Alibaba's AI strategy shift comes into focus with big bets on agents

* AI agents have potential to drive change in consumer behavior

* Alibaba ecosystem offers integration of e-commerce, logistics, AI capabilities

* Alibaba separates AI from cloud, forms Token Hub Business Group

* Wukong platform launched for enterprise automation, handling complex tasks with AI agents

By Casey Hall

SHANGHAI, March 18 (Reuters) - Alibaba is sharpening its artificial intelligence strategy by focusing on agents that connect the many businesses under its sprawling corporate umbrella.

In recent months, Alibaba has rolled out several AI agent integrations and this week, the firm said it would separate its AI businesses from its cloud computing arm. The newly formed Alibaba Token Hub business group, led by Chief Executive Eddie Wu, is the clearest sign yet that the company is shifting its focus to digital assistants powered by AI models that consume far more tokens - units of data used by models to generate language - than traditional Q&A chatbots.

Alibaba did not respond to a request for comment on this story.

The $325 billion e-commerce giant reports quarterly results on Thursday, with AI monetisation in focus as major tech firms in China and beyond wrestle with how to make the era-defining technology profitable. Analysts expect Alibaba's (BABA) third-quarter revenue to rise 3.8% and net income to fall 42.5%. The quarter included Singles' Day, China's biggest shopping festival.

Facing a prolonged slump in consumer confidence as shoppers save rather than spend, a weak macroeconomic outlook and a prolonged property crisis that has eroded household wealth, Alibaba has turned to new business models to encourage consumption.

Last year, the firm invested heavily in acquiring users for its instant retail platform, which competes in the one-hour delivery market with Meituan (MPNGF). This year, Alibaba's (BABA) AI chatbot Qwen has begun moving beyond answering questions to helping users make purchases directly through a chat interface.

In February, an early push to get users to try Qwen's new functions encountered some hurdles. Alibaba launched the first phase of a 3 billion yuan ($435.7 million) coupon campaign that allowed users to make in-app purchases on Alibaba-owned retail platforms using only chatbot prompts. The coupons proved too popular, prompting a temporary shutdown of the app.

According to Brian Wong, a former Alibaba employee and author of "The Tao of Alibaba," the company's wide-ranging ecosystem - spanning e-commerce, food delivery, travel, movie ticketing and more - means executing all those daily functions through a chatbot could fundamentally shift consumer behaviour.

"Think of it like having OpenAI, Amazon, Stripe, Uber, DoorDash, Ticketmaster, Expedia, Netflix and Charles Schwab all integrated into one text box you can just use natural language to execute," he said. "This is what the company has enabled through its restructuring and it's happening first in China. I don't see this happening in the U.S. because of the challenges of integrating different platforms from different companies."

Alibaba is not the only Chinese tech giant using AI agents to integrate consumer-facing functions, but rivals like Tencent (TCTZF) and TikTok-owner ByteDance would mainly serve as agent platforms interacting with third-party companies inside their apps. Alibaba's (BABA) ecosystem gives it an advantage, said Ed Sander, an analyst at China Digital Retail Report.

"Alibaba also has the fulfillment and logistics part built in, not to mention running everything on Alibaba's (BABA) cloud infrastructure, no other company has the ability to execute every part from the chatbot all the way through to the logistics in the way Alibaba does," he said.

WELCOME WUKONG

On Tuesday, Alibaba launched another enterprise-focused AI platform targeting automation. The platform, called Wukong, can coordinate multiple AI agents to handle complex business tasks like document editing, spreadsheet updates, meeting transcription and research within a single interface.

A key driver behind the shift to agents is not only tapping into the frenzy triggered by the launch of OpenClaw in China, but also the potential to make money from it. These agents, which can make decisions and execute tasks around the clock, consume tens to hundreds more tokens per day than a typical chat session, according to estimates from Poe Zhao, a China tech analyst and founder of Hello China Tech.

This matters especially for Chinese firms, most of which offer open-source AI models that are free to download and have seen token prices plunge amid intense domestic competition among leading tech companies.

Alibaba's (BABA) AI push comes as the company navigates turmoil in its AI leadership ranks. Lin Junyang, head of the firm's Qwen model division, left in early March - the third senior Qwen executive to leave this year.

"This has heightened concerns about morale in Qwen and Alibaba's (BABA) ability to retain AI talent and maintain its leadership in the AI model race," Morningstar analyst Chelsey Tam said. "Top AI talent is scarce. If Lin and core Qwen members join a competitor, it would be a setback for Alibaba."

"The AliCloud bench is deep and broad enough that while Lin's departure was not ideal, there's sufficient talent to fill in the gaps, particularly in light of the new restructuring that just took place," Wong said.

($1 = 6.8851 Chinese yuan renminbi) (Reporting by Casey Hall, additional reporting by Laurie Chen; Editing by Thomas Derpinghaus)