Measurement
Web design agencies benchmark research: a method protocol
Use web design agencies benchmark research with strict comparator rules, metric specifications and rejection criteria before calculating any figure.
What to take away
- No England-wide performance benchmark for web design agencies was found in the public sources checked here, and the gap stays visible.
- This page is a method protocol, not a benchmark result. It produces a specification and a search, not a number.
- Named candidates for comparator hunting include ONS statistics, Companies House records, IPA survey work and trade-body member surveys.
- A benchmark is comparable only when every field of the metric specification matches, including period, unit and exclusions.
- Do not pool England, other UK and global figures. Stop when provenance is absent.
The protocol was checked on 6 September 2026. Nothing below is a threshold, a ranking or a market result.
Fix the decision before the metric specification
State the decision first: contract renewal, delivery improvement, service diagnosis or investment appraisal.
Then define the service and the buyer population. "Web project" is too broad if it mixes a brochure redesign, a transactional build, hosting support and accessibility remediation.
Write these fields before collecting anything:
Fix the decision
- Decision, threshold owner and review trigger.
- Service definition and eligible buyer population.
- Numerator event and denominator event.
- Unit, period and timezone.
- Inclusions and exclusions.
- Source system and frozen query version.
- Evidence owner and reconciliation route.
- Missingness, uncertainty and publication-bias treatment.
For a delivery-acceptance share, the numerator is items accepted against predeclared evidence. The denominator is every contractually eligible item. The unit is a percentage.
The period runs 1 April 2026 to 31 March 2027 in Europe/London. Approved scope changes are the only exclusions. The source system is the acceptance register in the project tool, query version v3.2, frozen on 1 April 2026.
Do not summarise that field set in one line. The dictionary entry for every measure explains why a short definition invites mismatched denominators.
Example: a delivery-acceptance benchmark for one England agency
The figures here are illustrative, invented for arithmetic. They are not measured market data and must not be quoted as an England norm.
An agency in Manchester wants an external comparator for its 96 percent acceptance share. Its register holds 100 eligible items for the year. A UK-wide supplier survey claims 92 percent, and an England trade-body survey reports 88 percent for marketing retainers.
Delivery-acceptance benchmark example
| Source | Population and geography | Period | Verdict |
|---|---|---|---|
| Agency acceptance register | One England agency, 100 eligible items | Apr 2026 to Mar 2027 | Own baseline, usable if the definition holds |
| Supplier survey | UK, undisclosed client count | 2025 | Attributed claim, not a comparator |
| Trade-body survey | England, marketing retainers | 2025 | Different service boundary, own row |
Only the agency's own register survives as a baseline. The honest output of a defensible England benchmark today is a baseline, not a market norm.
England source candidates to search first
Search these by name, then check geography, service definition, sample and revision history before treating any figure as comparable:
- ONS business and sector statistics, including UK sector accounts.
- Companies House registration records, described on the Companies House organisation page.
- Published IPA survey work.
- Member surveys from design and digital trade bodies.
A clean register entry is shortlisting evidence only. The three England agencies with verified company records show why, because a registration proves little about delivery performance.
Record publication date, geography, collection method, sample, missing cases and revisions for every source you keep.
Treat evidence classes separately. A supplier case study stays an attributed claim unless its eligible population, measure, method and limitations can be examined.
The Infrastructure and Projects Authority's best-practice guide sets a benchmarking cycle for government projects and shapes the comparator-selection method here, though it holds no private web-agency norms.
Reject false comparability
Do not pool figures when service boundaries, populations, formulas, periods or capture methods differ. Keep England, other UK geographies and global evidence in separate rows.
A larger sample from the wrong population is still the wrong population. Comparability belongs to the specification, not to the sample size.
The 2025 AQuA Book separates checking that analysis follows its specification from checking that it fits its intended use. Both tests apply.
Government Analysis Function advice on communicating uncertainty sets out what must accompany results: sources, methods, revisions and limitations.
Refresh, stop and handover rules
Stop when provenance is absent, definitions cannot be reconciled, or publication bias is material and unassessed. An expired comparator should not stay in a dashboard because it once fitted.
If the eligible observations are defensibly comparable, the calculation can run:
comparator rate = sum of eligible numerator events / sum of eligible denominator events
Do not average percentages with unequal denominators. Report distribution, coverage and uncertainty as the analytical reviewer specifies.
Repeat the search and the method review before any 2027 use, with a named owner and trigger set when the specification is written.
Common questions
Is there an authoritative England-wide benchmark for web design agencies?
No. The sources checked produced none, and a UK-wide statistic or supplier case study cannot be relabelled as an England agency norm.
Which sources can supply eligible comparator data?
Candidates include ONS statistics, Companies House records, IPA survey work and trade-body member surveys. Check geography, service definition and sample first.
Why is a supplier case study not a usable benchmark?
It stays an attributed claim unless its eligible population, measure, method and limitations can be examined. The held agency review shows how that check ends when evidence is missing.
When must I refuse to compare figures?
When service boundaries, populations, formulas, periods or capture methods differ. Also refuse when provenance is absent and definitions cannot be reconciled.
