All posts
Testing

· 6 min read

By Syed Shariq, Co-founder & CEO of Estric AI · Editorial policy

How to test your AI receptionist before you let customers near it

Twenty minutes of deliberate rehearsal catches almost everything that would embarrass you in front of a real caller. Here is the checklist, in the order that finds problems fastest.

The gap between an AI front desk that impresses customers and one that embarrasses you is almost never the technology. It is whether anyone sat down and listened to it the way a customer will. Owners tend to skip this: the setup wizard says everything is connected, the dashboard is green, so the number goes live and the first real caller becomes the test.

That is an expensive way to find out that your Saturday hours are wrong. Deliberate rehearsal takes about twenty minutes and finds nearly everything. What follows is the order that finds problems fastest — greeting first, edge cases last — so if you only have ten minutes, the first half is the half that matters.

Call it yourself, out loud, on a phone

Not the chat window. A phone, held to your ear, in a room with normal background noise. Voice is where the flaws live: a greeting that reads fine on screen can sound stilted, too long, or oddly formal when spoken. If you would not say it to a customer standing in front of you, it is wrong.

Listen specifically for three things in the first ten seconds. Does it say the business name the way you say it? Does it get to the point, or does it deliver a paragraph before the caller can speak? And does it sound like it belongs to your business, or to a call centre? First impressions are decided before the caller has said a word.

Ask your three most-asked questions

You already know what they are: what do you charge, are you open on a particular day, and do you cover a particular suburb or service. Ask each one in the words a customer would use, including the vague version. "How much is it?" with no other context is a real question customers ask, and the answer should be useful rather than a request for clarification.

Check the answers against what is true today, not what was true when you filled in the wizard. Prices drift, hours change at Christmas, and service areas expand. An AI that confidently states a price you no longer charge is worse than one that says it will check — you will honour that quote or lose the customer arguing about it.

Book something, all the way through

Take one booking from first hello to confirmation, and then look at what landed. Is the appointment in the right calendar, at the right time, in the right timezone, with the right length for that service? Did the confirmation actually arrive, and does it read like your business sent it?

Then do the awkward version: ask for a time you are closed, or a slot that is already taken. What you want to hear is a specific alternative — "there is nothing at four, but I can do half past five, or Thursday morning" — not a dead end. Most lost bookings are lost precisely here.

Ask something it should not know

Ask a question you never gave it an answer to. Something plausible and specific: whether you service a brand you do not carry, or offer a treatment you do not provide. The correct behaviour is a straight admission and a route to a human. The dangerous behaviour is a confident invention, and you would far rather discover that yourself than read it in a complaint.

This is also the test that tells you whether your knowledge base has holes. Every question that gets a vague answer is a line you should go and write. Nine times out of ten the fix is one sentence in your own words.

Check the handover

Say the thing an unhappy customer says: "I want to speak to a person." The AI should stop selling, stop explaining, and hand over cleanly — take a number, flag the conversation, or transfer, depending on how you have it set up. A front desk that will not let go of a caller is the single fastest way to turn a small problem into a review.

Then confirm the flag actually reached you. A handover that raises a notification nobody sees is the same as no handover at all.

Rehearse without spending your customer allowance

One reason owners under-test is that it feels wasteful: every rehearsal call uses minutes bought for real customers, and every practice booking lands in the inbox looking exactly like a real one, to be cleaned up later.

Estric AI is adding testing credits to remove both objections — a separate one-off allowance for calling and chatting to your own front desk from the dashboard, kept out of your plan allowance, with test bookings kept out of your real inbox. Solo and Starter will start with five minutes of calls and a hundred chats; Growth and Scale with twenty minutes. The amount is set by the plan you start on and granted once. It is not switched on yet, so for now, rehearse anyway and delete the test booking: twenty minutes of your own allowance is cheaper than one caller hearing the wrong price.

Re-test whenever the facts change

This is not a one-time gate. Every time you change prices, add a service, move your hours, or take on a new staff member, the same twenty-minute pass is worth repeating on the parts you touched. Most of the drift that makes an AI front desk sound out of date is not the AI drifting — it is the business changing and nobody telling it.

Put it on the same schedule as anything else that goes stale: whenever the price list changes, and once a quarter regardless.

Frequently asked questions

How long should I spend testing an AI receptionist before going live?

About twenty minutes covers it: the greeting, your three most-asked questions, one complete booking, one question it should not be able to answer, and the handover to a human. Repeat the relevant parts whenever prices, hours or services change.

What is the most common problem found when testing?

Out-of-date facts. Prices, holiday hours and service areas drift after setup, and an AI states the old ones confidently. The second most common is a dead end on unavailable times, where the caller is told no without being offered an alternative.

Do test calls use my plan allowance?

Today, yes — rehearsal calls and chats come out of the same allowance as customer traffic. Estric AI is adding separate testing credits so they do not, but that is not switched on yet.

Put Estric AI on your phone

Estric AI answers every call, books the appointment, and texts the customer back, 24/7, on the number you already have.

Get started