After hours and overflow as the first use case

Why is after hours the honest first use case?

Because the alternative is nothing. An out-of-hours call today reaches a voicemail most callers will not use, or an unanswered ring, and the caller rings the next business in the search results. Any structured message beats that, which means the bar for success is low enough that ordinary performance is a clear gain rather than a contested one.

It also contains the damage of a wrong answer. Errors in the evening are discovered in the morning by your own team, before the customer is affected, and you can correct them with a phone call. The same error at midday goes straight into a conversation someone is having with a customer while your staff are busy.

And it gives you a clean measurement. Every out-of-hours call the agent handles is incremental by definition, because nobody was handling it. You do not have to argue about whether the agent did better than a human would have, which is the argument that stalls internal approval for daytime use.

The condition to check first is whether your out-of-hours calls are worth answering at all. For trades, clinics, lettings and anything where the caller has an immediate need, they are the most valuable calls of the day, because whoever answers wins the job. For businesses whose evening traffic is mostly wrong numbers and cold sales approaches, the volume report will say so plainly, and the pilot belongs at lunchtime instead.

What counts as overflow, and how do you detect it?

Overflow is any call your people were not going to reach in time. Mechanically there are four detectable cases: nobody answers within a set number of rings, every line or agent is busy, the queue is longer than a threshold you set, and the call arrives inside an exception such as a training day or a holiday. Each is a different setting, and most phone systems can act on all four.

The setting that matters most is the ring timeout, because it decides who gets the agent. Too long and callers hang up before the overflow triggers, too short and calls are diverted while someone was reaching for the handset. Somewhere around fifteen to twenty seconds is a common starting point, and you should tune it against your own abandonment data rather than accept a default.

Get the detection from your phone system rather than guessing. Ask your provider for a report of unanswered calls, calls answered after a long ring, and calls arriving outside opening hours, by hour of day. That report is both the case for the project and the configuration for it, and most businesses have never requested it.

What should it do at 2am that it should not do at 2pm?

Say plainly that the office is closed and what will happen next, then commit to a specific time rather than a vague one. Tomorrow morning is weak. Someone will call you before ten is a commitment the caller can act on, and it stops them ringing three competitors while they wait. Only make it if your team will honour it.

It should also be more willing to take a message and less willing to attempt an answer, because there is no one to correct a mistake for eight hours. Narrow the agent's scope out of hours: capture, book if the calendar allows it, answer only fixed facts, and hand anything else to a message with a promised callback.

The emergency path is the exception, and it needs deciding before you go live rather than after the first incident. If some out-of-hours calls are genuinely urgent, a burst pipe, a locked-out tenant, a failing system, then the agent needs a defined test for urgency and a real number to reach, and it must state clearly what it cannot do. An agent that takes a message about a flood is worse than a recording that gives the on-call number.

What should the morning handover look like?

One list, in the place your team already opens first, ordered by urgency rather than by time received. If they use a shared inbox, it goes there; if they use a CRM task list, it goes there. What does not work is a separate dashboard someone has to remember to check, because for the first week they will and after that they will not.

Each item needs four things: what the caller wants, the number, when they were promised a call, and whether the agent already made a commitment on your behalf. That last field is the one most systems omit and the one that causes trouble, because a caller who was promised a call before ten and hears nothing until three is more annoyed than if they had reached voicemail.

Then close the loop visibly. Mark items as called back, and count the ones that were not. An overnight capture rate of a hundred per cent with a morning callback rate of sixty is not a working system, it is a queue of disappointed people, and the number nobody measures is the second one.

When should you extend it into working hours?

When two conditions hold. Your morning callback rate is high, meaning the promises made overnight are being kept, and your review of out-of-hours transcripts has stopped finding wrong answers. If either is failing, extending the hours multiplies a problem rather than the benefit.

Extend by mechanism rather than by clock. The next step after out of hours is overflow during hours: the agent answers only when nobody picked up within the ring timeout, which means it never competes with an available human. That covers lunch, busy periods and the twenty minutes when everyone is on a call, and it is invisible when your team is keeping up.

Only then consider having it answer first during working hours, and treat that as a separate decision with its own justification. There are businesses where it is right, generally where call volume is high and repetitive, and businesses where it is a poor trade because callers value reaching a person. Answering first is a positioning choice, not a configuration change.

What should you set up in the first week?

Four things, in order. Get the unanswered and out-of-hours call report from your provider, so you have a baseline that predates the agent. Point out-of-hours calls at the agent and leave working hours untouched. Set one promise the agent is allowed to make and make sure your team knows about it. And book half an hour on Friday to read every transcript from the week.

That last one is the step people skip and the one that determines whether the pilot tells you anything. Five evenings of calls is a small enough volume to read in full, and reading them in full is how you find the questions you did not anticipate, which is the actual output of a pilot.

Then run one deliberate test before you trust it: ring your own number at nine in the evening, ask for something slightly unusual, and see whether the promise made to you is kept the next morning. If it is not, the problem is not the agent.

Keep the pilot to a fortnight and decide at the end of it rather than letting it drift. Two weeks covers a full pattern of evenings and a weekend, and it is short enough that nobody has built a process on top of something you might switch off. Write down before you start what result would make you stop, because a pilot without that sentence in it always concludes that more time is needed.