For banks, the question is not how many artificial intelligence (AI) applications they can deploy, but whether those applications improve decisions, controls and operating performance. Technology platforms draw on continuous streams of behavioural information; financial institutions must integrate AI with regulated processes, fragmented data and established responsibilities.

Amap applies spatial intelligence to real-world decisions

On the final day of The Asian Banker Beijing AI Innovation Study Tour 2026, the morning discussion at Amap examined how a navigation platform is evolving into a system that can interpret physical activity and support decisions based on location, movement and user intent.

During the visit, Amap cited 910 million monthly active users. Alibaba had reported in January 2026 that the platform’s monthly audience was approaching one billion, illustrating the scale of its location-based services. Amap’s spatial intelligence strategy combines mapping information, positioning signals, road conditions and points of interest with AI models capable of interpreting requests and coordinating services.

The company demonstrated Xiao Gao, an AI agent designed to move beyond answering navigation questions. It can consider a user's destination, preferences and travel conditions to recommend routes and related services. Amap also presented its Eagle Eye Guardian road-safety system, which uses changes in vehicle movement and other signals to identify potential hazards and issue warnings to approaching drivers.

The difference from conventional navigation is the use of incoming data to anticipate events rather than simply report known road conditions. Amap explained that a sudden slowdown involving several vehicles can indicate a possible accident before it is manually reported.

Amap said its system could generate warnings within seconds. The presentation did not establish how often warnings proved accurate, whether they reduced accidents or how its performance compared with conventional road-safety alerts.

Amap’s commercial model also extends beyond consumer navigation. The company supplies mapping services to vehicle manufacturers, developers and other businesses. It said its open platform supports more than 400,000 applications and nearly three million developers, while its overseas mapping coverage includes more than 200 countries and regions.

Amap’s location data raises questions for lending and fraud detection

The most relevant discussion for banks concerned whether Amap’s information about merchants and physical activity could support financial decisions.

Delegates asked whether information about customer visits, business locations and merchant activity could serve as alternative data when small businesses apply for loans.

Merchant visits and activity could help lenders assess whether a business is operating at its stated location or attracting customers. Neither observation establishes sales, profitability or repayment capacity, and the visit did not demonstrate a lending model using those signals.

Amap described data-related cooperation within the Alibaba ecosystem, including links with Alipay. The discussion did not establish that Amap currently supplies banks with merchant footfall data for credit underwriting.

Delegates also explored whether location information could strengthen fraud detection, particularly when a customer's reported position differs from the location associated with a payment or account activity.

Location mismatches may be more immediately useful as supplementary transaction-risk signals, but they do not establish fraud. Legitimate travel, device settings, signal accuracy and consent all affect whether such data can be used responsibly.

The discussion illustrated how AI's value can depend on access to information generated outside the banking relationship. It also exposed the difficulty of converting large volumes of behavioural data into reliable financial indicators.

Bank of Beijing shifts from AI assistants to redesigning entire workflows

The afternoon visit to Bank of Beijing provided a closer examination of how an established financial institution is deploying AI within its operations.

The bank presented its internally developed technology infrastructure, including a computing platform, enterprise knowledge resources and tools for creating AI agents. It reported deploying more than 140 agents across business activities and completing 276 AI-related business projects.

Its broader strategy covers customer service, risk management, software development and operations. The bank also demonstrated digital retail banking systems linking mobile services, relationship managers and remote banking channels.

The more consequential change was in how the bank approached automation.

Bank of Beijing said earlier efforts largely gave employees AI tools to complete existing tasks more quickly. In 2026, it began emphasising the redesign of entire workflows, assigning defined activities to AI while employees handled exceptions, supervision and decisions requiring judgement.

This distinction addresses a common limitation in enterprise AI adoption. An employee may save time drafting a report or retrieving information, but the organisation may realise little benefit if the surrounding approvals, data transfers and processing steps remain unchanged.

The bank illustrated the approach through three examples involving retail customer management, transaction reviews and credit card marketing.

Bank of Beijing reports lower manual effort in transaction reviews

One example involved reviewing transactions against internal rules and compliance requirements.

Bank of Beijing explained that manual post-transaction checks previously relied on sampling because employees could not review every case. Its redesigned process uses AI to conduct broader automated analysis and identify cases requiring attention.

Bank of Beijing reported a 60% reduction in manual effort for this review activity. The figure referred to work required for a specific process, not a reduction in the bank’s workforce. The baseline, measurement period and resulting cost savings were not established during the presentation.

The change matters because the benefit is not limited to processing speed. Moving from sampling towards wider transaction coverage could also change the effectiveness of internal controls, provided automated checks detect relevant problems accurately.

The presentation did not establish whether the new system improved detection rates or reduced false positives. Those measures would be necessary to assess whether the efficiency gain was accompanied by stronger risk management.

In retail banking, the bank described how AI could help relationship managers serve large customer portfolios by managing initial customer screening, engagement preparation and other repetitive activities.

For credit card marketing, it presented a workflow redesign initiated by four employees from different functions. Rather than automating their individual tasks separately, they examined how the entire campaign process could be reorganised.

The example showed why cooperation between business teams matters. Improvements made within a single department may be lost when work passes to another team. Redesigning the process across departments can address delays and duplicated activities that individual AI assistants cannot resolve.

Shared knowledge and employee participation shape wider AI deployment

Bank of Beijing said about 70% of its workforce of more than 30,000 employees had access to or were using its AI capabilities. The presentation did not distinguish employees with access from regular active users, limiting what can be concluded about adoption.

It identified three requirements for scaling deployment: enterprise-wide knowledge engineering, teams combining business and technology expertise and the redesign of existing workflows.

The knowledge component is particularly important. An AI agent cannot reliably complete a banking process if relevant product rules, operating procedures and customer information are fragmented across systems.

The bank's proposed direction is to provide employees with a common AI interface capable of selecting the necessary tools and information, rather than requiring them to navigate numerous separate applications.

It also described the use of locally deployed models for sensitive banking information, distinguishing internal processing from applications that might use external services.

Delegates questioned how the bank manages organisational change, particularly when management support and technology resources are limited. They also asked how AI-generated recommendations and customer-facing outputs are checked for compliance.

The discussion highlighted a tension between making AI widely accessible and retaining appropriate control over its actions. Bank of Beijing's approach combines shared infrastructure with business-led experimentation, but broader employee participation increases the importance of consistent permissions, validation and monitoring.

The bank reported a sharp rise in AI usage after integrating agents more deeply into operational processes. It said token consumption had increased substantially over the preceding six months. Such growth indicates wider use, but does not by itself demonstrate financial returns.

Banks must judge AI by changes in operating performance

Amap and Bank of Beijing approached AI from different starting points. Their experiences nevertheless showed why access to useful information, reliable controls and integration with operational decisions matter more than the number of models deployed.

Amap’s advantage lies in its continuously updated understanding of physical activity. Its navigation and merchant services generate information that can support predictions and actions based on what is happening in the real world.

Bank of Beijing's challenge is to make institutional knowledge, business rules and customer information accessible to AI without weakening governance. Its most consequential examples involved changing how work is performed rather than simply improving individual employee productivity.

For banks, the potential connection between these approaches is the use of external behavioural signals alongside internal financial information. Yet the visit did not demonstrate a deployed credit-underwriting application using Amap’s merchant traffic data. That remains a potential use case rather than an established outcome.

Bank of Beijing provided a more direct operational measure through its reported reduction in manual transaction-review effort. Even there, the presentation did not establish whether wider coverage improved detection, reduced false positives or produced lower overall costs.

Across the five-day tour, the gap between technical capability and business value emerged repeatedly. Sophisticated demonstrations showed what AI could do, while operational examples raised harder questions about deployment costs, accuracy, accountability and measurable benefits.

For banks, the next stage will depend less on adding assistants than on improving complete business processes. That requires clear ownership of data and decisions, measurable performance before and after deployment, and the ability to intervene when automated systems fail.