What 15,910 Hotel Calls Revealed About AI Voice Agents

When a hotel group removes an outdated phone menu, you'd expect callers to have an easier time reaching the front desk.

But what happens when the numbers suddenly get worse?

That's exactly what happened during an AI voice deployment across a European hotel network.

After removing its interactive voice response (IVR) menu, the group saw its empty call rate — calls ending before a meaningful conversation with the AI — jump to 19.5%.

At first glance, it looked like a problem with the AI voice agent.

The reality was more interesting.

The old phone menu hadn't prevented callers from abandoning their calls. It had prevented the hotel from seeing some of those abandoned calls in its AI reporting.

This discovery emerged from Polydom's analysis of 15,910 incoming calls handled by Una across 15 AI voice agents over 175 days.

The findings offer a practical lesson for hotel operators: before judging whether an AI receptionist is working, make sure you understand which callers your reporting actually includes.

The Study: 15,910 Calls Across a European Hotel Network

Over approximately six months, Polydom analyzed phone activity from a European hotel network using Una by Polydom, a voice-first AI Digital Employee.

The dataset covered:

• 15,910 incoming calls handled by Una.
• 15 AI voice agents deployed across the hotel network.
• 175 days of recorded activity.
• Approximately 91 calls per day across the network.

The analysis examined empty calls, changes in call-routing arrangements, and differences between hotel types.

An empty call was defined in the deployment reporting as a call where the caller disconnected before having a meaningful conversation with the AI.

That distinction matters because an empty call isn't automatically a failed booking.

Some callers may have dialed accidentally, decided they didn't need assistance, or disconnected for reasons unrelated to the hotel's service.

Others may have been potential guests who encountered enough friction to abandon the interaction.

The data cannot establish the intent behind every disconnected call.

But it can reveal where callers leave the process — and whether the hotel is measuring those departures consistently.

The Unexpected Finding: Removing IVR Made Performance Look Worse

For several months, the hotel network's reported empty call rate remained relatively stable, generally between 12% and 15%.

Then, in mid-February 2026, the group removed its traditional IVR menu.

The previous system asked callers to select a language before continuing — for example, pressing one for Czech or two for English.

The change was intended to simplify the experience by allowing callers to reach Una directly.

Instead of navigating a menu, guests could begin speaking to the AI voice agent.

But shortly after the change, the reported empty call rate increased to 19.5%.

That raised an obvious question.

Had removing IVR damaged the guest experience?

The initial numbers appeared to suggest it had.

However, the change had also altered which calls were visible to Una's reporting.

The Hidden Filter in Hotel Phone Systems

Before IVR was removed, callers first encountered the traditional telephone menu.

Some disconnected before selecting a language or reaching the AI.

Those callers never entered the AI voice agent's conversation records.

As a result, the AI's empty call rate described only the people who had already passed through that first stage.

When IVR was removed, callers reached the AI directly.

Disconnects that previously happened before the AI were now more likely to appear in its reporting.

The measurement boundary had changed.

This created a counterintuitive outcome: the system appeared to perform worse partly because it had gained visibility into callers who previously went unrecorded in the AI dataset.

The findings illustrate a broader analytics problem.

If one system reports only calls that reach an AI agent, while another measures every call entering the telephone infrastructure, their abandonment rates are not directly comparable.

A lower reported abandonment rate doesn't necessarily mean the overall caller experience is better.

It may simply mean that part of the journey is missing from the measurement.

Spa Hotels and City Hotels Behaved Differently

The network-wide average was only part of the story.

When the analysis separated properties by type, two distinct patterns emerged.

Spa and Wellness Hotels: Relatively Stable Call Abandonment

Across the observed period, spa and wellness hotels recorded an average empty call rate of approximately 11.7%.

In February, their rate was approximately 12.7%, showing comparatively little change after the IVR adjustment.

One possible explanation is the nature of their audience.

Spa properties may attract more repeat or locally based guests who are already familiar with the hotel and its services.

A caller contacting a wellness hotel to book a treatment may also have a more specific reason to complete the conversation.

These are plausible interpretations of the observed difference, not individually verified reasons for each caller's behavior.

Nevertheless, the relatively stable results suggest that the telephone experience was not affecting every property type in the same way.

City and Business Hotels: A Much Larger Shift

City and business hotels showed a different pattern.

Their overall empty call rate was approximately 16.7%.

In February, it reached 26.5%.

That meant more than one in four calls ended before a meaningful conversation with the AI, under the reporting definition used.

Compared with spa hotels, these properties experienced a considerably higher rate of early disconnects.

Why?

A likely contributing factor was the difference in guest populations.

City and business hotels often serve international travelers, business visitors, and guests who may be unfamiliar with the local language.

If the initial telephone experience doesn't immediately communicate that service is available in English or another familiar language, some callers may disconnect.

The original analysis did not establish language as the cause of every abandoned call. The IVR change also affected which calls were counted.

But the difference between property types raised an important operational hypothesis:

The language a guest hears at the beginning of a call may influence whether they remain on the line long enough to receive assistance.

Why the First Seconds of a Hotel Call Matter

Hotels invest considerable effort in optimizing their websites and booking engines.

They test booking forms, simplify checkout processes, and try to reduce the number of steps between interest and reservation.

Yet the telephone experience can receive much less attention.

For a guest calling directly, the opening greeting serves a similar purpose to the first screen of a booking website.

It tells the caller whether they've reached the right place, whether the hotel can assist them, and what they should do next.

A long menu may introduce unnecessary steps.

A greeting in an unfamiliar language may leave an international traveler uncertain about whether they can communicate successfully.

A delay before any response may create another reason to disconnect.

None of these conditions guarantees abandonment.

But they represent friction at the beginning of a potentially valuable guest interaction.

For hotel operators, this creates a measurement opportunity.

Rather than evaluating only the AI's ability to answer questions, operators should examine the entire path from the initial incoming call to a meaningful conversation.

That includes the telephone infrastructure, routing rules, initial greeting, language handling, and any transfer or escalation steps.

Does a Dropped Hotel Call Mean Lost Revenue?

Not necessarily.

An abandoned call represents a lost or interrupted conversation, but it doesn't establish that a booking would otherwise have occurred.

Some callers may have been existing guests seeking information. Others may have been suppliers, employees, or people calling the wrong number.

However, unanswered or prematurely disconnected calls can create a commercial risk when prospective guests are trying to make reservations.

At an average of approximately 91 calls per day across the network, a five-percentage-point difference in the empty call rate represents roughly 4–5 additional early disconnects per day.

Over 30 days, that is approximately 135 calls.

This is an illustration based on average call volume, not a measured number of lost bookings.

The actual revenue impact depends on who was calling, why they called, whether they contacted the hotel again, and whether any reservation would have been completed.

A guest who cannot complete a direct telephone inquiry might later book through an online travel agency, contact another property, or abandon the search altogether.

Without linked booking and caller-outcome data, the analysis cannot determine which outcome occurred.

The strongest conclusion is not that every dropped call costs a hotel a reservation. It's that early call abandonment can conceal commercially valuable demand — and operators need better visibility to distinguish that demand from other traffic.

The Next Experiment: A Bilingual Greeting Instead of IVR

The findings prompted the Polydom team to explore a different approach.

Rather than restoring the traditional IVR menu, the deployment moved toward testing a short bilingual greeting.

The proposed opening was straightforward:

"Dobrý den, welcome to [Hotel Name]. How can I help you?"

The idea was to communicate immediately that English-language assistance was available while retaining a familiar local greeting.

Instead of asking the caller to navigate a language-selection menu, Una could listen to the response and continue in a supported language.

This approach was intended to address two potential sources of friction:

• Language uncertainty: international callers hear an immediate indication that English is supported.
• Menu navigation: guests can begin speaking without selecting options through a keypad.

At the time of the original February 2026 analysis, the team was testing bilingual greetings across the network.

The stated target was to reduce the reported empty call rate below 12%, including callers who had previously been excluded from the AI reporting by the IVR stage.

That figure was a performance target, not a verified outcome.

The published analysis did not provide completed A/B test results demonstrating that the new greeting achieved it.

Nevertheless, the experiment illustrates an important principle of AI voice deployment.

A better guest experience may depend as much on how the call begins as on what the AI can accomplish after the conversation starts.

What This Means for AI Voice Agents in Hospitality

AI voice agents are often evaluated through conversation-level metrics.

Operators may examine the number of calls handled, the percentage requiring human escalation, average duration, or whether the AI completed a requested task.

These measurements are valuable, but they describe only part of the guest journey.

An AI voice agent can perform well after a conversation begins while the telephone system still creates unnecessary friction before that point.

That distinction becomes particularly important for hotels using existing PBX infrastructure, legacy IVR menus, or complex call-routing arrangements.

Una by Polydom is designed as a voice-first AI Digital Employee for hospitality.

Through supported integrations and configured workflows, Una can communicate with guests, retrieve relevant reservation information, support booking actions, and escalate situations requiring staff judgment.

But the 15,910-call analysis demonstrates why the effectiveness of an AI deployment cannot be judged solely by what happens inside Una's conversations.

Operators also need to understand how guests reach the AI in the first place.

For an AI receptionist to contribute meaningfully to hotel operations, it must be accessible, appropriately configured for the property's audience, and connected to a reliable process for handling the requests it receives.

Learn how Una works with existing hospitality systems and operational workflows.

Five Questions Hotel Operators Should Ask About Their Phone Data

Before introducing an AI voice agent — or evaluating an existing deployment — hotel operators should examine the reporting boundaries surrounding their telephone system.

1. Are We Measuring Every Incoming Call?

Determine whether reporting begins when a caller reaches the property's phone infrastructure or only after the call is transferred to the AI.

If callers can disconnect during an earlier IVR stage, those interactions should be measured separately.

Otherwise, the hotel may underestimate abandonment across the complete telephone journey.

2. Do We Know Where Callers Disconnect?

A caller hanging up before hearing the greeting represents a different operational situation from someone disconnecting after receiving an answer.

Segment calls by the stage at which they end.

This can help distinguish routing friction, early abandonment, and conversational performance.

3. Are We Comparing Different Property Types?

A spa resort, airport hotel, extended-stay property, and city business hotel may attract different caller populations.

Their language needs and reasons for calling can differ.

A single network-wide performance target may conceal meaningful differences between properties.

4. Does the Greeting Match Our Guests?

Review the languages commonly used by guests and whether the opening greeting makes the available language options clear.

For internationally oriented properties, consider whether a multilingual greeting or direct language detection would provide a more accessible experience.

Test changes rather than assuming that one greeting works best for every hotel.

5. Are We Connecting Phone Metrics to Business Outcomes?

Early abandonment deserves attention, but conversation completion is not the same as reservation conversion.

Where possible, examine how telephone inquiries relate to completed bookings, repeat calls, escalations, and follow-up actions.

That provides a stronger basis for assessing business value than relying on the empty call rate alone.

Frequently Asked Questions

What Is an Empty Call in Hotel AI Voice Reporting?

In the Polydom deployment analyzed here, an empty call was an interaction that ended before the caller had a meaningful conversation with Una.

The precise definition can vary by telephone platform and analytics system, so operators should confirm how their provider classifies such calls.

Why Can Removing IVR Increase the Reported Abandonment Rate?

Removing IVR may change which calls are included in the AI's reporting.

Callers who previously disconnected before reaching the AI can become visible once calls are routed directly to the agent.

This means the reported rate may increase even when the AI itself has not changed.

Does a Higher AI Empty Call Rate Mean the AI Is Performing Worse?

Not necessarily.

The rate may reflect telephone routing, greeting design, caller behavior, or changes in measurement.

Operators should compare equivalent stages of the call journey before attributing a change to the AI voice agent.

Should International Hotels Use Multilingual AI Greetings?

Multilingual greetings may help clarify language availability for international callers.

However, the right approach depends on the property's audience, the languages supported by its AI system, and the results of live testing.

The analysis identified multilingual greeting design as an area for experimentation, not as a universally proven solution.

Can an AI Voice Agent Help Hotels Capture Direct Bookings?

Yes, an AI voice agent can support direct reservation workflows when connected to systems that permit booking actions.

For example, in a separate deployment at Lee Gardens Extended Stay, Una made 232 direct reservations representing $80,968 in reported booking revenue over ten months.

Those results are separate from the European hotel network analyzed in this article.

Read the Lee Gardens hotel AI booking case study.

The Bigger Lesson: Your AI Metrics Are Shaped by What You Measure

The most surprising discovery from 15,910 hotel phone calls wasn't that AI voice agents could handle guest conversations.

It was that a change in telephone routing revealed a group of callers the existing AI reporting hadn't captured.

The old IVR menu had created a measurement blind spot.

Removing it didn't automatically solve call abandonment. But it made more of the caller journey visible.

The difference between spa and city hotels also reinforced the importance of evaluating AI voice performance in the context of each property's guest population.

For hotel operators, the lesson extends beyond telephone systems.

Before drawing conclusions about AI performance, understand how the data was collected, which interactions are included, and what the reported metric actually measures.

A low abandonment rate may look reassuring.

A higher rate may reveal a problem.

But sometimes the more uncomfortable number is also the more useful one.

Because you can't improve the guest conversations your reporting doesn't show you.

See How Una Works in Real Hotel Operations

Considering AI voice agents for your hotel or hotel group?

Talk to our human sales team about call handling, multilingual communication, reservation workflows, and how Una works within existing hospitality systems.

Book a quick call here to explore how Una could support your property's guest communication and operational needs.