Case 01 · Automated quoting

A request lands at 2:07 a.m. and gets an offer in minutes.

Quote requests arrive at all hours, and the people who answer them have limited capacity and waking hours. The quotes that wait often lose the booking to someone faster. This started as a sketch: read the request, check the rules, find something that fits, send a real offer. Today it’s a working system that answers in minutes, day or night.

the first sketch↓ what it became

12:07 a.m.

A message comes in.

Daniel wants a place for the weekend for himself, his partner, their nine-year-old and the dog. He’d love a hot tub, and he’d take the lake house if it isn’t much more. Nobody is at their desk.

Designed forNever missing a request, whichever door it comes through, whatever the hour.

22:07 a.m.

It reads the message the way a person would.

It notices what a good agent would: who’s coming, when, what they’d love, and the catch Daniel put on the lake house.

Designed forUnderstanding people the way they actually write, not the way a form would ask.

32:07 a.m.

It knows every partner’s rules.

Each host does things a little differently, and their business has rules of its own. The system keeps a playbook for all of it and turns to the right pages: Host B takes one dog for a $75 fee, and the house rule says every offer shows the full price.

Designed forContext. “Pet-friendly” means one thing at Host B and something else at Host C, and the system knows the difference. When someone corrects it, the lesson gets written down, so nothing is learned twice.

42:07 a.m.

Then it checks its own work.

The AI missed the dog on the first pass but the system caught it. Anything that has to be true gets checked a second time, separately, and the check always wins.

Designed forAccuracy. The things that must never be wrong are checked twice, and a single pass is never trusted on its own.

52:08 – 2:15 a.m.

It goes looking.

Four homes could work. It confirms each one is really free that weekend, then prices it in full: nightly rate, cleaning, pet fee and taxes.

Designed forResilience. Busy hours and slow or unavailable sources delay a request. They never lose one.

62:15 a.m.

It picks the one that fits.

The studio sleeps two. The loft doesn’t take dogs. The lake house is $344 more, just over Daniel’s limit, so it isn’t pushed. The hot-tub cottage fits everything.

Designed forHonesty with the customer. Their limits are respected, and any substitute is named as one, never swapped in silently.

72:16 a.m.

The offer goes out.

Daniel gets a real, bookable offer with the full price just minutes after he asked, and nobody on the team is even awake yet.

Designed forTrust. What the customer sees is built only from facts that were checked, never from what the AI happened to say.

What it took to make it hold up.

A demo reads one message. A real system reads thousands, from real people, on bad days too. Three things shaped every decision.

Accurate

The AI does the reading, but it doesn’t get the last word. Anything that has to be right is confirmed independently, and the customer only ever sees verified facts.

Robust

Not every source responds quickly, and some stop answering if you ask too often. So the system paces itself, pauses on purpose when it needs to, and picks up right where it left off. When something is truly stuck, a person hears about it right away.

Learns as it goes

When a person corrects a result, the lesson goes into the playbook with the reason attached, so the fix sticks and the next change doesn’t quietly undo it.

People stay in charge.

One switch decides whether offers go out on their own or wait for a person’s click.
Anything the system can’t answer cleanly stops and waits for a person.
A dashboard shows every request, what was found, and what was sent.

Same idea, other places.

It fits anywhere people ask in their own words, the answers depend on live availability or pricing, and the rules are super specific.

Travel and hospitality bookings
Freight and logistics quotes
Event and venue inquiries
Custom service estimates