: apply the same test to Ofcom
The team rejects Ofcom online-use research as an agency-performance benchmark. The Online Nation page was last updated on 10 December 2025, accessed on 5 September 2026, and states a UK geography. Its subject is different from the campaign outcome in the brief.
The underlying research populations and methods vary, so the team would need the precise report, table and methodology before using an observation. At the index-page level, a single observation has no specified sample, exclusions, privacy conditions or uncertainty. The index entry alone therefore does not fully describe a measure for this comparison.
Online use is not exposure to a specified paid advert, an agency-controlled population or an internal business outcome. A UK adult usage statistic cannot supply an England conversion denominator. Differences in samples, devices, age rules and periods also prevent a direct comparison with the campaign metric card.
Step four: keep Google Ads definitions inside their limits
Google Ads can help describe activity inside an account, but its definitions do not provide an independent England benchmark. The team treats the guidance as live product material accessed on 5 September 2026, with a global product record and a population drawn from the selected account and eligible product events.
An impression definition is supplier-specific, while conversion columns depend on account settings and may include modelled values. The definition pages provide no England comparison sample or weighting. For internal account use, the team would record settings, exclusions, privacy conditions, lag, model status and the uncertainty that remains unknown.
Those details do not create an external distribution or uncertainty interval. Event setup, attribution, audience, product, placement, period and data maturity can all differ between accounts. The team records cross-account differences as a comparability failure, rather than presenting account definitions as performance evidence.
Step five: publish the no-benchmark decision
None of the three candidates passes the team's test, so the report publishes no usable benchmark.
It names the failed comparison fields instead of combining business counts, online behaviour and account metrics into a synthetic England figure. The key metrics reference helps the team describe the measures without suggesting they are interchangeable.
To assess its source checks, the team applies the ONS questions of relevance, accuracy, timeliness and comparability from its quality guidance. These questions can help assess a source, but they do not turn commercial account data into official statistics or make unlike sources comparable. The team completes advertising substantiation separately.
Rules to carry into another measurement decision
The case leaves the team with one central rule: define the measure before selecting a source. The guide to measurement and reporting covers the wider evidence layers. A useful comparison needs a compatible campaign population, event, numerator, denominator, period, geography and method.
If a key field is absent or describes something else, the team states the gap rather than trying to repair it with an average.
- Record the publisher, publication date, period and access date.
- Check geography, population, sample size and method.
- Confirm the event, numerator, denominator and period.
- Compare each field with the campaign metric card.
- Reject sources without a compatible campaign population.
- Keep privacy and PECR checks separate from comparability.
A versioned internal baseline may support a future threshold only while its event, denominator, period, geography, query and uncertainty remain stable; it describes that operation alone. Accessibility, regulated-product and finance reviews also remain separate from statistical checks, so passing one kind of review does not make a source suitable for a benchmark.
