When people hear that we’re building an AI product, they usually picture a chat box.
That makes sense. For the last few years, “AI product” has mostly meant “a place where you type a question and a model answers.” Every app seems to be adding one. A little sparkle icon in the corner, a window that slides open, a friendly “How can I help you today?”
Santara doesn’t work that way. It doesn’t wait for you to ask anything. Every morning, it sends short-term rental hosts one briefing: who’s checking in today, which rooms need cleaning, and which guest messages need a reply.
That was a deliberate choice, and it shapes almost everything we build. This post explains why.
It’s also my answer, so far, to one of the open questions on my Now page:
How do you build AI that a busy small business owner actually trusts, every single morning?
A host’s morning, before any AI
To understand the choice, you have to picture the person we build for.
A small short-term rental host might manage three, five or fifteen apartments. Bookings arrive on Airbnb. More arrive on Booking.com. Guests write on WhatsApp, sometimes on the platform’s own inbox, sometimes both. A cleaner needs to know which rooms to turn over before 2 p.m. Someone is asking about early check-in. Someone else is asking where to park.
So the morning goes like this: open Airbnb, scroll. Open Booking.com, scroll. Open WhatsApp, scroll through twenty chats. Try to remember which guest is arriving at which apartment. Write a list on paper or in a spreadsheet. Message the cleaner. Answer the guests who seem urgent. Hope nothing got missed.
None of it is hard. All of it takes time, every single day. And the cost of a mistake, a room not cleaned, a guest not answered, a double booking spotted too late, lands as a bad review.
That’s the job we’re trying to take away. As I wrote in People Don’t Pay for Features. They Pay to Do Less., the host doesn’t want “AI.” They want a calmer morning.
Why a chatbot is the wrong shape
Imagine we’d given that host a chatbot instead.
To get value from it, the host would have to:
- remember to open it;
- know what to ask (“Who’s checking in today?” “Any messages I missed?”);
- read the answer and decide if it’s right;
- ask again, in different words, when it isn’t quite what they needed;
- do the same thing again tomorrow.
Look at that list. It’s more work, not less. We would have replaced “scroll three apps” with “interview a robot.” For a person who’s already busy, that isn’t a gift. It’s a new chore.
Chat interfaces are wonderful when the user doesn’t know what they need until they ask: research, writing, exploring ideas. They’re the wrong shape when the user needs the same answers every day. A host’s questions barely change. Who’s arriving? What needs cleaning? Who’s waiting for me? If you already know the questions, you shouldn’t make people type them.
There’s a second problem. A chatbot puts the burden of noticing on the user. If a guest asked something important at 11 p.m. and the host doesn’t think to ask “did anyone message me late last night?”, the chatbot will never mention it. The things people miss are exactly the things they don’t think to ask about.
A briefing flips that around. The product does the noticing. The host just reads.
What “one briefing” means
The principle is simple: the host should be able to start their day by reading one thing.
The briefing answers three questions, in this order:
1. Who’s arriving and leaving today? Check-ins and check-outs across all platforms, in one list, by apartment and time. If a guest mentioned an early arrival or a late flight in a message, that note belongs here, next to their booking.
2. What needs to be cleaned or prepared? Which apartments turn over today, and by when. Anything unusual, like a guest who asked for an extra bed.
3. What needs a reply? Guest messages that are still waiting for an answer, with the most urgent first. Not every message, just the ones where someone is waiting for the host.
That’s it. Not a dashboard with forty widgets. Not a feed. A short list a person can read with their morning coffee and then act on.
The hard part isn’t the layout. The hard part is making that short list correct.
Trust is the real product
Here’s the uncomfortable truth about AI for small businesses: the first wrong answer costs more than ten right ones earn.
If the briefing says a room is free and a guest shows up, the host gets a bad review and an angry phone call. After that, they’ll check every line of the briefing by hand. And once they’re checking everything by hand, the briefing is just one more app to open. We’ve added work again.
So the whole product is designed around one goal: a host should be able to trust the briefing without double-checking it. Not on day one, nobody trusts anything on day one, but after a few weeks of it being right.
In How to Tell If an AI Product Is Real or Just a Demo I wrote that demos are easy and daily reliability is hard. A briefing is a promise you make again every single morning. Here are the rules we use to keep it.
Rule 1: Facts come from data, not from the model
Some things in the briefing are facts: a booking, its dates, the apartment, the number of guests. Those come straight from the booking data. We don’t ask a language model to “figure out” who’s checking in. The system already knows.
The AI is used where the information is genuinely messy: guest messages. That’s where a model earns its place. A message like “hi!! our flight got moved, we’ll land around 9pm, is that ok? also can we leave bags after checkout” contains a late arrival, a question and a request. A person can read that in a second. Software couldn’t, until recently. Now a model can turn it into structured notes: arriving late, needs a reply, asking about luggage storage.
Keeping the facts and the interpretation separate means that the parts that must never be wrong aren’t left to a model’s guess.
Rule 2: Always show where it came from
Every note in the briefing links back to its source: the booking, or the guest’s original message. If the briefing says “Guest in Apartment 3 is arriving late,” the host can tap it and see the actual message.
This does two things. It lets the host check quickly when something looks odd. And over time, it builds trust, because every time they check, the source matches. A summary you can’t verify is a summary you’ll never fully believe.
Rule 3: Say “I’m not sure”
Language models are confident by nature. They’ll give an answer even when they shouldn’t. In a briefing, a confident wrong answer is the worst possible outcome.
So when the system can’t tell what a message means, the briefing says so. “Unclear message from guest in Apartment 5, please read.” That’s less impressive than a perfect summary. It’s far more useful than a wrong one. A host can deal with “please check this.” They can’t deal with a mistake they didn’t know about.
Rule 4: Don’t act without asking
It’s tempting to let AI do everything: reply to guests, message cleaners, change bookings. Some of that may come later. But anything that leaves the host’s hands, a message to a guest, a change to a booking, has to be something the host chose to do, at least until they’ve decided to trust the system with it.
The briefing tells. The host decides. That line matters more than any feature.
Rule 5: Be the same every day
Habits are built on predictability. The briefing arrives at the same time, in the same order, with the same structure. If nothing needs attention, it says so in one line, and that’s a good morning.
A product that’s a little different every day makes people read carefully every day. A product that’s consistent lets people skim, notice what’s unusual, and get on with their lives.
The pipeline, in plain words
Without getting into anything confidential, here is the general shape of what happens before a host wakes up:
- Collect. Gather the latest bookings and guest messages from the connected platforms.
- Match. Figure out which message belongs to which booking and which apartment. This sounds trivial and is not: the same guest can have a different name, phone number or spelling on each platform.
- Understand. Use a language model only on the parts that need it, mostly messages, to pull out what matters: questions, requests, changes, problems.
- Check. Compare what the model found against the facts we already have. If a message says “see you Friday” but the booking says Saturday, that’s worth flagging, not hiding.
- Prioritize. Decide what goes at the top. A guest arriving in three hours who hasn’t received the door code beats a question about restaurants.
- Write. Produce the short, plain briefing, with every item linked to its source.
Most of this isn’t AI. It’s careful, boring engineering: syncing data, matching records, handling missing information, retrying when a platform is slow. The model is one important step in the middle. It isn’t the product. The product is the morning.
What we gave up
Choosing a briefing over a chatbot had costs, and I want to be honest about them.
It demos worse. A chat box where you type anything and get an answer feels magical in a five-minute demo. A briefing that quietly says “3 check-ins, 2 cleanings, 1 message waiting” doesn’t. We’re betting that daily usefulness beats a good first impression.
It’s less flexible. If a host wants to ask something unusual, a briefing can’t answer it. We may add ways to ask questions later. But the default, the thing that arrives every morning, stays a briefing.
It’s harder to build. With a chatbot, you can let the user carry part of the responsibility: they asked, they read the answer, they decide. With a briefing, we decide what’s important. If we miss something, it’s on us. That forces us to be much more careful, which I think is a good thing.
What I’ve learned so far
We’re early, and I’m sure we’ll get parts of this wrong. But a few beliefs have become clearer while building Santara.
AI should remove steps, not add them. If using your AI product means opening something, typing something and checking something, ask whether you’ve actually saved anyone time.
The best interface is often no interface. The ideal briefing is so clear that the host reads it in a minute and closes it. Time spent in our product is not a metric we want to grow.
Trust is built in small, boring moments. Not in a demo. In the fortieth morning in a row that the briefing was right.
Say what you don’t know. The products people trust aren’t the ones that never make mistakes. They’re the ones that are honest about where mistakes could happen.
AI won’t replace small business owners. It shouldn’t try to. What it can do, if it’s built with care, is take the copying, sorting and remembering off their plate, so the first hour of the day belongs to them again.
That’s the job. One briefing, every morning, right.
Do you host short-term rentals? I’d love to hear how your mornings really look, and what you’d want in a briefing. Get in touch.
