Why the demo isn't the test
Every provider's demo is a happy path: a clear voice, a quiet room, a simple booking. Your customers ring from a ute with the window down, change their mind halfway through, and sometimes just want to talk to you. The gap between those two calls is where businesses get burned.
People who build these systems themselves describe days of tuning before they trusted one with real calls. You don't need to do the building, but you do need to do the calling.
Test it the way your customers call
Before you start, write down the three calls you get most often and the one call you dread. Those are your real test cases. Then run the list above, in this order:
- A normal booking. If it can't do this smoothly, stop.
- The messy version of that call. Interrupt, correct a detail, call from somewhere noisy.
- The escape hatches. Ask for a person, describe an emergency, ask if it's a robot.
- The edges. Something it can't know, a card payment, a double booking.
- The paperwork. What landed in the dashboard, who got told, how fast a change gets made.
What a good answer sounds like
On the tests that matter most, here's the difference between a system worth buying and one that will cost you customers:
- Asking for a person: good is an immediate transfer or message. Bad is a loop, an argument, or "I can help you with that instead" three times.
- An emergency: good is your own escalation rule, said calmly, plus 111 where there's danger. Bad is being offered a booking slot next Tuesday.
- Something it can't know: good is "I'll take a message and someone will confirm that." Bad is a confident, invented price.
- Am I talking to a robot? good is a straight yes. Anything evasive tells you how the vendor thinks about your customers.
The part almost everyone skips
After your emergency test call, go and see what actually reached a human. Not the dashboard: your phone. The quiet failure people report most often is an AI that answered politely, took a message, and then nothing happened, because the message sat in a system nobody watches. Test the whole chain, not the conversation.
Then run a real trial
Test calls prove it can work. A week of real calls proves it does. During a trial, read the transcripts daily for the first few days, and note every call where you'd have answered differently. That list is what a managed provider should fix for you within a day or two.
If you want a second opinion before you commit, the failure guide lists the things that break after go-live, and the cost calculator tells you whether the money makes sense at your call volume.