Selection guide

Choosing a front-desk humanoid: the seven questions to ask

The specific procurement questions that separate a hospitality-ready platform from an industrial or research one. Ask each of them at the sales stage; the vendor's answers will do more of the shortlisting than your requirements matrix will.

A hotel general manager considering a humanoid platform for a front-desk role in 2026 has, in principle, eight or nine platforms to choose from. In practice, most of those platforms will disqualify themselves within the first ten minutes of a serious conversation. The remaining shortlist is typically two or three platforms. Getting to that shortlist is a matter of asking the right questions, in the right order, and being willing to disqualify a vendor when the answer is unsatisfactory.

The seven questions below are those we ask, in this order, when we run a selection engagement for a hotel operator. Each one is a genuine disqualifier if the answer is wrong.

1. Which spoken languages are the platform's voice interface certified in?

Every hotel that operates internationally will need the platform to communicate in at least three languages. Most Asian and European hotels will need five. Some Gulf and Southeast Asian properties will need seven or more. The specific languages your property serves are non-negotiable; the platform must be certified in all of them at the vendor's shipping specification, not on a promised roadmap.

The important word is "certified." Every platform vendor will claim its voice interface supports every language on paper. The certification requires that the vendor has tested the language, published a phoneme coverage report, and warrants the interface's accuracy for guest-facing use. Certified language coverage is usually documented; roadmap coverage is usually not.

2. What is the platform's hearing accuracy at typical lobby ambient noise levels?

A hotel lobby, at check-in-heavy hours, produces ambient noise levels in the range of sixty-five to seventy-five decibels. Some are louder. Ask the vendor for the platform's speech-recognition accuracy at seventy-two decibels of ambient noise, from a guest speaking at a normal conversational volume, at a distance of one metre. This is the operating profile the platform will encounter every day. If the vendor's accuracy figures are quoted at "quiet indoor" conditions only, you do not have data suitable for your property.

3. Which atelier or contract manufacturer has already dressed this platform for a hotel deployment?

The most under-appreciated procurement question. A platform that has been dressed for a hotel deployment already carries the fit data, the fabric performance data, and the atelier relationship that your project will need. A platform that has not been dressed for a hotel deployment is asking you to be the pilot programme for its dressing pipeline. This is not by itself a disqualifier, but it changes the risk profile of the deployment substantially, and it should change the price you pay for the platform.

4. What is the maximum guest-interaction latency, from prompt to response?

Guests in a hotel lobby expect a response to a spoken question within roughly one second. Two seconds is uncomfortable. Three seconds is disqualifying in a guest-facing role. Ask the vendor for the ninety-fifth-percentile response latency, measured from the end of the guest's utterance to the start of the platform's response, at the sensor conditions your property will present. Anything over one point four seconds is a red flag for a front-desk role.

5. What is the platform's uninterrupted operating time between charge cycles, in the specific role you are procuring?

Vendor-published battery life is measured in idle conditions. Front-desk service is not idle. The platform will be in continuous interaction for hours at peak times. Ask the vendor for the specific figure for a continuous front-desk workload profile. If they cannot produce it, ask them to test at your property before you finalise. Real front-desk platforms need at least four hours of continuous operation between charges; six is comfortable; three or less will not survive an evening shift.

6. What is the failure recovery model?

Every humanoid platform will occasionally fail in front of guests. The important question is what happens next. Does the platform announce the failure gracefully and route the guest to a human staff member? Does it freeze and require staff intervention? Does it retry silently, causing an awkward pause the guest will remember? Ask the vendor to describe the failure recovery model in detail and, if possible, to demonstrate it live. A hospitality-ready platform will have a considered answer; a platform being sold to hotels for the first time will not.

7. What is the second-year cost of ownership?

Every vendor will quote you the first-year cost. Not every vendor will quote you the second-year cost. In the second year, the platform will need maintenance, potentially firmware upgrades that require on-property support, atelier refresh for the dressing, and continued service coverage. Ask for the year-two figure in writing before you sign year one. The delta between year one and year two is often the difference between a viable deployment and a mistake.

The seven disqualifiers, in one place

• Language: not certified in every language your property serves
• Hearing: not accurate at your typical lobby ambient noise level
• Dressing: no delivered hotel deployment with an existing atelier partner
• Latency: over one point four seconds at the ninety-fifth percentile
• Endurance: less than four hours continuous under a front-desk load
• Failure: no considered recovery model, or no live demonstration of it
• Year two: unwilling to quote, or the quote is unacceptable

Any one of these is a genuine disqualifier. Two of them together should end the conversation. Three should send you back to the directory.