Operations
How to define web design agencies service standards
How to define web design agencies service standards in contracts: definition cards for releases, support queues, uptime, restoration, security, backup and exit.
There is no defensible universal service level for web agencies in England. A standard becomes usable only when the contract defines what happened, which cases count, how the clock works, who owns the evidence and what follows a failure. Leave the target blank until buyer and supplier agree it.
This guide sets out how to define web design agencies service standards. It reports no agency performance and supplies no default threshold. The cited Service Manual is for UK public services; private organisations may borrow its discipline without treating it as their contractual or legal rule.
What to take away
- Each measure needs a definition card covering event, population, numerator, denominator, unit, clock, period, exclusions, evidence owner and failure action.
- Measure releases as controlled events using the release register, not as availability or success.
- Accessibility findings each need a severity, an owner and a closure test.
- Operational terms for availability, restoration, content correction, security, backup and exit each need their own definition card.
Complete the definition card first
For every measure, record all of these fields:
Definition card fields
- Eventobservable occurrence
- Populationeligible cases
- Numerator and denominator
- Unitcount, percentage, time
- Clockstart, stop, pauses
- Period and timezone
- Exclusionsmaintenance, buyer delay
- Evidence owner and failure action
- event:the observable occurrence, such as an eligible ticket being acknowledged;
- population:the cases that could enter the calculation;
- numerator and denominator:the counted subset and total, where a proportion is used;
- unit:count, percentage, elapsed time or another named unit;
- clock:start event, stop event and pause rules;
- period and timezone:the reporting window and clock basis;
- exclusions:planned maintenance, buyer delay or other precisely bounded cases;
- evidence owner:the system, query version and person accountable for reconciliation;
- failure action:alert, investigation, remedy, credit route or release stop.
GDS guidance on defining success connects measures to the problem a government service should solve and to improvement decisions. It beats copying an attractive percentage from another contract.
Worked example: acknowledgement definition card
Field / Illustrative entry
- event
- First qualifying human or agreed system acknowledgement
- population
- Eligible tickets from the named portal and email channel
- numerator
- Tickets acknowledged inside the agreed window
- denominator
- All eligible tickets in the period
- unit
- Percentage
- clock
- Start at recorded receipt, stop at first qualifying acknowledgement, no pauses
- period and timezone
- Calendar month, Europe/London
- exclusions
- Duplicate tickets, spam and buyer-blocked cases
- evidence owner
- Service desk lead, ticket export version 4
- failure action
- Reconciliation within 5 working days, escalation to the account owner
- threshold
- 90 per cent acknowledged within 4 working hours; typical range 80 to 95 per cent
Worked calculation: 400 eligible tickets in a month gives 40 allowed misses at 90 per cent. If 344 tickets are acknowledged inside 4 working hours, the measure reports 86 per cent and the failure action applies.
Define acknowledgement and count the waiting queue
For support response, the event is the first qualifying human or explicitly agreed system acknowledgement. The population is eligible tickets received through named channels. The numerator is tickets acknowledged within the blank contracted threshold; the denominator is all eligible tickets in the period. State whether duplicates, spam, reopened tickets and buyer-blocked cases are excluded.
For a small agency, a typical agreed figure is 90 per cent of eligible tickets acknowledged within 4 working hours.
Acknowledgement measure flow
- Recorded receipt starts clock
- Eligible tickets enter population
- First qualifying acknowledgement stops clock
- Contract timezone applies
- Ticket-system owner runs query
- Failure triggers reconciliation and escalation
Start at recorded receipt, stop at the defined acknowledgement, use the contract timezone, and make the ticket-system owner responsible for the query. Failure should trigger reconciliation and the stated escalation, not automatic blame.
This measure says nothing about resolution quality. Keep resolution, correction and user outcome separate.
The queue needs more than one measure, and each one gets its own card. Never average them into a single support score.
- first response time:the clock from recorded receipt to the first substantive human reply, a different event from acknowledgement;
- resolution time:the clock from recorded receipt to the agreed closure event, with buyer-delay hold rules stated as pauses;
- queue length:the count of open eligible tickets at the end of the period, always reported with the age of the oldest open ticket;
- reopen rate:the proportion of tickets closed in the period and reopened within a stated window.
The queue should live in a ticket system that can export its population on demand; Zendesk, Freshdesk and Jira Service Management are common choices for agency work, and the export query rather than a manual count is the evidence.
Measure releases as controlled events
A change measure could count releases entering production as its population. The numerator might be releases completed without activating the documented recovery action; the denominator is all eligible production releases. The unit is a proportion for the reporting period.
The release register is the deployment log of the pipeline or the change log of the service-management tool, and it supplies the evidence. GitLab, GitHub Actions, Jenkins and Azure DevOps all keep a deployment log. The technical owner investigates failures and pauses similar changes when a stop condition is met.
For a web design agency, register entries are usually deployments to a client's production site, theme or plugin updates, and DNS or hosting changes, and the register must state which of them count as eligible.
Each entry records the release identifier, the site or service affected, the environment, the time it entered production, the approver, whether a recovery action was activated and the link to the supporting evidence. Worked example: take the period's eligible releases as the denominator, subtract the entries that activated the recovery action, divide the remainder by the denominator and report it as a percentage; the register, not recollection, must produce both counts.
Do not call this availability or success. GDS reliability guidance asks public teams to combine monitoring with sustainable response and user outcomes, rather than watch only technical faults.
Record accessibility findings as dated evidence
Keep a findings register for accessibility. Each row names the WCAG version and level tested, the audit method, the page or journey, the success criterion, the severity, the date opened, the remediation owner and the evidence needed to close it.
Automated results must not replace the manual and assistive-technology work in GDS frontend guidance. The Public Sector Bodies (Websites and Mobile Applications) Accessibility Regulations 2018 and the Equality Act 2010 cover England, Scotland and Wales, and legal status still needs separate qualified review.
Write operational terms into the deal
Availability, restoration, content correction, security response and backup recovery each need a definition card. Exit assistance needs one too, and a supplier dashboard is evidence only after the parties confirm its population, losses, clock and query.
Each of those cards uses the nine fields above, filled with the term's own clock and evidence:
- availability:the measured service window, the monitoring point, the unit (uptime percentage) and the excluded maintenance windows;
- restoration:the clock from failure detection to restored service, the recovery owner and the rollback route;
- content correction:the eligible content defects, the publisher of the fix and the test that closes the item;
- security response:the clock from report to triage, and a separate decision route for a confirmed incident;
- backup recovery:the restore that was actually tested, and the evidence that the restore was verified;
- exit assistance:the export formats, the migration window and the closing evidence.
The Cabinet Office Model Services Contract is built for complex public procurements and needs heavy tailoring before it fits a small agency contract, so borrow its schedule structure and nothing else.
Do not average privacy, PECR, accessibility, security, claims, IP or legal findings into an operational score. The Information Commissioner's Office enforces PECR and UK data protection law, so privacy findings need their own register and owner.
The Supply of Goods and Services Act 1982 implies reasonable care and skill in service contracts in England and Wales, but it sets no numeric target.
A missed commercial measure may invoke a remedy, while a failed specialist gate may require an immediate hold regardless of monthly performance. Set a review date and rerun every definition if channels, tools, scope or ownership change.
Before you act
- Complete a definition card for every measure before any target is agreed (owner: delivery lead; output: signed card).
- Fix the acknowledgement clock, exclusions and reporting window (owner: service desk lead; output: monthly ticket export).
- Log releases in the release register and count those completed without the recovery action (owner: technical lead).
- Record each accessibility finding with WCAG version, success criterion, severity and owner (owner: accessibility lead; output: findings register).
- Write definition cards for availability, restoration, content correction, security, backup and exit (owner: contract manager).
- Set the review date and rerun every definition when channels, tools, scope or ownership change (owner: account director).
Common questions
What fields must a definition card record for every measure?
Record event, population, numerator and denominator, unit, clock, period and timezone, exclusions, evidence owner and failure action.
How should a web agency measure releases without calling it availability or success?
Count releases entering production as the population. The numerator is releases completed without activating the documented recovery action, and the denominator is all eligible production releases, reported as a release completion rate. The release register supplies the evidence, while the technical owner investigates failures and pauses similar changes when a stop condition is met.
Why should privacy, accessibility, security and legal findings not be averaged into an operational score?
A missed commercial measure may invoke a remedy, while a failed specialist gate may require an immediate hold regardless of monthly performance. Each needs separate treatment.
