Custom is not the voice, the name, or the greeting. Those are paint.
Everything that actually distinguishes one AI receptionist from another is invisible on a demo call.
Almost every AI receptionist is described as customisable, and almost always the word covers the voice, the business name, the opening line and the list of services. Those are the parts a demo shows, and they are the parts that matter least. What genuinely differs between two well-built lines is what each one refuses to say, what makes it stop and fetch a human, what it treats as available, and what it is allowed to write into your systems.
What are the layers, and which ones matter?
Three layers sit between a caller and a booking, and they are customisable to very different degrees. Presentation is trivially adjustable and nearly irrelevant. Knowledge is easy and useful. Judgement is hard, and it is where lines actually differ.
This is why demos are a weak evaluation tool and why the pilot matters more. In fifteen minutes on a demo you can establish that a line sounds good and knows the prices. You cannot establish what it does with a caller describing chest pain, because nobody demos that.
| Layer | Examples | How much it differentiates |
|---|---|---|
| Presentation | Voice, name, greeting, tone, pacing | Almost none: every vendor offers it |
| Knowledge | Services, prices, hours, staff, locations | Some, but it is data entry, not design |
| Judgement | What to refuse, when to escalate, what counts as available, what to write | All of it |
A demo call shows you the first row clearly, the second row if you ask, and the third row not at all.
What does the judgement layer actually contain?
Four things, each of which has to be decided by someone who knows the business: refusal boundaries, escalation triggers, availability logic, and write permissions. None of them can be inferred from the industry alone.
The availability example is the clearest illustration of why industry knowledge is insufficient on its own. Two Alberta med-spas can differ on this: one with a medical director on site full-time has a different constraint from one where the prescriber attends two days a week. Same industry, same regulation, different line.
- Refusal boundaries. The questions the receptionist answers confidently, and the ones it captures and routes instead. A med-spa's per-unit price is a fact; whether a specific patient is a candidate is not.
- Escalation triggers. Deterministic conditions that stop the call and fetch a person, biased deliberately toward false alarms.
- Availability logic. What actually constrains a booking. In an Alberta med-spa an injectable slot depends on the prescriber being on site, not on the room being free, so a line booking against room availability generates appointments the clinic has to phone back and move.
- Write permissions. What the receptionist may commit to your systems versus what it proposes for a human to confirm. Getting this wrong produces either a useless receptionist or a calendar nobody trusts.
How can you tell whether a line is genuinely custom?
Ask it something it should refuse. A line built for a business has a considered, specific refusal that names what happens next. A line wearing a business's name gives you a generic apology, or worse, an answer.
All four take about four minutes and none of them require technical knowledge. They are also the questions a vendor's demo will not have rehearsed, which is precisely what makes them informative.
- Ask a question your business is not licensed to answer. Listen for whether the refusal names a next step.
- Describe an urgent situation in unusual words. A rule fires or it does not.
- Ask for something outside your published hours or constraints, and see whether it knows the real constraint.
- Ask for a human, insistently. Whether there is a path is a design decision someone made or failed to make.
Custom usually means the voice, the name and the greeting: the three things a demo shows and the three that matter least. What actually differs between two well-built lines is what each refuses to say and what makes it stop and fetch a human.
Four minutes on the phone, asking things a line should refuse, tells you more than any feature list.
Start free pilotWhere this argument is weakest
We benefit from this definition. Here is the pushback.
- We sell the thing this page argues for. AnswerAI builds custom lines rather than selling a configurable product, so this cluster describes our own commercial position. The arguments are made from mechanism and you should check them against a vendor who disagrees.
- The layers are not perfectly separable. Tone genuinely affects outcomes at high price points, so calling presentation irrelevant overstates it. The claim is that it is the least differentiating layer, not that it does not matter.
- Configurable products keep improving. The judgement layer is increasingly exposed as settings in some products. The argument is about who decides, not about what is technically possible.
Questions this raises
- What makes an AI receptionist custom rather than configured?
- Whether the judgement layer was decided for the business (refusal boundaries, escalation triggers, availability logic and write permissions) as opposed to the presentation and knowledge layers being filled in on a standard flow.
- Can I tell from a demo call?
- Not reliably. A demo shows presentation clearly and judgement not at all. Ask it something it should refuse, and something urgent phrased unusually.
- Why does availability logic need customising?
- Because what constrains a booking varies by business, not just by industry. An Alberta med-spa's injectable availability follows the prescriber's schedule, and two clinics in the same city can have different prescriber arrangements.
- Is a custom line harder to change later?
- It should not be. The decisions are written down, so changing one is a decision to revisit rather than an archaeology exercise. A line nobody documented is the hard one to change.
- How long does building one take?
- For us, about 14 days to a working line you can phone yourself. Most of that is deciding the judgement layer with the business rather than engineering.
Sources
- Health Information Act, RSA 2000, c H-5: s. 66, information managersAlberta King's Printer / CanLIIconsulted 2026-08-09
- Recording of Customer Telephone Calls: guidance for organizationsOffice of the Privacy Commissioner of Canadaupdated 2018-04-18, consulted 2026-08-09
The only real test is calling one yourself.
We build a working line on your calls, your pricing, your booking rules and your escalation rules, then you phone it and tell us what does not sound like you. About 14 days. No call limit during the pilot.
Start free pilot