The competitive question for banks is no longer confined to developing AI models. Technology platforms already control important customer touchpoints, distribution channels and operating data. Their ability to combine these assets with AI could reshape how financial products are marketed, credit is assessed and customers receive advice.

During visits on the second day of The Asian Banker Beijing AI Innovation Study Tour 2026, delegates examined three distinct approaches. ByteDance demonstrated the connection between algorithmic distribution and AI-assisted content production. Xiaomi showed how a connected-device ecosystem supports digital services and supplier financing. Ant Group presented applications in identity security, financial advice and risk-management workflows. The discussions also exposed limits in attribution, data access and the measurement of commercial results.

ByteDance combines AI-generated content with targeted distribution

At ByteDance, the discussion traced the development of its recommendation systems and their application to content distribution, advertising and AI-generated video.

Its early news aggregation platform, Toutiao, used algorithms to recommend content according to inferred user interests, rather than relying on users to search for it. Smartphone application pre-installation agreements also helped the company acquire users outside conventional advertising channels.

The model brought content recommendations and advertising into a common distribution system. Its relevance to banking lies in the potential to improve customer acquisition, although reaching an audience does not necessarily translate into financial-product sales.

ByteDance demonstrated AI-assisted video production through a 68-episode short drama, with episodes lasting about two minutes. Presenters estimated production costs of about $1,000 for a basic version and up to $40,000 for a more elaborate production, against about RMB 300,000 (about $42,000) for a comparable live-action drama. The presentation did not establish whether the estimates covered equivalent production requirements, limiting conclusions about the savings attributable to AI.

The workflow covered scripts, character and scene generation, visual references, character replacement and director-level controls. Lower production costs may widen access to video content, although copyright compliance, platform review and audience demand still determine whether it is commercially viable.

For banks, ByteDance’s Volcano Engine technology platform outlined privacy-controlled ways to match customer information with external behavioural insights without transferring raw datasets between participants.

The presentation covered audience matching and advertising arrangements linked to agreed return-on-investment targets. A presenter said about 70% of Chinese banking institutions cooperated with Volcano Engine in some form; the scope and basis of that figure were not independently established during the session.

Delegates questioned data ownership, responsibility for advertising content and the durability of performance guarantees when creative quality varies. The banking test is whether better targeting produces incremental, compliant sales, rather than simply more impressions or engagement.

Xiaomi uses its device ecosystem to support financing and distribution

At Xiaomi, delegates examined how smartphones, household devices, wearables and electric vehicles contribute to a business combining hardware sales, software and internet services.

Xiaomi reported more than 700 million monthly active users across its MIUI ecosystem and more than one billion connected devices on its AIoT platform. Its move into electric vehicles, beginning with the SU7 in 2024, has extended the ecosystem beyond phones and homes.

Xiaomi described a model combining competitively priced hardware with recurring digital-service income. It said it limits the overall net profit margin of its hardware business to no more than 5%, while earning additional revenue from application distribution, pre-installation agreements, advertising and other services.

The presentation cited quarterly internet-services revenue of RMB 10 billion (about $1.4 billion) and gross profit of RMB 8 billion (about $1.1 billion). The figures indicate the contribution of the installed device base, although they do not establish the profitability of individual products or financing activities.

The most directly relevant discussion for banks concerned Tianxing Digital Technology, whose activities include virtual banking, consumer finance, insurance brokerage and supply-chain financing.

Xiaomi described financing suppliers within its commercial ecosystem and businesses in which it has invested, drawing on purchase orders, transaction records and procurement information to assess financing needs.

According to the presentation, financing personnel work alongside automotive procurement teams to verify supplier orders and delivery arrangements. This gives the financing business access to operational information close to the underlying transactions rather than relying only on periodic financial statements.

For banks, the approach illustrates the potential of embedded supply-chain finance: lenders with timely, reliable transaction information may identify changes in supplier activity sooner. That does not remove credit or concentration risk, particularly when borrowers depend heavily on one commercial ecosystem.

Delegates also discussed bank partnerships in application pre-installation, payments, instalment financing and customer engagement. Xiaomi representatives said bank application pre-installation partnerships had declined as marketing budgets tightened.

Xiaomi emphasised regional data storage, privacy requirements and customers’ ability to delete personal information. Its device offerings vary across markets in response to certification and privacy rules.

The banking opportunity therefore concerns access to customers and transaction information within permitted arrangements, not unrestricted use of data generated across the ecosystem.

Ant Group applies AI to identity verification and financial workflows

The final visit was to Ant Group’s Beijing T Space, where delegates viewed demonstrations covering Alipay+, OceanBase, AntChain, privacy-preserving computing and AI-enabled services.

Ant Group said Alipay+ connected 56 digital wallets globally, supporting cross-border payments and merchant access to overseas customers. A healthcare assistant using a proprietary medical large language model was also demonstrated, although the subsequent banking discussion centred on identity security and financial AI.

Representatives of Ant Digital Technologies outlined how identity verification and AI-assisted processes are being extended across financial services.

The company described a progression from facial recognition and electronic know-your-customer (eKYC) checks to know-your-business (KYB) and proposed know-your-agent (KYA) capabilities. KYA addresses the emerging challenge of identifying and governing AI agents interacting with financial systems.

Its identity-security approach combines liveness detection, hardware-related controls and network-graph analysis to identify manipulated documents, deepfakes and fraudulent identities.

The presentation showed how historical identity records can expose patterns such as one face linked to multiple documents or one identity linked to different faces. Ant Digital Technologies said its technology helped GCash in the Philippines reduce account-registration time from five days to three minutes. The session did not establish how much of that improvement was attributable specifically to its technology.

Ant Digital Technologies also presented AI agents for retail customers, bank employees and small businesses. Its wealth-management example addressed the difficulty of extending personalised service when relationship managers can handle only a limited number of customers.

In a case study presented by the company, an AI-assisted operating model increased customer coverage per relationship manager from about 200 to 3,000, a fifteenfold rise. It said about 70% of conversations concerned financial enquiries, product comparisons and market interpretation, while 30% related directly to investment products. These are reported operating measures, not independently verified improvements in advice quality or customer outcomes.

The company identified obstacles including graphics-processing-unit costs, weak data quality, model hallucinations, fast-changing technology, limited explainability, shortages of specialist talent and difficulties translating business requirements into technical specifications.

Its proposed model combines technology deployment across computing, data, models and applications with human oversight. Agents support information preparation, standardised tasks and broader customer coverage, while relationship managers retain responsibility for judgement, complex cases and client relationships.

Delegates questioned whether Ant Digital Technologies competes with banks. The company described itself as a technology supplier rather than a provider of banking operations or financial-product marketing.

In risk management and anti-money laundering, the company described agents that process alerts, incorporate external information, classify risks and prepare summaries for human review. It said its KYA product was planned for release in the first quarter of 2027, rather than already deployed in production.

Commercial value depends on data access, distribution and accountability

The visits presented three different routes to applying AI commercially, although their relevance to banks varies.

ByteDance combines content generation with algorithm-driven customer distribution; Xiaomi uses connected devices and procurement information to support services and financing; and Ant Group is integrating AI into identity checks and financial workflows. These approaches draw their commercial potential not from models alone, but from access to customers, operating data and established commercial relationships.

For banks, each approach also has limits. Better advertising targeting does not establish incremental revenue. Supply-chain visibility may improve underwriting without eliminating counterparty risk. AI-assisted relationship management can increase customer coverage without proving that advice, suitability or investment outcomes have improved.

The second day therefore offered practical examples of AI innovation alongside a more demanding test of its value: whether these systems produce measurable commercial and operational improvements while preserving customer-data controls, sound credit decisions and accountability for AI-assisted actions.