AEVOMIND INSIGHTS — AI CALLING AGENTS
AI receptionist vs human receptionist: coverage, hours and what actually changes
This is not a contest, and treating it as one produces bad decisions. There are 168 hours in a week; one full-time receptionist covers 37.5 of them. Across a year, after statutory leave, that is just under 20%. The real question is not who answers better — it is what happens during the other 80%, and whether you have ever measured it.
The short answer
A good human receptionist is better than any AI at the things that matter most: reading distress in a voice, exercising judgement, taking responsibility, making a caller feel looked after. Nothing in this article argues otherwise.
But a human is available for about a fifth of the year, and during that fifth they are frequently doing something else. An AI receptionist is available for all of it, at a consistent but lower standard. Those are different products solving different halves of the same problem, and the businesses that get value from AI answering are the ones that understand which half they are buying.
The coverage arithmetic
This is the part almost nobody calculates, and it is the whole argument.
A week contains 168 hours. A standard UK full-time contract is 37.5 of them — 22.3% of the week.
Across a year it is lower. Statutory holiday for a full-time employee is 5.6 weeks including bank holidays, so a full-timer works about 46.4 weeks. That is 37.5 × 46.4 ≈ 1,740 hours, against the 8,736 hours in a year:
1,740 ÷ 8,736 ≈ 19.9%. One full-time receptionist covers roughly one fifth of the year — before sickness, training, lunch or a queue of two calls at once.
These are contracted figures, derived from public statutory entitlements, not survey data. Substitute your own contract hours if they differ. The conclusion holds at any plausible input: most of your phone line's existence is uncovered, and the calls that arrive then leave no trace, which is precisely why the gap feels smaller than it is.
What each is genuinely better at
The human wins on
- Judgement. Knowing this caller needs the manager, now, regardless of the script.
- Distress. Hearing that someone is upset before they say so, and changing approach.
- Authority. Making a decision, granting an exception, taking ownership of a problem.
- Relationships. Recognising a regular caller and what happened last time — as a person, not a record lookup.
The AI wins on
- Availability. The 4 in 5 hours nobody is there, including the evening a customer finally gets round to calling.
- Never being busy. Ten simultaneous calls are answered on the first ring; there is no queue and no engaged tone.
- Consistency. The same questions, asked the same way, captured the same way, on the 400th call as on the first.
- Recording everything. Every call produces a structured record rather than a note that may or may not get written up.
Read those two lists again and notice they barely overlap. That is the actual finding: they are complements, not substitutes.
What changes on the ground
The change people expect is "we stop missing calls". The changes that matter more are quieter:
Your receptionist stops being interrupted. The hidden cost of a phone is not the call, it is the interruption — the person doing something that needs concentration who has to stop, take a message, and start again. Routing routine calls away is often worth more than the calls themselves.
The out-of-hours calls become visible. Most businesses genuinely do not know how many they get, because a call that rings out leaves no record. The first month of data is usually the most surprising part of the deployment.
Enquiries arrive as records, not memories. A call at 9pm becomes a structured entry in your CRM with a next action, rather than a voicemail somebody may listen to on Monday.
And one thing that does not change: the calls that need a person still need a person. The AI's job is to recognise those quickly and hand over, which is a design decision you make up front — not something the technology decides for you.
What it should never handle
Decide this before deployment, not after an incident. In our view an AI receptionist should hand over immediately on:
- Complaints, and anyone who sounds upset
- Safeguarding concerns or vulnerable callers
- Medical, legal or financial advice of any kind
- Anything involving payment or card details
- Any caller who asks to speak to a human — first time of asking, without negotiation
A deployment without a written escalation rule and a named person at the end of it is not finished. We have written more on what AI employees are and what should never be automated, and on the practical questions to ask any answering-service provider.
Working out whether it is worth it for you
Two ordinary weeks of counting beats any vendor's estimate.
Log every missed call, split into in-hours and out-of-hours. Then apply two of your own numbers: what proportion would have been genuine enquiries, and what one converted enquiry is worth to you.
A worked example with illustrative inputs. A business missing 12 calls a week — 4 in-hours, 8 out-of-hours — where one in three is a genuine enquiry and one in four of those converts, at an average job value you already know: 12 × ⅓ × ¼ = 1 converted job a week that currently does not happen. Substitute your own three figures.
If the answer is small, do not buy anything. That is a legitimate outcome and we would rather tell you than sell you something. The arithmetic is the same one we use for automation time savings generally.
Frequently asked questions
How much of the week does one full-time receptionist actually cover?
About 22% of it. There are 168 hours in a week and a standard UK full-time contract is 37.5 of them, which is 22.3%. Across a year the figure is lower: after 5.6 weeks of statutory leave a full-timer works roughly 46.4 weeks, or about 1,740 hours against the 8,736 hours in a year — just under 20%. That is before sickness, training or lunch cover.
Does an AI receptionist replace a human receptionist?
It should not, and framing it that way usually leads to a bad deployment. The honest framing is coverage: a human covers about a fifth of the year brilliantly, and an AI receptionist covers the other four fifths adequately. The work that needs judgement, reassurance or authority still belongs to a person. What changes is that nobody has to choose between answering the phone and doing the job in front of them.
What should an AI receptionist never be allowed to handle?
Anything where being wrong is expensive or distressing. Complaints, safeguarding concerns, vulnerable callers, medical or legal questions, anything involving payment details, and any call where the person is clearly upset. These need a defined escalation path to a named human, and the AI should hand over early rather than late. A deployment without an escalation rule is not finished.
Do callers have to be told they are speaking to an AI?
Tell them, as a matter of practice. Beyond the legal position on recording and data handling under UK GDPR, which you should take your own advice on, disclosure is simply better business: callers who discover it later feel misled, and that costs more than the disclosure ever does. A short, plain sentence at the start works better than an apologetic one.
How do I work out whether it is worth it for my business?
Count missed calls for two ordinary weeks, split into in-hours and out-of-hours. Then estimate what proportion would have been genuine enquiries, and what one converted enquiry is worth to you. Multiply. Most businesses have never measured the out-of-hours number because the calls leave no trace — which is exactly why it is usually the larger of the two.