In a 30-day sample of 900,000 guest messages on Akia, about 4 in 5 were routine questions with one right answer, like the wifi password or where to park, and an AI agent can handle those from day one. The rest split into requests that need an agent trained on the property's own policies, plus a small set of complaints, refunds, and safety issues that should always go to a person.
The short version.
- Tier 1 (about 80%) is ready for AI on day one. Wifi codes, parking, check-in times, breakfast hours, and policy questions all have one right answer that's already written down.
- Tier 2 needs an agent trained on your property. Multi-step requests and property-specific judgment calls are the work a chatbot can't do and an agent that has learned your policies can.
- Tier 3 stays with your team on purpose. Complaints and refund disputes go to a person whenever reputation, safety, or money is on the line.
Why most guest messages don't need a person.
Most guest messages ask for something already written in a binder, a PMS field, or on the property's website. Picture a guest texting at 9:47pm for the wifi password, another asking where to park the rental car three minutes later, and a third asking for the gate code at 11:14am, two minutes after the front desk finally sat down.
None of those messages needed a person, and the pattern held across the thousands of properties in our sample. Guests ask the same questions over and over, and each one has a single correct answer.
How we sorted 900,000 guest messages.
We classified every message on two axes, which together form a 2x2 we call the Akia Message Tier framework.
- Factual or judgment. A factual message has one correct answer in a knowledge base, policy doc, or PMS field. A judgment message depends on context, history, or the specific guest writing in.
- Low or high stakes. Low stakes means a wrong answer costs little. High stakes means reputation, safety, money, or a guest who's already upset.
Anything factual is AI work at any stakes, and so are low-stakes judgment calls that a clear policy can settle. High-stakes judgment is the one quadrant that stays human.
Figure 1: The Akia Message Tier framework. Factual vs. judgment runs along the horizontal axis and low vs. high stakes along the vertical. Three of the four quadrants are AI work, and only high-stakes judgment stays with people.
The AI side splits into two tiers, the work that's ready on day one and the work that gets sharper as Akia learns your property. Add the human tier and you get three.
Figure 2: Where guest messages land across the three tiers. Tier 1 holds most of the volume, from logistics and service requests to compliments and local recommendations. Tier 2 covers multi-step requests and property-specific exceptions, and Tier 3 holds complaints, refund disputes, and sensitive escalations. Source: Akia inbound message data, 30-day sample.
Tier 1: Which guest messages can AI handle on day one?
About 80% of guest messages are Tier 1, which means the answer already lives in the PMS or on the property website and an AI agent can reply from the day it's switched on. Four categories make up most of it.
- Logistics and arrival. Check-in times, addresses, gate codes, wifi passwords, and parking. Every property's knowledge base already has these answers.
- Service requests. When a guest asks for extra towels, Akia creates the housekeeping task and confirms with them before anyone at the desk would have glanced at their phone.
- Compliments and thanks. "Loved our stay" doesn't need a person, though it does need a reply so the guest doesn't feel ignored. Your team stops typing "we're so glad!" forty times a week.
- Time-flex requests. Early check-in, late checkout, and an extra night usually get a yes for a fee. Akia quotes the policy, applies the charge, and closes the upsell while the guest still has their phone in hand.
Amenities, local recommendations, booking policy, and pet and kid questions land in Tier 1 too.
Tier 2: Which guest messages need an AI agent trained on your property?
Tier 2 is multi-step requests, property-specific exceptions, and edge cases. To handle them, Akia has to learn your property's policies and how your team likes things escalated, and this tier is where an AI agent and an FAQ bot stop looking alike.
Take a guest who writes, "We had a really late night. Any chance we can stay until 2pm before heading to the airport?" A chatbot spots the word "checkout" and either replies with the standard 11am time or gives a flat yes or no without checking anything. Akia reads the reservation, checks whether the room is booked again that afternoon, and applies the late-checkout policy. Once the guest confirms, she updates the housekeeping schedule so the room isn't flagged for an 11am turn.
Two more Tier 2 messages show the same pattern:
- "We're running 30 minutes late, will the kitchen still be open?" Akia checks the kitchen hours against the guest's ETA, offering an alternative if it'll be closed and holding a table if it won't.
- "My partner has a peanut allergy. Is breakfast safe?" Akia looks up the dietary information and checks with the kitchen contact, then escalates if the answer still isn't definitive.
These messages quietly eat up the front desk's day, because each one means pulling from a few different systems and making a small judgment call.
Figure 3: How Akia answers a late-checkout request. The guest asks, "Can we stay until 2pm before our flight?" Akia reads the PMS (Room 207 vacant at 1pm), the reservation (Smith, party of 2), the $30 late-checkout policy, and the calendar (next arrival 3:30pm), then sends one reply that confirms the time, applies the charge, and wishes the guest well.
Tier 3: Which guest messages should always go to a person?
Active complaints, refund disputes, in-progress room issues, and any message that names a staff member or describes how a guest was made to feel should always escalate to a person. Each of these is a judgment call with real stakes, and a fumbled reply is how a property ends up with an angry review or a refund it didn't need to give.
When Akia hands one of these off, your team sees what's been said, what's already been tried, and what the guest needs next.
What can an AI agent do that a chatbot can't?
A chatbot matches keywords against canned replies. An AI agent like Akia works from live property data and takes action across your systems.
- Pulls live context. Akia reads the PMS, the reservation, the property's knowledge base, and the room status in real time.
- Handles multi-step work. In one exchange, she can check housekeeping availability, apply a policy, update the PMS, and reply to the guest.
- Reads the situation. "What's my door code?" and "I'm at the door and the code isn't working" share most of their words, so a chatbot sends the code both times. Akia sees a guest stuck outside and treats it as urgent.
- Escalates with full context. When a person is needed, Akia passes along a summary and the steps already taken, so your team can help the moment they pick it up.
How should a property start automating guest messages?
Start with the type of question that comes up most, which for most properties means logistics and amenities like wifi, parking, and check-in times. Then layer in policy requests such as early check-in and late checkout, an order that gets you most of the time savings in the first few weeks.
Set firm rules so complaint, refund, and health-and-safety language always reaches a person, however confident the AI's reply would be. When you compare vendors, ask what your team actually sees when the AI escalates.
Frequently asked questions.
What percentage of guest messages can AI handle?
Across a 30-day sample of 900,000 guest messages on Akia, roughly 80% were routine questions with a single correct answer, such as wifi, parking, and check-in times. An AI agent can answer those from the first day it's switched on.
What's the difference between an AI agent and a chatbot for hotels?
A chatbot matches keywords to canned replies. An AI agent reads live context from the PMS and the reservation, takes multi-step actions like applying a late-checkout fee, and escalates to a person with a full handoff summary when it can't resolve a request.
Which guest messages should always go to a human?
Active complaints, refund disputes, and safety issues should always escalate to a person. A good AI agent passes along a summary of the conversation so staff can step in without rereading the thread.
Should hotels automate guest messaging?
Yes, starting with the most repetitive questions. Automate logistics, amenities, and policy questions first, then judge any vendor by how cleanly it hands escalations to your team.
