Across every customer we work with — clinics, salons, logistics firms, tuition centres, spread across fourteen countries now — there is one question that shows up almost everywhere, in almost the same words: some version of "do you have anything available?" I've read enough of these conversations at this point to notice something that isn't obvious from the outside: those four words mean completely different things depending on who's asking, and configuring AIVA as if they mean the same thing is the most common early mistake we see.
The same four words, wherever you go
"Do you have anything today?" from a salon customer is a browsing question. They're mildly hoping, not particularly invested, happy to hear "not today, how about Thursday" and move on without a second thought. The right response is quick, upbeat, efficient — get them a slot or a clear no and let the conversation end.
The identical sentence from a patient calling a clinic is rarely a browsing question. It's frequently attached to something that hurts, or worries them, or has been getting worse since yesterday. The words are the same. The emotional weight underneath them is not even close.
Same words, different stakes
This is the pattern I keep coming back to in onboarding calls: businesses assume the hard part of setting up AIVA is teaching it their services, hours, and prices. That part is genuinely the easy part — it's a spreadsheet. The hard part is teaching it the register the conversation should be held in, which has almost nothing to do with the words a customer says and everything to do with what's actually going on when they say them.
A logistics customer asking "where's my shipment" wants speed and precision, full stop — the emotional register is closer to checking a scoreboard than anything personal. A parent calling a tuition centre asking "is there still a spot" is often anxious about their child, not just information-seeking. A clinic patient asking about availability may be in pain. Treat all three the way you'd treat the salon customer — quick, upbeat, efficient — and you'll answer the question correctly and still leave the caller feeling unheard.
More of the same pattern, in businesses we didn't expect
It kept showing up in verticals I wouldn't have predicted before I started reading transcripts for a living. A property manager asking "is the unit still up" is often mid-decision on a lease they're anxious about losing to someone else — that's urgency, not browsing, even though the sentence structure is nearly identical to the salon question. A wedding planner fielding "do you have our date open" is fielding a question loaded with months of planning and, often, real stress about a fixed, unmoveable day — get that register wrong and a factually perfect answer can still feel dismissive of how much the date matters to the person asking. Even a veterinary clinic's version of "can you see us today" splits in two depending on whether it's a routine vaccination or a pet that's stopped eating, and a good configuration has to leave room for AIVA to pick up on which one it's actually hearing rather than assuming the calmer version by default.
The FAQ content is almost never the problem. The tone a business needs for its specific customers, in that specific moment, almost always is.
Where new customers usually get it wrong
The mistake I see most often: a new customer configures AIVA with generic, cheerful default language because it reads well on a setup screen, without thinking hard about who's actually going to be on the other end of a real call. It's an easy mistake, because in isolation, cheerful and efficient sounds like a safe default. It's only wrong in context — and the context is different for every industry we serve, sometimes for every customer within the same industry.
We've seen it go wrong in the other direction too. A logistics dispatcher once configured their AIVA with an unhurried, deliberately warm tone modelled on a healthcare customer's setup they'd seen in a demo. Their actual customers wanted speed above all else, and the warmth read as slow. Same fix, opposite direction: match the register to the real emotional stakes of the actual question, not to what sounded nice in someone else's demo.
"Isn't this overthinking a phone tree?"
I hear a version of this pushback often enough that it's worth answering directly: the business already answers correctly, so why does the emotional register matter as long as the facts are right? It's a reasonable question from anyone who hasn't sat through a call where it went wrong. The honest answer is that a factually correct call can still fail a customer, and it usually fails quietly — no complaint gets filed, the caller just doesn't come back, or mentions to a friend that the place "felt a bit off" without being able to say why. Our engineering team ran into almost the same lesson from a completely different angle — a call that was correct by every measurable metric still needed a person to notice something the transcript couldn't show. Tone problems and escalation problems are cousins: both are invisible to a dashboard and obvious to anyone actually listening. This isn't only a customer-service nicety, either — accessibility guidance on cognitive load and plain language makes a version of the same point: how something is communicated determines whether a person under stress can actually use it, independent of whether the underlying information is accurate. Register is a usability question as much as a warmth one.
What we actually onboard people on now
We changed our onboarding conversation because of this. We used to start by asking new customers for their FAQ list and service catalogue — the spreadsheet stuff. We now start with a shorter, more human set of questions, roughly in this order:
- What is your customer usually feeling when they pick up the phone to call you?
- Which of your regular questions carries the most anxiety, even if it sounds routine?
- Is there a moment mid-call where the mood usually shifts — and how do your best staff handle that shift today?
- What would a customer notice if a good call became a slightly-off one?
That last question does more work than it sounds like it should. Most owners have never articulated it out loud, and answering it tends to surface the register decisions that actually matter, faster than any FAQ spreadsheet does. The mechanics of writing FAQs an AI agent can use well still matter — they're just no longer the first conversation we have.
I work remotely, out of Bengaluru rather than the Rajkot office, which turns out to suit this part of the job better than sitting in the same room as the product team ever would have. A meaningful share of my week is still spent the low-tech way: reading transcripts, and increasingly listening to the audio itself rather than trusting the text alone — the same instinct our design lead has written about from his side of the product, just applied to real customer calls instead of test ones. A transcript can tell you a patient asked the same question three times. It's much harder for a page of text to tell you how she asked it — and register, more often than not, lives in the how, not the what.
Getting it right, not just avoiding getting it wrong
It's worth saying plainly that most of this goes right, not wrong — I spend more of my week reinforcing a configuration that's already working than fixing one that isn't. A coaching institute we onboarded early this year nailed its register almost immediately, mostly because the owner came into setup already able to describe, precisely, what a worried parent's call sounded like versus a returning parent's routine one. We didn't have to extract that insight from her. She already had it; our job was just translating it into a configuration. That's the ideal version of this process, and it's more common than the mistakes are — the mistakes just make for more useful stories. It's also a fair chunk of why customer success existed here before we had a dedicated sales team: getting the register right isn't a one-time setup task, it's an ongoing relationship with the account.
Four million conversations, one lesson
Four million conversations a month across this many industries has taught me one thing above everything else: the words people use to ask for something are far less interesting than what's actually happening in their life when they ask it. Build for the first and you'll answer questions correctly. Build for the second and people will actually feel taken care of — which, across every language AIVA answers in, turns out to be a harder and more universal problem than getting the facts right ever was.