To start using artificial intelligence at reception, choose a common, clearly defined task, prepare the information needed and compare the work with and without AI assistance. Measure the time until the reply is ready to send, including corrections, and check that the information is correct. You can organise this trial with your team before extending AI to other tasks.
An enquiry arrives while you are welcoming a family. The guest wants to know which accommodation is available, how parking works and whether arrival after dinner is possible. A draft can help organise the response. To assess its usefulness, however, you need to see how much work remains before sending it.
Choose a reply you know well
A good starting point is an enquiry your staff handles often: for example, a stay enquiry with questions about services. Choose a case where you can easily recognise a complete, correct reply.
Write the trial’s aim in one sentence: “Prepare a clear reply to the guest’s enquiry, with all information confirmed and a final staff check.”
For the first trial, leave out situations requiring a specific management decision, such as a disputed refund or an exception to commercial terms. AI can also support writing in complex situations, but starting with a defined task makes it easier to understand what works.
Here we are discussing help with drafting a reply. Website chatbots, telephone assistants and automatic sending are different processes requiring their own trials. For an overview of data and tools, read about CRM and artificial intelligence for hotels.
Prepare the information before drafting
Check that the information used in replies is up to date: opening hours, services, terms, accommodation features and arrival arrangements. For prices and availability, use your property’s current operational source.
A general parking description, for example, may not explain whether a space must be reserved, whether there is a charge or whether parking is available for all accommodation types. Both the writer and the reviewer need these details.
In the composition assistant in FastReplyMail, AI prepares a draft using the property’s Knowledge Base and available integration data. Staff can review, edit or regenerate the text before sending. The Knowledge Base is the hotel information repository consulted by the assistant: its usefulness also depends on its contents and how up to date they are.
Before starting, decide who checks this information and who can confirm answers when details are missing. Use test enquiries without personal data, or the environments and procedures authorised by your property.
Compare similar tasks under the same conditions
Prepare a small set of comparable enquiries: the same language, similar complexity and equivalent information available. You could start with ten test cases. This is a practical number for organising the work, not a threshold that makes the result statistically representative.
Measure the usual process and then the assisted process, keeping the same endpoint: a checked message ready to send. In both cases include reading the enquiry, finding information, writing and reviewing. For AI, include generation time, any regenerations and edits too.
If you repeat exactly the same enquiries, whoever handles them second may already remember the answers. Alternate the order or use equivalent cases, noting this limitation. Record interruptions and particular difficulties too: a phone call during a reply does not measure the assistant’s performance.
This comparison helps you make a decision about your process. On its own, it does not show what results you will get on every shift or in peak season.
Measure time, corrections and quality
For each enquiry, complete a worksheet with these fields:
- Case and task: an anonymous code, language and question type.
- Method: usual process or AI-assisted process.
- Total time: from reading the enquiry to a checked, ready-to-send reply.
- Corrections needed: information you had to add or change.
- Completeness: have all the guest’s questions been answered?
- Outcome: ready after checking, needs reworking or requires a manager.
- Context: interruptions, missing data and other relevant conditions.
Separate stylistic edits from substantive corrections. Shortening a sentence and correcting a cancellation condition carry different weight. Check dates, guest numbers, accommodation, prices, services, terms and the proposed next step.
To interpret the result, compare mean or median times for equivalent cases and review the more difficult cases too. Record initial Knowledge Base preparation, training and maintenance time separately: they count in the overall assessment, even if they do not recur with every reply.
Quickly generated text is useful when it helps you reach a correct reply with a manageable overall workload.
Open and download the evaluation worksheet
The fillable worksheet has ten spaces for comparing similar enquiries: usual and AI-assisted times, completeness, corrections, outcomes and notes. The summary calculates time differences only for complete pairs; quality remains for the team to assess.
Open the fillable evaluation worksheet.
Download the blank HTML worksheet and open it in your browser, even offline. No Excel or account is required.
Data is neither sent to the website nor saved automatically. Use “Save completed worksheet” to keep an editable copy before closing the page; you can also print it or save it as a PDF from the browser. Use anonymous codes and do not enter guests’ personal data.
An example of how to interpret the results
Consider this hypothetical example, which does not represent measured FastReply results: for one enquiry type, the usual process takes eight minutes on average. With the assistant, generation, checking and editing together take six minutes.
The difference is two minutes per enquiry, or 25% of the original time. Across twenty genuinely comparable enquiries, that would be forty minutes. Before using this estimate, check that the sample and final quality are comparable and account for setup, maintenance and usage costs.
If some replies contain substantive errors, the time figure alone does not make the trial successful. Identify the information or steps that need correcting, then repeat the comparison.
Time freed up can be used for other tasks. A financial return also requires measured financial costs and benefits; attributing increased bookings to AI requires further investigation.
Decide what to improve before expanding use
At the end of the trial, bring together those who prepared and checked the replies. Which cases needed less work? Which errors recurred? Did the problem stem from missing information, unclear instructions or a task requiring more judgement?
Agree in advance on the conditions for continuing: correct essential information in the final reply, staff review every time and a review workload the team can manage. Then decide whether to expand the trial, update the information or keep the usual process for some cases.
Measurement also matters in industry research: in the sponsored study h2c AI Opportunity 2026, only 13% of participating chains report a measurable financial return from AI initiatives. This figure refers to that sample, not to Italian hotels as a whole or your property’s results.
Start by choosing a common enquiry, preparing the information and completing the worksheet with a colleague. You will have a practical basis for deciding where the assistant helps and which checks to retain.
A practical checklist for your first trial
Use this checklist with the people preparing guest replies. For each item, mark done, to complete or to correct. Copy it into your working notes and use it alongside the evaluation worksheet above.
- Define the task. Choose a common enquiry type and write down what a complete reply should contain.
- Prepare the information. Check that services, hours and terms are current; identify the current source for prices and availability.
- Choose comparable cases. Use enquiries of similar complexity and language, without personal data or handled within the property’s authorised environments.
- Assign the reviewer. Identify who will reread the draft and who can help when confirmation is needed.
- Measure all the work. Record time from reading the enquiry to a ready reply, including generation, checking, edits and regeneration. Note interruptions separately.
- Check before sending. Verify dates, guest numbers, accommodation, prices, services and terms. Make sure every guest question has been answered.
- Distinguish corrections. Separate tone edits from errors or missing information; record cases needing rework too.
- Decide the next step. Compare time and quality, including preparation and costs. Decide whether to expand the trial, correct information or retain the usual process for some cases.
If essential information is unconfirmed, finish checking before sending. An open checklist item shows where to act before expanding the trial.
FROM READING TO PRACTICE
Discover how the FastReplyMail AI assistant works
See how your property’s Knowledge Base and available data help the assistant prepare a Composer draft, which staff check and personalise before sending.
Discover the FastReplyMail AI feature →