Before hiring an AI consultant, a hotelier should ask what exactly will be delivered, which tool costs sit outside the fee, where guest data goes, who owns the set-up afterwards, and how hand-over works. A sound first month delivers a ranked task list, one tested workflow, a data note and a hand-over guide.
Consultant pages for hospitality tend to promise a full transformation across reservations, revenue, marketing and housekeeping. A first month cannot deliver that, and a hotel that tries will end up with five half-working experiments. The better question is what one process the consultant will make reliable, and how you will know it is reliable.
Twelve questions, grouped by what they protect
Print these and take notes during each call. Every answer should be something the consultant can show you, not just say.
Scope: what you actually get
- "Which single process will you make work first?" A good answer names a job in your hotel, such as replying to pre-arrival messages, not a category such as "guest experience".
- "What is out of scope?" Listen for specifics. A consultant who cannot name anything out of scope has not scoped.
- "How will we test it?" The answer should involve your real past messages or reviews, scored by your own staff.
Money: what it costs and what it does not
- "Is the fee fixed, and for which deliverables?" You want a list of things you can hold in your hand, with a price for the list.
- "What do the tools cost per month, separately?" The answer should be a table of products with list prices and who pays them. The consultant's fee should not hide a software resale margin.
- "What happens to costs if guest volume doubles in July?" Many tools charge per conversation or per seat, so a seasonal property needs a number.
Data: where guest information goes
- "Which accounts will guest data touch?" You own them, you can see the log, and the settings keep your content out of model training where that option exists.
- "What will never be pasted into an AI tool?" Payment card numbers, ID or passport details, health and allergy information tied to a named guest, and anything about a child. Expect a short written list.
- "How do you handle guests from the EU?" If you have such guests, data-protection law applies to how you collect and use their data. A consultant should raise this unprompted and suggest you ask a qualified adviser about the final set-up.
Hand-over: what survives after they leave
- "Who owns the prompts, templates and automations?" You do, on payment.
- "Who on my team runs this after you?" Naming a person is part of the deliverable.
- "What does the review look like at day 60?" A date, a person and three numbers to check.
If you want a general version of this for any small business, what an AI consultant cannot do for you is a useful companion, and comparing a consultant, agency and freelancer will help you decide whom to call.
What a first month should deliver
I would expect five items on paper by the end of month one, in this order of importance. They are deliberately modest.
| Deliverable | What it looks like | How you check it |
|---|---|---|
| Ranked task list | Eight to twelve tasks, each with hours a week and a risk rating | Front desk and housekeeping leads agree the hours are about right |
| One tested workflow | For example, drafted replies to routine pre-arrival messages | 50 past messages redrafted; staff score them 1 to 5 |
| Data note | One page of banned, restricted and open data types | A manager can explain it in two minutes |
| Prompt and template library | Filled in with your tone, your policies and your room names | A new receptionist can use it on day one |
| Hand-over guide and owner | Recorded walkthrough, named person, review date | The owner runs it once without the consultant |
Notice what is absent: a website chatbot, a revenue-management overhaul and a staff-wide workshop. Those can be month two or three, once one workflow is stable and the team trusts the process.
A sample plan and its numbers: a 30-room independent hotel
This is an illustration with invented figures, to show the shape of a sensible plan. The hotel receives about 90 guest messages a day across email, a messaging app and a booking-site inbox. Staff reckon about 55 percent are routine: check-in times, parking, breakfast hours, wifi, early arrival, directions to the property.
Time sum, before: 90 messages times 55 percent is about 50 routine messages. At roughly 2.5 minutes each, that is 124 minutes, or just over two hours of front-desk time each day.
Time sum, after (target): a drafted reply that a person reads and sends, at about 45 seconds each. The saving per routine message is 1.75 minutes, or about 87 minutes a day. Over a 30-day month that is roughly 43 hours.
Money: at a loaded staff cost of $18 an hour, 43 hours is about $780 a month. Tool cost for the assistant is separate: ChatGPT Business Standard is $20 a seat a month on annual billing ($25 monthly), and for three front-desk seats that is $60 a month on the annual rate. Net monthly benefit is about $720. A $3,000 fixed fee would pay back in a little over four months, if the target holds. If the first test shows only 45 percent of messages are suitable for drafted replies, the benefit drops and the fee should be reconsidered before work begins.
Here is how the four weeks might be laid out. It is a sample scope, not a promise.
- Week 1: Observe the desk for two shifts, export 200 past messages with names removed, rank tasks, write the data note.
- Week 2: Build reply templates and prompts for the top six message types; test on 50 past messages; log every failure.
- Week 3: Staff use drafts live for three days with a "review before send" rule; track time per message on a simple sheet.
- Week 4: Fix failures, write the guide, record the walkthrough, set the day-60 review.
The related tutorial on how small hotels use AI for guest messages shows what the finished workflow looks like in day-to-day use.
Keeping tool costs on a separate ledger
Hotels get burned when the consultant's fee quietly includes a tool subscription that renews in the consultant's name. Insist on a two-column budget: fees on one side, tools on the other, with the account owner named for each tool.
| Item | Illustrative cost | Billed to | Renewal |
|---|---|---|---|
| Consultant, fixed fee | $3,000 one-off | Hotel | None |
| Assistant seats, 3 at the annual rate | $60 a month | Hotel | Annual |
| Messaging or chatbot tool (if chosen) | Check the vendor page; many are priced by conversation | Hotel | Monthly |
| Staff time for testing | About 10 hours in month one | Internal | None |
| Review day 60 | 1 hour call | Included | None |
Chatbot and guest-messaging products vary widely in how they charge, so do not accept a monthly figure without the unit behind it. A free plan is not the same as a free product: some free tiers cap total lifetime conversations or delete inactive agents. The comparison of hotel AI guest messaging tools covers the products side so the consultant's recommendation can be checked against it.
A weak scope against a strong one
These two proposal snippets are invented, but typical.
Weak: "Implement AI across front desk, marketing and operations to enhance guest experience. Ongoing optimisation. $1,800 a month." There is no end point, no test, and no way to decide whether to renew.
Strong: "Draft-and-review workflow for the six most common pre-arrival message types; tested on 50 past messages; staff-scored; data note; templates in the hotel's own folder; hand-over and day-60 review; $3,000 fixed, two payments; tools billed to the hotel directly." You can hold the consultant to each phrase.
A mistake that shows how data rules matter
Consider an illustrative case. A receptionist, trying to be helpful, pastes a guest's message into a personal free AI account to draft a reply. The message includes a booking reference, the guest's full name and a note about a medical dietary requirement. Nothing visibly goes wrong, which is the problem: nobody knows it happened, and the hotel cannot show where the data went. A data note would have banned the practice and named the approved account. The fix is not blame but a simple rule, an approved account on every front-desk machine and a one-line reminder on the template page: "No names with health details. Use the approved account only."
A second mistake: the answer that was true in winter
A drafted reply tells a guest that breakfast runs from 7 to 10. That was correct until the hotel moved to summer hours, and the template was never updated. The guest turns up at 10:15 and finds the room closed. The consultant's work should include a "facts sheet" that the prompts draw from: one document with check-in times, breakfast hours, parking rules and pet policy, dated and owned by a named manager, with a quarterly review. Ask each candidate how they keep factual content current, because the AI will repeat stale facts with confidence.
Scoring candidates on one sheet
If you are talking to three consultants, a simple score sheet stops the most fluent speaker winning by default. Give each question from earlier 0 to 2 points; add the totals. Here is a filled-in illustration.
| Area | Candidate A | Candidate B | Candidate C |
|---|---|---|---|
| Scope (3 questions, max 6) | 5 | 3 | 6 |
| Money (3 questions, max 6) | 4 | 5 | 5 |
| Data (3 questions, max 6) | 6 | 2 | 5 |
| Hand-over (3 questions, max 6) | 5 | 3 | 4 |
| Total (max 24) | 20 | 13 | 20 |
A and C tie on 20, so look at the weakest area. Candidate C's hand-over score of 4 may be fixable by asking for a sample guide. Candidate B's data score of 2 is a reason to drop them regardless of price. Any single area at 2 or below is worth treating as a deal-breaker for a hotel.
One test reply you can ask to see in week two
Ask the consultant to show you a drafted reply built from your own facts sheet, with the prompt visible. This is the sort of thing to expect; the output is illustrative.
Prompt (kept in the hotel's folder):
You draft replies for the front desk of a 30-room hotel. Use ONLY the
facts sheet below. If the answer is not on the sheet, write "NEEDS STAFF"
and nothing else. Keep it under 90 words, warm, no exclamation marks.
Never include payment details. Sign off "[your name], front desk".
Facts sheet (reviewed 1 June): check-in from 15:00, check-out by 11:00,
breakfast 07:00 to 10:30, parking 12 spaces, $10 a night, first come.
Guest message: "Hi, we land early. Can we drop bags before check-in
and is parking free?"
Draft reply:
Hello [first name], thank you for getting in touch. You are welcome to
leave your bags with us from the morning, and we will store them safely
until your room is ready from 15:00. Parking is $10 a night, with 12
spaces available on a first-come basis. Early check-in cannot be
promised, but we will let you know if the room is ready sooner.
[your name], front deskHere is what I would fix. The draft promised bag storage "from the morning", which is not on the facts sheet; a stricter prompt should have returned "NEEDS STAFF" for that part. This is the exact kind of failure a test on 50 past messages is meant to catch, and the failure log should record it with the correction. If a consultant shows you only flawless examples, they have not tested enough.
What staff change-over does to a hotel's AI set-up
Hospitality has seasonal and high-turnover teams, so a set-up that depends on one enthusiast is fragile. Ask how the guide copes with a new receptionist arriving in May. A workable answer is a one-page card at the desk with three rules (use the approved account, check the facts sheet date, send only after reading), plus a 20-minute recorded demo. Also ask who removes a leaver's access: it belongs in the existing staff exit checklist, not in the consultant's scope.
Contract points to check before you sign
- Payment in stages, tied to deliverables in the month-one table, not paid entirely before work starts.
- A named stop point. If the week-two test fails badly, you can stop and pay only for work delivered.
- Insurance and liability. Ask whether the consultant carries professional cover and what the agreement says about errors. Have a qualified adviser read anything that touches guest data.
- Confidentiality covering room rates, occupancy and guest lists.
- No exclusivity. Nothing should stop you using another supplier for month two.
What to ask about integrations
Most small hotels do not need an automation connecting the property management system to an AI tool in month one. If a consultant proposes one, ask three things: which system, how access is granted, and what happens if the connection breaks. A failed link in the middle of a Saturday check-in is a bigger problem than a slower manual process. Start with exports and drafted replies that a person sends; integrate only when the manual version is stable and the time saving justifies the risk.
Prepare your own side first
A consultant's month is only as good as what you hand them in week one. Before the first call, gather four things, and tell candidates you have them, because it changes the quality of their answers.
- Two hundred past guest messages with names, booking references and contact details removed. Mix email, messaging-app and booking-site messages, and include the awkward ones.
- A facts sheet draft. Even a rough page of check-in times, breakfast hours, parking, pets, cancellation terms and local transport. Writing it often exposes that staff give different answers to the same question.
- A time count. Ask the desk to note, for three ordinary days, how long they spend on messages. A rough tally on paper is enough.
- A named owner. One person, with two hours a week free for the month, who will test drafts and run the process afterwards.
If you cannot supply the named owner, say so openly. It changes the plan: a consultant may suggest a lighter workflow, or a longer, slower engagement, instead of the standard month. Hotels that skip this step often discover in week three that nobody has time to review the drafts, which is the most common reason a first workflow stalls. Gathering the material yourself also lets you judge candidates by what they do with it: the good ones ask to see the awkward messages first.
Red flags in the sales conversation
- A guaranteed percentage. "Cut your workload by 40 percent" cannot be promised without seeing your messages. Ask them to show the sum.
- A demo on a different property. A smooth demo on invented data tells you nothing about your messages.
- Pressure to sign this week. A genuine first project does not expire on Friday.
- No mention of staff. If the plan does not say who on your team uses the tool and what they stop doing, it will not stick.
- A tool bundled into the fee with no list price. Ask for the vendor price and the account owner.
Checking it worked on day 60
Use three numbers, all countable from the desk sheet: minutes per routine message, share of messages sent from a draft, and the number of guest complaints that mention a wrong answer. If time per message has not fallen by at least a third, find out whether staff are using the drafts at all. If wrong-answer complaints have risen, pause the workflow and fix the facts sheet before anything else. A good consultant will propose this check without being asked. The cost-and-scope logic behind a retail version of this list is in the retail hire-or-not checklist.
If you would rather map this with someone before you speak to any consultant, that is what my AI implementation consultation covers, via the contact page.
Questions hoteliers ask before signing
Should a consultant get access to our property management system?
Not at the start. Begin with exports or sample data with guest names removed. Give system access only for a named task, through a separate login with limited rights, and remove it at hand-over. Ask your PMS vendor whether their terms restrict third-party access, since some contracts do.
Is a hotel too small to need an AI consultant?
A property with one person who handles messages and reviews can often start with a business-grade assistant account and a good prompt library. A consultant becomes worth the fee when guest data, several channels or staff change-over make mistakes costly, or when nobody on the team has time to test.
Who should own the finished prompts and automations?
The hotel, outright, on payment. Ask for them in a shared folder you control, with a short guide on how each one works. If the consultant keeps them inside their own account, you will lose the work the day the relationship ends.
What if the consultant recommends a tool they resell?
That is allowed, but it must be disclosed in writing, with the price you would pay buying direct. Ask for one alternative they considered and why it lost. A consultant happy to show their comparison is usually sound.
Further reads
- Hotel Guest Complaints Handling: Service Recovery Scripts With AI — A guest-facing job to consider once messages work.
- How Can AI Help Hotels? Housekeeping Checks and Maintenance Requests — A second job to consider once guest messages work.
- Hotel Guest Feedback: Turn Post-Stay Surveys Into Fixes With AI — Turn survey comments into fixes, a good month-two task.
- ChatGPT Prompts for Hotels: 10 for Guest Emails and Rate Copy — Ready prompts for guest emails and rate copy.
- WhatsApp for Small Hotels: Meta's AI Rules, Templates and Handover — Messaging-app rules to check before a consultant builds on it.
- Hotel Email Marketing With AI: Past-Guest Segments and Subject Lines — A later-month task that uses past-guest data.
- Best AI SEO Agency for Hotels? Questions to Ask Before You Sign — Judge an AI SEO agency for a small hotel by its proof, contract terms and reporting, and compare it with doing the steps yourself.
- Channel Manager for a Small Hotel: Cost and What AI Adds — Price a channel manager three ways, work out the break-even for a 12-room hotel and test whether the AI extras are worth paying for.
- AI Agents for Hotels: What They Can Do Today and Where They Fail — What a hotel AI agent can handle now, where it fails on bookings, guest messages and housekeeping, and the tests and hand-off rules to set first.
- Smart Room Automation Through AI in Hotels: What a Small Property Can Add — What a small hotel can add to its rooms first: thermostat setback linked to the booking system, window and door sensors, and clear guest consent.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Clutch AI pricing guide (September 2026) and Contra's hiring page for market rates; OpenAI and Anthropic business plan pricing pages.