(Yicai) Sept. 12 -- User retention is the single most important factor investors look at when deciding whether an artificial intelligence startup is worth backing, according to the managing partner of venture capital firm GSR Ventures.

“In the AI era, many people are keen to try out new things. When a new product or service comes out, people will pay for it the first month, but by the second month they may stop paying. There is no retention,” Zhu Xiaohu said during a panel held at the opening forum of the 2025 Inclusion·The Bund Conference yesterday.

Once a user leaves, it can cost more than 10 times to bring him or her back, something that is almost impossible to do, he said. Therefore, user loyalty is the main factor when it comes to deciding whether a project is worthy of investment or not.

“Good user retention will prove whether a company has the potential for future growth,” he added.

Startups in the AI era are not very different from those in the personal computer or mobile internet eras, Zhu said. In order to survive long term, companies still need to stick to the basics of running a good business.

As for which types of AI applications have real growth potential, it is often the ‘boring’ but practical technologies that do best, rather than the flashy ones, Zhu said.

For instance, the most successful AI commercialization cases worldwide last year were tools for taking meeting minutes, such as Abridge and Plaud, he added. The technology behind them is not particularly complicated, but they solve practical needs and can be easily scaled up.

Consumer-facing AI applications are likely to boom next year, Zhu said. Chinese entrepreneurs have a strong advantage here as they are very good at creating standout user experiences. The recent surge in the valuations of gaming firms is proof that delivering excellent user experiences pays off, he added.

“Last year, the AI industry saw an explosion in hardware led by US chip giant Nvidia. This year the focus is infrastructure, with companies such as the US’ Oracle and Snowflake, and next year will be the year of applications. There is likely to be an explosion in applications next year,” he said.

“The next ByteDance, the next Kuaishou or the next Xiaohongshu is probably already formed,” he added.

Editors: Tang Shihua, Kim Taylor