Is Human-Reviewed AI Outdated? | The Future of AI Oversight in Customer Support

Reasons Chatbots Fail in the Travel Industry

― Why “Solutions Meant to Improve Efficiency” End Up Increasing the Workload―

In recent years, the travel industry has increasingly expressed interest in:

Delegating customer inquiries to AI
Using chatbots to address staffing shortages

However, in practice, many organizations find that although the system is implemented, it is not fully utilized, and front-line staff ultimately have to cover the gaps

This article examines the reasons why chatbot implementations are prone to failure, with consideration given to the specific characteristics of the travel industry

Reason for Failure (1)

Customer Inquiries Vary More Than Expected

Inquiries in the travel industry have the following characteristics:

There are many combinations of dates, group sizes, cities, and flight numbers.
Relative expressions such as “last time,” “last year,” and “the other day” are common.
Tour numbers and product names are often written ambiguously.

What happens if we take the same tour as last year, but with one more child this time?

Inquiries like this
cannot be framed as a standalone FAQ question.

With an FAQ-based chatbot:

The question may not match any FAQ.
It may return a similar answer instead.

These problems are likely to occur.

Reason for Failure (2)

Too Many Exceptions

In the travel industry, exceptions such as the following occur frequently:

Weather, natural disasters, and strikes
Sudden changes due to local circumstances
Cancellations that require individual judgment

These situations
are governed by rules, but the same rules do not necessarily apply every time
which means:

As a result:

The more precisely the rules are written, the more they may diverge from reality.
Vague rules can lead to misunderstandings.

This creates a dilemma.

Reason for Failure (3)

Thinking “Automation Means AI Does Everything”

One of the most common misconceptions about introducing chatbots is the idea that
**“If we leave it to AI, people will no longer be needed.”**

In reality:

Triage incoming inquiries
Provide initial information
Answer obvious, standard questions

These are areas suited to AI.
Responses that require judgment and accountability must be handled by people.

If AI is introduced without drawing this line:

Staff may spend more time checking AI responses.
Responsibility in the event of a complaint may become unclear.

These are common outcomes.

Reason for Failure (4)

Knowledge “Grows on Its Own”

More systems now automatically read websites and PDFs
to build a knowledge base.

However, in the travel industry:

Outdated terms and itineraries
Multiple documents containing the same information
Limited-time notices
Options and wording that are no longer in use

These often coexist, so
deciding what the AI should learn is itself important.

Automating this process can create risks such as:

Not knowing when the information was current
The AI providing answers that staff are unaware of

These risks can arise.

How Should We Think About This?

When training AI for the travel industry,
**“what it can do” and “what it should not do”**
must be clearly distinguished.

Areas Handled by AI

Drafting replies
Executing reply commands
Organizing information

Areas Handled by People

Selecting the knowledge the AI should remember
Codifying decision criteria
Handling exceptions
Final review

One Approach

For this reason, TuneAIBot adopts the following design principles:

Separate automatic replies from human review.
Knowledge is managed by people.
Design with the variability unique to the industry in mind.

This also reflects the decision
not to let AI run unchecked.

Conclusion

AI is a powerful tool when used properly.

On the other hand,
implementation that ignores the characteristics of the industry
may end up increasing the workload for staff.

The important questions are:

How much should be delegated?
At what point should a person step in?

These boundaries
should be set from the start.

For Those Who Want to See It in Action

For those who find it difficult to judge from text alone,
we have made a demo AI using travel industry data available.

Questions answered automatically
Questions reviewed by a person
Questions the AI does not “judge”

You can see the difference in actual conversations.

無料デモを見る

※ No sales calls
※ No registration required