Branding agencies service standards: how to define them

Do not borrow a turnaround promise from another agency or publish an unsourced target. Define each service measure from the purchased work: eligible event, unit, clock, evidence, tolerance, response and remedy.

Targets are deliberately left blank in this guide until both parties agree on them, because a defensible figure depends on the purchased work, the evidence system and the risk the buyer carries. A generic published number would be unsourced.

This guide addresses private engagements delivered to organisations in England. The government sources below are operational references, not standards imposed on branding suppliers. No service records or contracts were tested.

What to take away

Start with the decision the measure controls

A measure needs a consequence. "Initial response time" might help the engagement owner detect an unattended request. "Accepted-output rate" might reveal a recurring briefing or production defect. Neither proves creative quality or business effect.

Government performance guidance recommends beginning with service purpose, forming hypotheses, finding data sources and giving measures context. Apply that sequence to the actual engagement — the pitch, the creative rounds and the asset handover — rather than treating its public-service examples as agency benchmarks.

Complete the definition card

For every proposed standard, fill in:

Worked example: initial response time. The table fills every field except the target, which stays blank until agreement.

A time measure requires both endpoints. An acceptance percentage needs all eligible submissions in its denominator, not only approved work. If the buyer delays an input, pause the clock only under an agreed rule and preserve the evidence.

Separate four kinds of promise

Availability covers whether the agreed route can be used. Responsiveness records when an eligible request is acknowledged or acted upon. Delivery tracks a defined output reaching its contracted state. Quality records observable acceptance criteria and defects.

Keep them apart. A prompt acknowledgement cannot compensate for a rights failure. An on-time delivery does not prove accessibility. A low defect count may reflect incomplete reporting rather than sound production.

The Cabinet Office's contract-management principles support outcome ownership, controlled change and performance reporting for government contracts. In a private agreement governed by the law of England and Wales, the contract terms and any governing law clause decide the definitions, authority and enforceability. Where the client is an organisation, the Supply of Goods and Services Act 1982 implies a term that the service be carried out with reasonable care and skill; where the client is a consumer, the Consumer Rights Act 2015 does the same. Named commercial and legal reviewers must approve them.

There is no ombudsman for branding services in England, so agree the dispute route in the contract: a named contact, an escalation step, and mediation before court proceedings if the disagreement continues. Membership of the Design Business Association, the IPA or ISBA is worth checking, because those bodies publish codes of conduct and professional standards that their members sign up to.

Agree rights, assets and revision rounds

Copyright in a brand identity stays with the agency that created it unless the contract assigns it, and under the Copyright, Designs and Patents Act 1988 that assignment must be in writing. Name what transfers, when it transfers and in what format: logo files, fonts, source artwork, colour and type specifications and the brand guidelines.

Agree how many revision rounds the fee includes and what counts as acceptance before the work starts. Acceptance criteria should be observable, such as whether the mark holds up at small sizes or whether the palette reproduces in print and on screen. The client's written approval of a round is the evidence the delivery measure records.

Add failure and recovery rules

Specify what enters the exception queue, who triages it and when release stops. The response may be correction, rework, withdrawal, service credit, another contractual remedy or no remedy, depending on agreed terms. Do not imply that a particular result follows automatically.

An illustrative tolerance and remedy pattern, not a default: sample 10 outputs each quarter. A breach count of 2 or more requires correction within 5 working days. A third breach in the same contract year triggers a review meeting.

For an inaccessible output, the standard should name the sampled artefact, criterion, test method and retest evidence. W3C's evaluation report template includes the evaluation scope, reviewers, process, results and recommended action. It is a reporting aid, not proof of conformance.

Record corrections as a separate measure when they matter to the contract. Define whether the count covers reported defects, accepted defects or corrected items, since each answers a different operational question.

Check definition drift before comparing periods

At each agreed interval, inspect missing records, paused clocks, exclusions, disputed states and changes to the work. A definition change can break comparison with earlier periods. Start a new series or transparently restate prior figures rather than silently moving the denominator.

Set a stop rule for unreliable evidence. Suspend the affected metric when events cannot be reconstructed, sampling has changed without record or the source system loses required fields. Resume only after the owner documents the repair and a reviewer confirms comparability.

Contract the blank, then test it

The useful template provides the fields. Populate it with the parties, scope, calendar, timezone, evidence system, target, tolerance and remedy from the signed arrangement. Keep the evidence in a ticketing system such as Zendesk or Freshdesk, so each event carries a timestamp and an owner. Run synthetic cases for an ordinary request, missing client input, disputed acceptance, inaccessible file and withdrawal. Record what was actually tested and leave every untested path open.

Worked synthetic case: a written request arrives at 16:55 on a working day. Expected result: acknowledgement logged before 17:00. Recorded evidence: ticket timestamp, inbox rule screenshot and signed test note.

Before you act

Common questions

What should a time measure include?

A time measure requires both endpoints, so the definition card must state clock start, clock stop, pauses and timezone. If the buyer delays an input, pause the clock only under an agreed rule and preserve the evidence for later review.