The biggest surprise was not that the genie handled calls well after hours. It was that callers behaved completely differently after hours than they did during the day.

I expected after-hours to be a graveyard of frustrated people leaving their name and number. What we actually saw was something closer to a confessional. People asked longer questions. They went into more detail. They were, frankly, more honest about what they needed than they ever were with a human on the other end of the line.

That caught me off guard. And it changed how I think about what voice AI is actually for.

The after-hours caller is a different person

When someone calls a business during office hours and gets a human, they tend to compress. They give you the headline. “I need a quote.” “Is this part in stock?” “What are your hours on Saturday?”

But when someone calls at 10pm and they know there’s no human waiting, something shifts. They tell you the whole story. “We moved in three months ago and the hot water system has been making this sound, and I’ve been putting off calling because I wasn’t sure if it was covered under warranty, and my wife is getting frustrated, and I just want to know what to do.”

That level of detail is gold. The genie picked up on it, worked through the knowledge base, and in a lot of cases resolved the question without any follow-up needed. The caller got an answer. The business got a transcript with every useful detail already captured.

If you’re in trades, appliances, or home building, you’ll know exactly what I mean. The worried homeowner who calls at night is often your most valuable customer. They’re not shopping around. They just want someone to help them, and they’ll explain exactly what they need if you give them the space to do it.

Where the genie beat a human. Genuinely.

I want to be careful here because it’s easy to overclaim. There are plenty of things a skilled human does on a call that voice AI simply cannot replicate. Reading hesitation. Picking up that a customer is about to walk. Knowing when to stop talking.

But there is one area where the genie was consistently better. Product and pricing detail.

A human who answers the phone at a car dealership or a marine dealer is juggling a lot. They might know the headline specs on the top-selling models, but ask them something specific about a particular trim level or a finance comparison, and you get “let me find out and call you back.” Which, if you’re honest with yourself, means a 30 to 40 percent chance they never do.

The genie does not have that problem. The knowledge base does not get tired. It does not forget what’s in the brochure. It does not have a bad day. If the information is in there, it comes out. Every time, for every caller, at 2am if that’s when they’re asking.

Automotive and marine businesses saw this most clearly in our data. The kinds of calls that used to require a specialist callback were being resolved in the conversation itself. Callers were getting to a decision point faster, not slower.

That was not something I predicted going in.

Where the genie had to hand off. And what that taught us.

Here is the thing nobody tells you about voice AI. The escalations are not where you think they’ll be.

I assumed the genie would struggle most with complex, emotional situations. Complaints. Disputes. Someone furious about a delayed delivery. And yes, it escalated those. But those were the expected handoffs. We had designed for them.

The unexpected escalations were transactional. Small, specific requests that should have been easy. Things like “can you move my appointment from Thursday to Friday afternoon?” Or “I need to update the delivery address on my order.”

The problem was not understanding the request. The genie understood it perfectly. The problem was that completing those requests required writing back to a system the genie did not yet have access to. It could read the knowledge base. It could not write to the booking system or the CRM.

That is a real limitation and I am not going to pretend otherwise. It shaped a lot of how we think about integration now. If you are looking at our explore page to figure out what a genie can and cannot do, that distinction matters. Answering questions from a knowledge base. Capturing leads. Routing callers. All of that works well out of the box. Triggering actions inside third-party systems requires the API and webhook layer, and that takes a bit more setup.

The businesses that got the most value, early, were the ones who leaned into what the genie does well and did not try to force it to replace their whole CRM workflow in week one.

What callers actually ask after hours

If you are thinking about deploying a genie and wondering what you’ll get, here is an honest picture of the enquiry types we saw most often outside business hours.

Pricing and availability questions dominated. Not surprising. But the specific framing often was. People rarely asked “how much does X cost?” They asked “is X worth the extra money compared to Y?” They wanted a recommendation, not a price list. The genies that had been given enough context to offer a genuine comparison performed much better than the ones that had only been given specs.

Warranty and service questions came up constantly, especially in appliances, industrial, and manufacturing. Customers wanted to know what was covered, how to make a claim, and how long they would be waiting. Most of that information exists somewhere in a manual or a policy document. The problem is it was never accessible at 9pm on a Sunday. Now it is.

Qualification questions were the third big bucket. “Am I the right kind of buyer for this?” “Do you do commercial work or just residential?” “Do you work in my area?” These are exactly the kinds of questions where a genie shines. Answer them well and you filter in the right leads. You also save your sales team from spending Tuesday morning calling back people who were never going to buy.

If you want to put a number on what that’s worth, the ROI calculator is worth a few minutes of your time. It is not a magic number generator. It just forces you to be honest about how many after-hours enquiries you are currently missing and what each one is worth to your business.

The lesson about caller trust

One thing I did not expect to be thinking about this much is trust.

There’s a school of thought that says customers don’t want to talk to AI. That they’ll hang up the moment they realise it’s not a human. We heard that concern a lot before we went into production.

What we actually observed was more nuanced. Callers who were given a clear, competent answer to their question did not care much about the mechanism. Callers who hit a dead end, who got a vague non-answer, who felt like they were going in circles, those callers did care. And they blamed “the robot.”

The variable was not AI versus human. It was whether the caller got what they came for.

That reframes the whole conversation for me. The question is not “will people accept voice AI?” The question is “is your genie actually useful?” If the knowledge base is thin, if the conversation logic is sloppy, if the genie does not know when to say “I don’t have that information, let me make sure someone follows up with you,” then yes, callers will be frustrated. Rightfully.

But a genie that knows the product, speaks in the brand’s voice, and gives real answers? Callers finish those conversations and move on with their evening. Which is exactly what they wanted to do.

Real estate and events businesses found this particularly relevant. Those callers often have a lot of specific questions. Is this property still available? What’s included in the venue hire? What’s the cancellation policy? If your genie can answer those questions clearly and completely, the caller does not sit there questioning whether they’re talking to a person.

What this means if you’re considering voice AI

Be honest about your knowledge base before anything else. The technology is not the hard part. Getting your information into a shape where it’s genuinely useful, specific, and complete, that is the work. If your knowledge base is vague, your genie will be vague.

Start with the use cases that are read-only. Answering questions. Capturing lead details. Routing to the right person. Do those well first, then layer in more complex workflows once you understand how your callers actually behave.

And expect to be surprised. I was, and I helped build this thing.

If you want to see what a genie looks like in practice for your industry, start at the explore page. If you want to get concrete about the numbers, the ROI calculator will give you a place to start.

The calls are happening whether you pick up or not. The question is just whether you’re capturing anything useful from them.