How to answer customer calls after hours

This page may contain affiliate links from which Blur Studio earns commissions.

A call that goes to voicemail after hours is usually a job that goes to whichever competitor answers. If you run a field service, property management or similar business, this is a solved problem with three real shapes, not one obvious answer, and the right one depends on what the caller actually needs when they call. Calls missed during the working day are a separate problem, covered in phone goes to voicemail while you are on a job.

Your three real options, from least to most effort

• Voicemail (free): costs nothing and catches nothing. Most callers who hit voicemail after hours hang up and call the next name on the list rather than wait for a callback. Fine only if your after-hours call volume is genuinely tiny. A better greeting helps a little, as what to say on a business voicemail explains.
• A human answering service (Ruby, Smith.ai and similar): a real person answers, follows your script, and can book or triage. Reliable and simple to set up, but it costs real money at real volume, typically a few hundred dollars a month once you're past a bare-minimum call count, and you're paying for every minute whether the call is simple or complex.
• An AI voice agent: answers instantly, every time, following rules you write once. It's genuinely good at the repetitive 80% of after-hours calls (hours, pricing, booking a slot, "can someone come tomorrow"), and it should hand off or take a message for the other 20% rather than pretend to handle an emergency.

Most businesses that try this don't replace a human service entirely on day one. They start with the AI agent catching the routine calls and route anything it can't confidently answer to voicemail or a real person, then tighten the rules once they see what callers actually ask. A wider look at the options and costs is in AI phone answering for a small business.

Where an AI voice agent fits

ElevenLabs' agent platform (it calls this ElevenAgents) builds a phone-answering agent from a written prompt, a chosen voice, and a set of rules for what it can and can't do, then connects it to a phone number. As of today, its plans run Free, Starter, Creator, Pro at $99 a month, Scale at $299, and Business at $990, each with a monthly minutes allowance built in rather than a flat per-minute charge from the first second: roughly 15 minutes on Free up to about 12,375 minutes a month on Business. Once you're past your included minutes, voice usage runs about $0.08 a minute on the paid plans, and the language model powering the conversation is billed separately on top of that, so the real cost per call depends partly on which model you pick. You can compare ElevenLabs' current plans directly before committing to a tier.

Each plan also caps how many calls the agent can handle at the exact same moment: 4 on Free, 6 on Starter, 10 on Creator, 20 on Pro, and 30 on Scale and Business. If you briefly get more simultaneous calls than that, the agent can burst up to roughly three times its normal limit, with those extra calls billed at double the usual per-minute rate. For a single-location business, this ceiling is rarely the limiting factor; it matters more once you're fielding calls across several locations at once.

Step by step

1. Write down the 5 to 10 questions your after-hours callers actually ask (hours, pricing ranges, how to book, whether you service their area), since the agent is only as good as the script you give it.
2. Decide what counts as an emergency for your business, and write an explicit rule: if the caller mentions one of those words or situations, the agent stops trying to help and either transfers the call or takes a detailed message and flags it for an immediate callback.
3. Set up the agent in ElevenLabs' ElevenAgents builder with that script and rule set, connect it to a phone number, and test it yourself by calling at odd hours with both routine and "trick" questions.
4. Run it for two weeks alongside your existing voicemail, listen to the call transcripts, and tighten the script where it guessed wrong.
5. Only then turn off voicemail as the fallback and let the agent be the first and, for most calls, only answer.

Practical tips from actually running one of these

A few things that aren't obvious until you've set one of these up:

• Latency is the thing callers actually notice. ElevenLabs publishes its own real-world numbers: a median time to the agent's first word of around 680 milliseconds, and a worst case around 1,560 milliseconds. That's fast enough to feel like a real conversation most of the time, but on a bad connection or during a busy moment it can feel like a half-second pause, which some callers read as the agent "not understanding." Tell your script to use a short acknowledgement ("let me check that") rather than silence while it thinks. Per ElevenLabs' own changelog (October 26, 2026), a Fast LLM service tier is now available on supported models, trading a higher per-token price for noticeably lower latency; worth testing against the standard tier if those 1,560-millisecond worst cases are actually costing you callers.
• Build the emergency handoff before you launch, not after the first bad call. This is the single most common thing that goes wrong with after-hours agents: someone calls about a genuine emergency, the agent tries to be helpful instead of immediately escalating, and the business finds out about the gap from an angry customer, not from testing.
• Treat ElevenLabs' own status page as part of your setup, not an afterthought. The platform had a real incident in late September 2026 that caused agent conversation failures and speech-to-text errors for about 90 minutes, and a separate multi-hour period of elevated response delay on a slice of text-to-speech requests in early October. Neither is unusual for a cloud AI service, but it means an after-hours phone line built entirely on one vendor has a small, real chance of a bad night. If being unreachable for an hour during an outage would be genuinely costly for your business, keep voicemail live as a silent fallback rather than removing it entirely.

A worked cost comparison at three call volumes

The math changes meaningfully with volume, which is why there's no single right answer. Here's roughly how it plays out for a single-location business, assuming most after-hours calls run 2 to 3 minutes:

• Light volume, around 20 after-hours calls a month: a human answering service's minimum monthly fee usually costs more than the calls themselves are worth. An AI agent on ElevenLabs' Creator plan easily covers this inside its included minutes allowance, so the marginal cost of each call is close to zero once you're past the flat monthly fee. Voicemail is also genuinely defensible here if most of those 20 calls aren't urgent.
• Moderate volume, around 150 after-hours calls a month: a human service bills per minute regardless of call complexity, so 150 calls at 2 to 3 minutes each starts adding up to a real monthly bill, often several hundred dollars. An AI agent on Pro or Scale handles this volume comfortably inside its minutes allowance for a flat fee, which is where the cost gap between the two options becomes obvious.
• High volume, 500+ after-hours calls a month: at this point you're also weighing the simultaneous-call ceiling, not just cost. A human service scales by adding more staff, which costs more per added capacity. An AI agent on Scale or Business handles 30 calls at once with short bursts beyond that, so the constraint shifts from "can we afford enough coverage" to "do we have the knowledge base and routing rules tight enough that the agent doesn't need to escalate most of these calls."

The crossover point in practice is usually somewhere in the low hundreds of calls a month; below that, it's closer to a toss-up depending on how much you value the extra judgment a human provides, and above it, the agent's flat-fee pricing starts winning clearly on cost alone.

What I'd tell a contractor weighing this against a night shift

If you're a field service business, the real question isn't "AI or human," it's "how much of my after-hours call volume is actually routine." For most trades businesses, that's the large majority: people wanting hours, a rough price, or a booking for tomorrow. An AI agent handles that cleanly and never gets tired of answering the same question at 2am. What I'd push back on is treating it as a substitute for judgment on anything that smells like an emergency; write that rule narrowly and conservatively, and don't let the agent improvise on it. The cost math usually favors the agent once you're past a trivial call volume, since a human answering service bills by the minute regardless of how simple the call is, and the agent's included-minutes allowance covers a lot of "what are your hours" calls for a flat monthly fee. Trade owners can read a fuller breakdown in AI receptionists for trade businesses.

How I'd approach this

Most of what I just described, a working after-hours agent with a clean emergency rule, is something you can build yourself on ElevenLabs' own dashboard in an afternoon, and I'd encourage you to try that first. Where I actually get called in is when a business wants the agent to do more than answer generically: check real availability in their scheduling system before offering a slot, look up a customer's job history before talking to them, or log every after-hours call straight into their CRM instead of leaving it in a call transcript nobody reads. That's a custom build on top of the same agent platform, not a replacement for it, and it's exactly the kind of connection work I do.