Robot, Kiosk and Avatar Receptionists: How Much of the Front Desk AI Handles

Four ways AI reaches the front desk

Receptionists answer the phone, greet visitors, take payments, keep the calendar and field complaints; O*NET lists those among the role's core tasks, and the BLS figures on O*NET count 947,500 people in the job in the US in 2025.¹ AI arrives at that desk in four main set-ups. A character or humanoid robot stands at the counter and talks to visitors. A self-service kiosk does the transaction: it finds the booking, takes payment and issues a key, sometimes with speech or face recognition. A digital human is an animated face on a screen that listens and answers. And a remote operator is a real person working several sites at once through an on-screen avatar or video link: the software supplies the on-screen character, but a person does the answering. Separately, voice agents can answer the phone. Most real deployments mix them: at Henn na Hotel the robot greets and the kiosk transacts.⁴

Robot front desks: the Henn na Hotel story

Henn na Hotel opened in July 2015 beside the Huis Ten Bosch theme park in Nagasaki with 72 rooms.¹⁶ Its developer said the aim was a low-cost hotel, with automation and service robots meant to cut labor costs; robots from Kokoro, Aldebaran Robotics and Yaskawa Electric were to work the front desk, porter service, lockers and room cleaning, with check-in and check-out done by guests themselves.¹⁷ In December 2017 the hotel's general manager said it ran 144 rooms around the clock with 7 staff, 5 on any given day, against 30 at opening; guests checked in at a counter terminal by saying their name into its microphone, which pulled up the booking for them to confirm and sign in about 20-odd seconds.⁴

The robots drew most of the problems. In January 2019 the hotel was reported to have retired more than half of its 243 robots after complaints from staff and guests; reported problems included an in-room assistant that couldn't answer basic questions and a check-in dinosaur that couldn't scan passports.⁹ The hotel's own account, as reported by Nikkan Kogyo Shimbun, was more measured: robot types fell from a peak of 27 to 15, with machines installed four years earlier no longer satisfying guests, some limited by water or weather, and guests arriving with higher expectations than the robots met. The third building, added in 2018, took only robot vacuum cleaners.¹⁶ The brand kept growing all the same, from four hotels in December 2017⁴ to a much longer list today: as of October 2026 the H.I.S. group site lists 22 Henn na hotels, with a dinosaur robot assisting check-in at 10, a humanoid robot at 5 and a projected "Holograms of Light" faceless check-in at 4.³ At the Ginza hotel, the operator says human staff are always at reception.¹⁸

The kiosk does the transaction

Strip away the robot and most of the front-desk work is a self-service kiosk. At Henn na Hotel Tokyo Ginza, guests are welcomed by two humanoid robots and human staff, then make payment and receive their room key at the kiosk.⁶ The AI part is narrow: recognizing a spoken name, as at the first hotel's terminal, or a face, which the first hotel planned in 2015 as an alternative to a room key.⁴¹⁷ The "Holograms of Light" hotels replace the robot with projected characters at check-in.¹⁹ That is why payment and check-in score as tasks AI does in production, while the robot on the counter is mostly the greeting.

Pepper: the humanoid receptionist that stopped production

SoftBank's Pepper was the best-known reception robot. Its business model, Pepper for Biz, came with a reception app that emailed an employee when their visitor arrived, for ¥55,000 a month on a 36-month rental (July 2015 prices).¹⁴ The consumer model cost ¥198,000 plus compulsory subscriptions of ¥14,800 and ¥9,800 a month over 36 months.²⁰ In June 2016 two Belgian hospitals, AZ Damiaan in Ostend and CHR Citadelle in Liège, announced Pepper robots that would greet patients, take their appointment details and give directions or walk them to the doctor, at €30,000 per robot for a version with its software maker's support.⁵ Reuters reported in 2021 that SoftBank had stopped producing Pepper in 2020 after about 27,000 were made, with sources citing limited functionality and unreliability.²¹ The Pepper cases on this page are historical.

Information robots: what a hospital found when it checked

Peking University Shenzhen Hospital introduced a consultation robot, "Xiaoyi", on March 14, 2018. A China Daily-run government site reported that it averaged 1,090 interactions a day and held answers to 565 frequently asked questions, and quoted the hospital's deputy director of nursing as saying three such robots served patients.¹⁰ A study by clinicians at the hospital, published in Cureus in 2021, examined an intelligent guidance robot in a Shenzhen outpatient lobby over 20 days in November 2019; the authors did not name its maker or model, so the paper does not confirm it was Xiaoyi. Of 25,526 encounters, the robot answered 12,543 and gave a default reply, "I am still learning. Please ask the nursing staff", in 13,183. Unanswered questions were in languages other than Mandarin and English, numbers, play or nuisance, or unrelated to the hospital, and the authors found that noise degraded answers. Among the answered questions, manual checking found 92.32% accuracy, against the manufacturer's claim of near-perfect accuracy. The authors declared no competing interests.¹¹

Digital humans: speech in, a face and voice out

A digital human runs a pipeline: a microphone and speech recognition turn the visitor's words into text, a dialogue engine (an intent-based assistant, or in newer systems a large language model) picks the answer, and speech synthesis drives an animated face that speaks it. At Dallas Fort Worth International Airport, "Digital Iris" was built from Soul Machines' animated face and emotion detection plus IBM's Watson Assistant for understanding and answering, on kiosks with a directional microphone, speaker and camera.²² The airport tested her near gate A-34 in Terminal A until June 25, 2021.²³ Soul Machines said in October 2022 that she had been in Terminal B since June 2022 and had handled "thousands of conversations" about flights, gates, restaurants and maps.²² We found no later status, and receivers were appointed to Soul Machines on February 5, 2026.²⁴ In this set-up the face, the voice and the knowledge base can all come from one vendor, as they did at DFW.

Remote people behind avatars

The fourth set-up puts a person back at the desk, from far away. Since November 2022 Lawson has piloted avatar staff with AVITA: remote operators appear as avatars on in-store monitors, guide shoppers and explain the self-checkouts, and one operator can serve several stores at once. By July 2026 the pilot covered 78 stores with about 70 operators, and Lawson said self-checkout use rose by 15% or more and register work fell by about 1.5 hours or more a day per store.¹⁵ Timeleap, the maker of the RURA remote reception service, said in December 2024 that Hokkaido Airports would use it at New Chitose Airport's 12 information counters after a one-week trial.²⁵ The airport operator's own information-counter page doesn't mention it, so that case remains a vendor claim.²⁶

The phone line, and where every set-up fails

On the phone line, AI already does the switchboard's first step at some employers. At Tampa General Hospital, Hyro's voice agents answer calls and route patients to the right service, with the first phase focused on transfers that pass on the caller's context and on appointment management; the partners reported abandonment and wait times falling within two weeks.² Off-the-shelf AI receptionists sell the same: a warm transfer that tells staff who is calling and why, bookings, and texted summaries.⁷ The limits repeat across set-ups. Noise and off-script questions defeat speech systems, as the Shenzhen data showed.¹¹ Robots date quickly, and some can't work in water or bad weather.¹⁶ Products end: Pepper's production stopped, and Digital Iris's maker went into receivership.²¹²⁴ And complaints stay human: even NextPhone's own guidance has its AI hand angry callers to a person and never promise refunds.⁸ BLS figures on O*NET project a decline of 1% or more in receptionist employment from 2025 to 2035, and 105,100 projected job openings over the same period; those projections don't say what share of the change, if any, comes from AI.¹