On May 22, at the 2025 Global Business Travel Association (GBTA) Digital Intelligence Forum in Shanghai, Gu Qing, founder of DTalk.org, delivered a speech titled “Users, Scenarios, and Industries: An Interpretation of the China Tourism AI Application Trend Report.” The speech revealed the current state of AI applications in the tourism industry and the challenges faced by enterprises.
Gu emphasized that AI is not meant to replace humans but to collaborate with them. He stressed the need for companies to address the hidden issues of AI application by starting with organizational structure and data foundations to drive digital transformation in the industry.
Gu has worked in the internet sector for nearly two to three decades and has been closely following and researching AI technology. However, the “2025 H1 China AI Tourism.Application Trend Report” he presented was not written by him. His role was to interpret the report and share its content and trend judgments with the audience.The data for the report came from several partners, including Fliggy, Mafengwo, Tuling Travel, Zhongxin Travel, and GBTA. Due to time constraints, Gu focused only on the enterprise and industry sides of the report.
The survey included approximately 3,000 user-side samples from first- and second-tier cities as well as third- and fourth-tier cities. On the enterprise side, about 70 managers participated, including 900 business travel leaders. The industry side surveyed around 115 subjects.
Looking at the current state of enterprises, the adoption rate of AI tools is not high. For example, nearly 70% of participants have never had the opportunity or are unwilling to use AI tools, with only 31% trying them out, and their functions are limited to booking and expense control. Why is this the case?
The analysis revealed that enterprises have a low level of trust in these tools. Implementing them requires robust process management and significant integration with various systems, which many enterprises are not fully aware of or willing to invest in.
AI tools have a high threshold and require substantial investment, but their effectiveness is uncertain. This makes many enterprises hesitant to adopt them. As a result, the promotion of AI within enterprises has become a major point of contention.
Since last year, Gu has communicated with many enterprises. The promotion of AI projects within enterprises often involves a preliminary step: although the decision-making level of many companies agrees to use AI, they question its return on investment and whether existing AI tools match the actual needs of the enterprise.
Let’s take a look at the situation with travel management and expense reporting.
Within the tourism industry, travel management has become a clear and rigid demand. The arrangement of all routes, checkpoints, and the ticketing process from point A to point B has become very important in terms of information management.
However, the penetration rate of expense reporting is still insufficient.Expense reporting occupies a significant amount of working time for salespeople and frequent travelers. But why hasn’t this area been improved?The reasons lie in system compatibility and the adaptability of internal management processes within companies.
Let’s look at another figure. The popularity of different types of AI applications is related to the city in which they are located.The higher the city tier, such as first-tier and new first-tier cities, the higher the adoption rate of travel management and smart expense reporting. In contrast, cities below the fourth tier have lower adoption rates. This shows that the digitalization level of first- and second-tier cities is relatively higher.
Looking at medium-sized enterprises with 10,000 to 100,000 employees, their adoption rate of intelligent systems is the highest, close to 85%. In contrast, enterprises with more than 100,000 employees have a lower usage rate. Companies with 500 to 2,000 employees are more reliant on such systems.
The distribution chart shows that the penetration rate of AI applications is not even.
From the perspectives of state-owned enterprises, private enterprises, and foreign enterprises, the usage rate of foreign enterprises is relatively lower.
This may be because the Chinese branches of foreign companies have higher requirements for IT compliance and data privacy. Their management requirements and decision-making processes are much more complex than those of domestic private enterprises.
In fact, not only in the field of AI, but also in terms of digital sampling rates or digital maturity in China, foreign companies are relatively slower compared to Chinese companies.
When I was in charge of Ctrip’s online business and competed with overseas companies like Booking in the domestic market, the difference was very clear. The iteration speed and intensity of domestic software in domestic enterprises are much stronger than those of foreign companies.
Overall, the penetration speed of AI within Chinese private enterprises is definitely ten times faster than that of foreign companies.On the industry side, leading companies such as Fliggy, Meituan, and Mafengwo have a very different pace and choice of AI technology implementation and scenarios compared to traditional hotel groups and travel agencies.
Looking at the data, if we divide companies based on whether they have already implemented AI technology in their business, 80% of large companies are using it, while the proportion decreases as we go down the corporate ladder.
Traditional AI applications such as recommendation systems, supply chain optimization, price forecasting, and sentiment monitoring are all based on demands like machine learning and deep learning.The difference between generative AI and traditional AI lies in the emergent aspects, which bring many compliance issues.
Large companies, with more mature digital infrastructure and higher internal data quality, can use structured data to constrain generative AI and avoid hallucinations. However, small companies may not have mature digitalization, and even if they use AI, the emergent results of generative AI remain uncontrollable.Looking at AI applications across different industries and companies, the application structure is also very interesting.
In comparison, the AI adoption rate in the hotel industry is not high, at only 25%.
The biggest issue is the database problem.Enterprises purchase multiple systems, each with a different data structure, operating independently. Each time an AI application is to be implemented, a lot of manual data processing is required, which is costly, not to mention the application of AI technology.
Therefore, the first step for the hotel industry is for the entire enterprise management process to move towards modern enterprise management systems. This is a problem of scientific management of operations, not an AI technology issue.Looking at the organizational structure of many companies, they have not made any adjustments for the introduction of AI.
If you compare with leading companies like Ctrip and Meituan, you will find that the job content of their product managers and marketing managers is very different from that of the same positions in your company, with different processes and outputs. This is an issue of organizational design.
In terms of applications, whether it is Fliggy’s personalized recommendations, Ctrip’s recommendation system, or the personalized recommendations of platforms like Douyin and Ele.me, the reason for the increased click-through rate is that generative AI can help companies use algorithms to supplement a large number of user characteristics.
This shows that the better the data foundation, the greater the possibility of using generative AI to improve business results.
Mafengwo’s good travel planning tool is also because its POI and review content and other structured data prerequisites are well done. Any team that does a good job with travel planning tools has invested a lot of time in the pre-processing of structured language materials.
Moving forward, whether it is large enterprises, medium-sized enterprises, or traditional small businesses, the options available now are either local deployment or renting someone else’s model through cloud computing, that is, model service.
Although it seems there are many choices, there are some issues with local deployment. For example, the computational power cost, graphics card cost, and maintenance cost of open-source models all require investment.
Moreover, the business side rarely participates in the decision-making process of introducing large models, which leads to technology becoming a tool that cannot create value for the business, and the business will not use it. This is a dilemma.The general mentality is to wait and see who can implement it first and then directly imitate it. No one is willing to be the first to try and do it end-to-end.
This is mainly because everyone is concerned about the effectiveness, considering data privacy issues, costs, and internal organizational resistance, especially the fear among employees of being replaced by AI.Let me ask everyone here again, are you worried about being replaced by AI? In fact, I believe that AI is not here to replace users but to collaborate with everyone.
Today’s time is limited, and I will conclude by mentioning data foundations, cost technology, and organizational adaptation.In my view, the first two are explicit reasons and are solvable. As long as there is investment in cost and time, they can be addressed. The hidden issue is the organizational problem. Organizational issues are internal and are the problems that each company needs to solve.
