Guest AI chat · observability · self-healing

The guest AI that tells you when it fell short — and fixes itself

One AI assistant across your property sites, group site, and guest portals — every reply graded, gaps explained, and one-click fixes that heal it across every property.

See the guest experience

What makes it different

  • One assistant on every surface: property sites, group sites, guest portals, and custom domains
  • Public answers stay public; door codes, Wi‑Fi, and arrival details unlock only after a booking
  • Every reply graded — answered, low confidence, couldn’t answer — with the reason it struggled
  • Train it by talking to it: chat with your own sites and guest portals, see the replies, feed it the right context
  • Fix a gap once and it applies to one site or all of them; insights re-run on their own

A chatbot you can’t see is a chatbot you can’t trust

Most guest chatbots are fire-and-forget: you paste in some blanket info, switch it on, and hope. You never learn which questions it fumbled, which properties it knows nothing about, or what guests actually ask before they book. Haven treats the guest AI as a full-cycle tool. It grounds every answer in the specific listing and the guest’s access level, grades its own replies, tells you exactly where it fell short and why, and gives you a one-click way to teach it — for one property or across every site you run. Then it keeps checking, so the assistant gets better every week without you digging through transcripts.

Blanket prompts break at property two

One generic instruction block can’t know that Unit B parks behind the building and the Loft has a crib. Haven scopes knowledge per site, per group site, or account-wide — and layers it automatically.

No idea what it got wrong

Without grading, a bad answer just disappears. Haven marks every reply as answered, low confidence, or couldn’t answer, and explains the gap in plain English.

Fixes never reach the other properties

Teach it once in Haven and choose the scope: this site, this group site, or all sites. The next guest, on any surface, gets the confident answer.

How the full cycle works

1

Guests chat on any surface

The assistant lives on your property sites, your group site, and every guest portal. It answers from the listing itself — amenities, rules, pricing, cancellation policy, local area — plus your FAQs and AI knowledge.

2

It respects who’s asking

Anonymous visitors get public answers. Guests who open a gated guest portal after booking also get private stay details — door codes, Wi‑Fi, check-in specifics, and what they told you in guest forms.

3

Every reply is graded automatically

Haven pairs each guest question with the assistant’s reply and grades it: answered, low confidence, or couldn’t answer — with a short reason and the context that would have fixed it.

4

Insights surface the pattern, not the noise

Analytics → Chats groups flagged replies into topics (Parking, Check-in, EV charging…) with counts, cited conversations, and which properties they came from.

5

Fix it once, everywhere

Click Add context: the dialog is prefilled with the suggested wording and the right scope. Save it to one site or all sites, public or private. It’s live for the next chat — no redeploy, no re-training.

6

Train it by talking to it

Open any of your property sites, your group site, or a guest portal and ask what a guest would. Your own chats show up in the Chats tab, graded like every other — so you can see the reply, add the missing context (private or public, one property, a group site, or everywhere), and ask again.

7

The loop keeps running

Insights re-run on their own as new chats arrive. Ask “what couldn’t it answer this week?” in plain English and click straight into the cited chats. Escalations still land in your Messages inbox.

See the full cycle

Detect → explain → fix → verify, across every property

Watch how one assistant answers on every surface, how Haven grades each reply and explains the gap, how you train it by talking to it, and how a single Add context heals it for one site or all of them.

One assistant

Answers guests on every site — and knows what to keep private

The same AI chats on your property sites, group site, and guest portals. Public visitors get public answers; door codes, Wi-Fi, and arrival details only unlock for guests with a booking.

One assistant · every site you run
Property site
Public
Is there parking?
Group site
Public
Which unit fits 6 with a crib?
Guest portal · after booking
Private stay details
What’s the door code?
Public info stays public · stay details unlock after booking

Key capabilities

Built for

Multi-property operators

A blanket prompt can’t know ten properties. Scoped knowledge, per-site grading, and all-sites fixes mean the assistant scales with your portfolio instead of breaking at property two.

Hosts who want automation they can trust

Automate guest messaging without going blind. You see exactly what the assistant answered well, where it struggled, and why — and fix it in one click.

Boutique inns and small hotels

Room-type questions, late arrivals, and local recommendations answered instantly on your site, with private arrival details reserved for confirmed guests.

Anyone tired of repeating themselves

The assistant handles the pre-booking and during-stay questions; you handle the ones that need a human — with the full transcript already in your inbox.

Because it’s one platform

The assistant is not a widget bolted onto your site — it shares the same listings, bookings and knowledge base as everything else in Haven. Teach it once and every channel knows.

It already knows the listing

Answers are grounded in the same property content that powers your booking site — amenities, rules, pricing, policies. Edit the listing and the assistant is current on the next chat.

See the website builder

The same brain answers OTA guests

The knowledge base you build fixing site-chat gaps is the one Haven’s AI uses to reply to Airbnb, Vrbo and Booking.com guests in their own threads, through your PMS.

See AI guest messaging

Chats sit next to replays

Every graded conversation lives beside session replays and the booking funnel — so “guests keep asking about parking” and “guests drop off at checkout” are answered from the same screen.

See analytics & replay

Observability, self-healing, and automation — not a black box

Haven’s guest AI is a first-class, full-cycle tool. It detects where it isn’t satisfying guests, guides you to the exact context to add across any or all properties, and gives you complete insight into how your AI messaging performs — without digging through transcripts.

Detect

Every reply is paired with its question and graded. Low-confidence and unanswerable replies are flagged with a plain-English reason.

Explain

Insights group flagged replies into topics with counts, affected properties, and cited conversations — so you see the pattern in seconds.

Fix

Add context is prefilled with the suggested wording and scope. One save teaches the assistant on one site or every site you run.

Verify

The next guest’s reply is graded like every other. Insights re-run automatically, so a fixed gap disappears from the list on its own.

Train

Talk to your own chats on any surface — site, group site, guest portal — see the answers, and feed them the right context: private or public, one property, a group site, or global.

Ask

Ask your chats in plain English — “what couldn’t it answer this week?” — and jump straight into the cited chats to act.

Escalate

When a human is needed, the full transcript lands on your Messages thread and you’re emailed. Automation never means guests fall through.

Built for hosts — not a generic chatbot widget

Most chat tools give you a prompt box and a transcript list. Haven gives you a grounded assistant per property, a grading loop that tells you what to fix, and a knowledge system that scales across your whole portfolio.

Generic website chatbots

One prompt for the whole account, no idea which property a guest means, no grading, no private stay details. You find out it was wrong when a guest tells you.

PMS auto-replies

Templated messages inside a reservation thread — useful after booking, silent before it. They don’t answer the visitor deciding whether to book on your site tonight.

Guidebook-only assistants

Helpful during the stay, but they know nothing about pricing, availability, or the rest of your portfolio, and they can’t report back on what they missed.

Haven

One assistant across sites and guest portals, public/private aware, every reply graded, insights with evidence, one-click fixes across all properties, and hand-off to you when it matters.

Frequently asked questions

On every property site, on group sites, inside gated guest portals, and on your custom domains. It is one assistant with knowledge scoped to whichever listing the guest is looking at.

Listing content is labelled public or booking-only. Public site visitors only ever get public answers (amenities, rules, pricing, policies, local area). Door codes, Wi‑Fi, check-in specifics, contacts, and guest-form answers are booking-only and are only available to a guest who opened a gated guest portal after booking. Your own AI knowledge entries can be marked public or private too.

Haven pairs every guest question with the assistant’s reply and grades it as answered, low confidence, or couldn’t answer, with a short reason. Analytics → Chats shows the totals, lets you filter to any signal, groups flagged replies into topic insights with cited conversations, and opens the exact transcript with the flagged reply highlighted.

When a reply is flagged, Haven suggests the context that would have fixed it. Click Add context and the dialog is prefilled with that wording, the right property or group site, and the right visibility. Save it to one site or all sites and it is live for the next chat. Insights re-run automatically, so a fixed gap disappears from the list on its own.

No. The assistant already knows each listing from its content. On top of that, Messaging → AI knowledge lets you add entries per property, per group site, or account-wide, and build reusable knowledge sets you attach to any number of sites. Your FAQs feed the assistant automatically too.

Talk to it. Open your property site, your group site, or a guest portal and ask the questions a guest would ask. Those conversations appear in Analytics → Chats graded like any other, so you can see exactly how it replied. If an answer is thin, add context right there — private or public, scoped to that property, to a group site, or to every site — and ask again. There is no separate sandbox to configure; you train it in the same environments guests use.

Yes. The Chats tab has a plain-English question bar — “Are people asking about parking?”, “What couldn’t the assistant answer this week?” — and every answer cites the exact conversations, which open scrolled to the cited message. The same tool is available to the Haven assistant in your dashboard.

It says so honestly instead of inventing property details. If the guest agrees, the full conversation is passed to your Messages inbox and you are emailed, so nothing is lost. The unanswered turn is also graded and surfaced in insights so you can add the missing context.

No — it complements them. PMS templates fire inside a reservation thread; the Haven assistant answers visitors before they book and guests during the stay, on your own branded surfaces, and reports back on what it could not answer.

Next step

Automate guest messaging without going blind.

One grounded assistant across every site you run, every reply graded, and one-click fixes that heal it across all your properties.

Book a demo