When a commercial app degrades, users complain in reviews and churn quietly. When a citizen portal degrades, people take a day's leave, travel to an office and stand in a queue — often the very people the digital service was meant to spare. That asymmetry is why service-level objectives for citizen-facing systems deserve more rigour, not less, than their commercial cousins get.
Choose SLIs a citizen would recognise
Infrastructure metrics — CPU, memory, pod restarts — tell you the machine's mood, not the citizen's experience. Good service-level indicators map to moments that matter: did the search return? did the payment confirm? did the certificate download? We define three to five journey-level SLIs per service and resist the temptation to measure everything that is measurable.
Availability of the top three citizen journeys, measured end-to-end
Latency at the 95th percentile for search and form submission
Payment round-trip success rate, including the gateway's failures as yours
Freshness of status information — 'submitted' must become 'processed' honestly
Write objectives that respect real usage
Government traffic is brutally seasonal: deadline days for tax, admission windows, scheme launches. An annual 99.9% that quietly spends its whole error budget during the one week citizens actually need the service has met its number and failed its purpose. We set tighter objectives for declared peak windows and demand a documented capacity rehearsal before each one.
# SLO definition — portal search latency
slo: portal-search-latency
service: citizen-portal
objective: 99.0 # of requests under threshold
threshold: 800ms # p95, measured at the edge
window: 28d
peak_windows:
- name: admission-season
objective: 99.5
alerting:
fast_burn: 14.4x budget over 1h # page
slow_burn: 6x budget over 6h # ticketMake the error budget a governance tool
The quiet power of SLOs is the error budget: an agreed, numeric answer to 'how much unreliability is acceptable?'. When the budget is healthy, teams ship. When it is spent, the next sprint goes to reliability work — and that trade is written down where programme boards can see it. This converts the eternal features-versus-stability argument from opinion into arithmetic.
An error budget is the first artefact many programme boards have seen that makes reliability a decision rather than a hope.
Report where citizens can see
The most effective reliability practice we have introduced in public-sector programmes is also the simplest: a public status page and a monthly reliability figure reported alongside programme spend. Sunlight changes incentives. Teams that publish their uptime defend it; departments that see the number fund it. Citizens, for their part, get what they were promised — a service that works on the day they need it.
