scoco.ai
Category & features

Emergency detection on the phone: how AI spots urgency

An AI phone assistant spots emergencies from word choice plus context: terms like "burst water pipe", "heating failure" or "power cut", together with the situation described, signal an urgency the business defined in advance. Recognising one, it puts the caller through to a stored emergency number or reports the request immediately — instead of leaving it in the ordinary inbox.

How the system spots urgency

Detection does not rest on a single source of signal but on the interplay of several cues in the conversation. The most important signal is the choice of words: a language model understands that "there's water standing in my cellar" is a different request from "I'd like a quote for a new bathroom" — even when the word "emergency" is never said. Then there is context: what is affected, since when, is it getting worse, is anyone at risk? A good system asks targeted follow-up questions when the description stays unclear. The decisive point: the classification only becomes reliable through rules the business stores — what counts as an emergency for a plumbing business does not for a law firm.

The business defines what counts as an emergency

Emergency detection does not begin with the AI but with a business decision: which requests justify disturbing someone immediately? That definition is stored in the assistant's configuration, at scoco.ai as a rule — typically as a list of concrete cases ("burst pipe, heating completely down in winter, power out across the whole building") plus counter-examples ("a dripping tap is not an emergency"). The more concrete the list, the more accurate the detection. Part of this is configuring the emergency numbers: which number is dialled when — the on-call service, the master craftsman, staggered by time of day? Businesses with an on-call service should run through this whole chain once before going live; more on that in "Automating the on-call phone".

Putting through or an immediate alert — two response paths

There are two sensible responses to a recognised emergency, and they differ in the speed they promise. Putting the call through connects the caller straight to the stored emergency number — the conversation switches live from the assistant to a human. That is the right route when an on-call service really can be reached. The immediate alert takes the request down in structured form and notifies the business at once, for instance by SMS or email flagged as urgent — the callback then comes from a human. It is the right route when nobody can be guaranteed to pick up: a caller who is first put through and then lands in voicemail after all is more thoroughly lost than one who is promised a fast callback and gets it. Technically, good systems handle both routes; the choice is a question of organisation. How putting a caller through fits into the chain of the conversation is shown in "How does an AI phone assistant work?".

The honest trade-off: false alarm against missed emergency

Every emergency detection balances two kinds of error, and no configuration eliminates both. If the system puts calls through too generously, the on-call staff get woken at night for dripping taps — after the third false alarm they stop taking a put-through call seriously. If the threshold is too strict, a real emergency lands in the inbox as an ordinary message and waits until morning. The honest recommendation: when in doubt, put the call through generously and sharpen the rules against the real cases — an unnecessary call costs sleep, a missed burst pipe costs a customer and in the worst case more. And one boundary holds throughout: a business assistant is no emergency service and no substitute for an emergency number. Where there is fire or a risk to life or limb, the call belongs with 112. Anyone who smells gas should get out of the building and only call the gas network operator's fault line from outside (as the Stadtwerke Kaiserslautern advise, for example); if anyone is in danger, the call belongs with 112. Whether an assistant points callers in danger to this of its own accord depends on how it is set up — check it rather than assume it.

What of this is standard today

How emergency put-through can be configured differs from provider to provider. At scoco.ai, emergency put-through with an announcement is part of what is delivered. What counts as urgent is set by the business as a rule: the condition is the request, described in its own words, the action is putting the caller straight through to a named transfer target such as the on-call service; some industry templates already bring suitable rules with them, the one for trade businesses for a burst pipe, a heating failure in the cold, a gas smell and a power cut, for instance. The call is only put through if a target is stored; without one, the assistant takes an urgent message. If nobody answers at the target, the assistant takes a message with a callback number that is marked as urgent in the inbox, and the team gets an e-mail and — where a mobile number is stored — a text message, even when the plan's text messages are used up. Calling several targets one after another cannot be set up at present, nor can an emergency target that changes with the time of day. And scoco.ai is no emergency service: the template for property managers tells callers in danger to get to safety and dial 112; the template for trade businesses puts the call through on a gas smell, for instance, and first tells a caller who reports a smell of gas, fire, smoke or a beeping carbon monoxide alarm to leave the building and call 112 from outside — nothing more. So add the rest of the advice yourself, in the greeting or the knowledge base: scoco.ai's terms oblige the customer to set the service up so that callers in urgent cases are referred to suitable places, such as 112, the out-of-hours medical service on 116 117 or its own emergency line (AGB § 7 (4a)). The advice also belongs wherever your customers find your emergency number, on your website for instance. After every call, the entry in the inbox shows how the call ended — put through, transfer failed or a message, marked as urgent where applicable — and the transcript shows why; so the threshold can be readjusted against real conversations.

Frequently asked questions

Does the AI detect every emergency reliably?

No, there is no hundred per cent hit rate — not in software and not with people on the phone. The more concretely the business defines what is urgent, the better the detection. The classification can be sharpened continuously against the transcripts.

What happens if the emergency number does not pick up?

Good systems then take the request down with an urgency note and notify the business at once; some try another number first. At scoco.ai the assistant takes an urgent message with a callback number, and the team gets an e-mail and — where a mobile number is stored — a text message; it does not try another number at present. Ask every provider specifically about this escalation chain.

Can I store different emergency numbers for different times?

Some systems can — the workshop during the day, the on-call mobile at night, for instance. At scoco.ai an emergency target that changes with the time of day cannot be set up at present. Check this with every provider during the trial before you build your on-call service on it.

Does this replace the emergency number or the out-of-hours medical service?

No. Where life or limb is at risk, calls belong with 112; the out-of-hours medical service on 116 117 helps with complaints outside surgery hours that are not life-threatening but cannot wait until the practice next opens. A business assistant's emergency detection handles operational urgency — a burst pipe, a heating failure, a fault.

Status & sources

As of 2026-11-15.

Further reading

From now on, someone
always picks up.

Call our demo business and hear for yourself what your reception sounds like.

We only use your address for your access. Privacy

4-week free trial. Or call the demo right now: +49 621 490799-22