Working hours and SLA: make promises you can keep

By the Onsites AI team · Last updated August 28, 2026 · 4-minute read

Nothing buys back customer goodwill like a stated response window, and nothing erodes it faster than a window the team cannot keep. Working hours and an SLA are how a desk goes from "we answer when we can" to "we answer by when we promised" — and the moment you state them, the queue gains something precious: the ability to tell you it is slipping before customers do. This guide sets both up in your free workspace and keeps the promises honest.

Hours first: what you will actually staff

Start from capacity, not ambition. If one person covers the desk alongside other work, a realistic promise is same business day, Monday to Friday. Two to three people sharing the queue staffs 09:00–18:00 with a few hours of cover each day. An "24/7" promise with a three-seat team is a lie you will apologize for by Wednesday.

Set the window in Settings, in your display timezone, and — this is the part teams miss — write the outside-of-hours sentence customers actually read: "We reply within {one business day}; messages after 18:00, on weekends and on {holidays} join the queue for the next working morning." A weekend inquiry should never expect a Saturday reply; it should expect a Monday 09:30 one, stated on the widget. Expectation set at the moment of writing is the single cheapest satisfaction win available.

Holidays and absences get their own line: a public-holiday note and an out-of-office template from the saved replies library turn "silence" into "informed absence", which customers accept remarkably well compared to the alternative — an auto-reply saying "we're on it" from a mailbox nobody reads.

The SLA: one number you can move

Resist the corporate reflex to write a four-tier SLA with twelve response codes. A small team needs one number: first reply inside N working hours. Pick N by measuring, not hoping: look at your median first response over the last two weeks and set the SLA a notch below it. A desk whose median is six hours should not promise four; a desk at two hours can promise four and beat it comfortably. The published guidance — what "fast" means per channel in practice — is in the first-response benchmarks guide.

The second number, only if the team is two or more: high priority inside one working hour. That tier only works if "high" remains rare and honestly defined — the priority definitions guide keeps it from inflating, and inflation is the failure mode that kills SLAs not by breach but by meaninglessness.

Mondays, and the arithmetic of peaks

Volume is not flat. Weekend accumulation plus Monday morning is typically the double-peak of every small desk: the queue you could clear by noon on Tuesday is two days of work at 10:00 on Monday. Three mitigations, in order of cheapness:

• Monday triage — clear "high priority" and unclaimed-invoice questions before anything else; the queue views make this a ten-minute ritual instead of an archaeology session.
• Stagger the window: one person starts at 08:00, the other closes the day — coverage without overtime, and the SLA gets two fresh hands a day.
• Deflect the repeatable: the twenty saved replies from the template library and, when you want automation, an AI draft at 3 credits per reply absorb the predictable share of the peak. When AI does the drafting, the Monday peak measures people, not typing speed.

Time zones and distributed teams

Two people in different time zones can double the coverage without doubling the headcount — on paper. In practice, distributed desks fail on handoff, not on hours: the thread that crosses midnight needs a note saying what it waits on and who owns it next. Three rules keep it honest. Every working day ends with the end-of-day sweep from the shared inbox guide. Nobody starts a conversation they cannot finish inside their own window. And the follow-the-sun guide's handoff summary — three lines, always in the thread, never in a side chat — is the difference between a relay and a drop.

If the team spans four or more time zones, per-channel hours start to matter: chat widget promises during one coverage block, email within one business day, and messaging channels set to explicit "away" hours rather than a silence that reads as neglect.

When the SLA breaches: the honest mechanics

An SLA that never breaches is either superb or fictional, and staffed reality includes sick days, launches and the printer room's ability to attract disasters. What separates teams that keep customers through breaches from teams that lose them: the breach gets acknowledged before it is discovered. When the queue shows an old conversation, the reply is "this took longer than our {N}-hour promise — here's where it stands and here's what I'm doing now." Customers forgive a stated window that slipped once; they do not forgive a window that never existed, which is exactly what an unstated SLA is.

Watch the two numbers monthly — median first response and the worst-decile evening (the number your worst customer experienced) — via the queue's views; if the worst decile drifts past the promise, either capacity or the promise moves, and moving the promise is a legitimate decision when made deliberately. There is no shame in widening a window you cannot staff; there is permanent damage in pretending one you missed.

Where this goes next

Hours and an SLA are the desk's promise; the metrics are how you keep it. The KPI guide for small teams picks the numbers worth watching, and the FCR guide covers the metric that buys time back rather than spending it. Free workspaces set all of this in minutes, and when compliance or data residency moves the goalposts, the same desk runs self-hosted — same SLA mechanics, your servers, your hours enforced by your own infrastructure. The promise is the product here: state it, staff it, and let the queue tell the truth about it.