Shanghai AI Innovation Study Tour and Retreat 2026 - Day 3

Shanghai AI Innovation Study Tour and Retreat 2026 - Day 3

Day 3 of the Shanghai AI Innovation Study Tour and Retreat moved the discussion from the visible surface of artificial intelligence into the harder operational questions underneath it. 

The day’s visits and briefings brought the delegation into three very different but connected worlds: Transwarp’s enterprise data and AI infrastructure, RedNote’s content-led platform economy and Bank of Shanghai’s attempt to rebuild traditional banking around silver finance, small business lending and a new core system. Taken together, the sessions showed that in China, AI is not being treated as a standalone technology theme. It is increasingly being built into the operating logic of institutions, whether in data architecture, customer acquisition or frontline banking services.

At Transwarp, the discussion stayed firmly on the plumbing beneath AI rather than the front-end applications that attract most of the attention. Founded in 2013, the company started in big data and has since expanded into AI infrastructure, with its products divided between data and AI platforms. Its positioning in the domestic market is that of an infrastructure software company serving large institutions, especially banks, securities firms, funds and insurers. The company said it has the largest market share in China’s big data industry and highlighted the breadth of its financial services client base, including many of the country’s leading banks. The speakers pointed in particular to work with Shanghai Pudong Development Bank, where they said Transwarp had supported more than 10 open-source large language models and over 20 agents for internal banking use.

What was striking in the Transwarp session was not just the scale of its ambition, but the way it framed the evolution of AI inside banks. The discussion set out a sequence that starts with basic question-and-answer functions, moves into retrieval-based systems using internal knowledge and then progresses towards deeply embedded agents connected to business processes. In that view, the problem is not simply whether a bank can deploy a model. It is whether it has the data foundation, governance and internal operating structure to make those tools useful at scale. This was why the company repeatedly came back to data cleaning, data governance and platform architecture. In its telling, AI adoption in banks still starts with getting data into usable form.

The implementation path it described was cautious and staged. Banks typically begin outside the core business, using models for customer service, legal advice or internal assistants. The next step is to combine external models with industry knowledge and internal information so that the outputs become more professional and more specific to banking. Only then does the process move into highly customised agents linked to the bank’s own business. The company said some banks it works with are already running more than 50 agents. The most common types described in the discussion included internal knowledge retrieval tools, customer-facing assistants, risk analysis tools and application-generation agents. There was also a notable claim that in some cases one AI agent could replace work previously done by engineers in building small internal applications.

The conversation also surfaced a deeper issue relevant to every institution on the tour: China’s AI journey has emerged from a highly data-centred and increasingly localised technology environment. One of the reflections after the session was that many Chinese institutions consolidated and disciplined their data infrastructure years ago and are now moving from a “data centre bank” towards an “AI-centred bank”. That is not the same path taken in many overseas markets, where cloud-led and more distributed architectures often came first. The point was not that one model is automatically better, but that the institutional starting point matters. For banks outside China, the real question is not whether to copy the Chinese route, but which parts of it are useful in accelerating their own.

If Transwarp represented the infrastructure layer, RedNote represented the consumer-facing layer where data, behaviour and commerce are brought together in real time. The session there focused on why the platform sees itself not simply as a social media company, but as a community where people search for answers and discover products, services and experiences through other people’s real-life posts. The distinction matters. According to the presentation, RedNote had reached 350 million monthly active users by December 2024, with about 80% still in China. Around 50% of users are under 30. The platform said 70% of users conduct monthly searches and 45% discover new products there. It also noted that more than 170 million users come to the platform specifically looking for purchase or product-related information.

The company argued that its biggest difference from platforms such as Instagram lies in the way content is distributed. On RedNote, what users ultimately see is shaped less by who they follow than by what they have actually read, searched, browsed and engaged with. That makes it less of a pure broadcasting platform and more of an interest-driven discovery engine. The speakers described the platform as a place where people go to solve practical questions in daily life, whether about parenting, medicines, travel, financial products or opening a bank account in Hong Kong. This becomes especially important in categories where consumers need more explanation and reassurance before making a decision.

That was why the RedNote team spent so much time discussing “seeding”, which it described as planting ideas rather than pushing direct hard-sell conversion. For financial institutions in particular, the argument was that users rarely start with loyalty to a brand. They start with a need, a question or a life decision. The task is therefore to shape the content environment around those needs, work with relevant creators and make sure the brand’s own content is consistent with what users have already seen from others. Search behaviour, user interest clusters and community discussion all then become inputs into campaign design. The company’s examples around Hong Kong were especially telling, with strong year-on-year growth in search interest around opening bank accounts, insurance and cross-border financial arrangements. The implication was that users on the platform are not merely browsing. They are already moving along a decision journey.

At the same time, the RedNote briefing made clear that platform growth has not come at the expense of patience. Unlike short-video platforms built around speed and impulse, RedNote said its community tolerates and even values longer-form content, longer captions and slower product explanation. For sectors such as finance, that is significant. It means the platform may be particularly suited to products that require context, trust and comparison rather than instant purchase. Monetisation, in turn, comes through advertising, search, campaign services and platform-native brand collaborations, but the company stressed that it delayed meaningful monetisation for years in order not to damage user experience too early.
The final stop at Bank of Shanghai brought the day back into the world of regulated finance and showed how a traditional institution is trying to respond to demographic change, digital expectations and competitive pressure without losing its local identity. The bank repeatedly positioned itself as Shanghai’s own bank, with over 360 branches across China, of which about 120 are outside Shanghai, but with the city’s development still its first priority. One of the most compelling parts of the session was the silver finance presentation. What began in 1998 as the operational responsibility of distributing pensions in Shanghai has developed into a much broader service model for the elderly.

The bank said it now serves six million pension customers and distributes more than RMB100 billion (about $14 billion) in pension payments each year. It also said elderly customer assets under management exceed RMB500 billion (about $69 billion).

What made the presentation notable was that the bank did not frame silver finance narrowly as a deposit or product opportunity. It framed it as an ecosystem. Examples included ATMs that accept both passbooks and cards because elderly customers still trust passbooks, low-volatility investment products designed specifically for seniors, telemedicine services available through branches, remote service through tablets and mobile interfaces, and home visits by staff where in-person due diligence is required. The aim is to break down the boundary between financial and non-financial services and make banking part of ageing well. The bank’s view is that this is not just a social mission. It is also a durable business logic in a rapidly ageing city.

Its inclusive finance strategy for small and micro businesses showed a similar attempt to move beyond standard product thinking. The bank said its number of SME customers had risen by more than 80% over the past six years, and that the share of revenue contributed by this segment had increased from around 5% to more than 13%. The examples included automobile supply chains, online merchants and platform-linked lending through internet ecosystems.

That led directly into one of the most important themes of the session: the role of data and technology in making the bank more agile. Bank of Shanghai described a restructuring of the way technology and business teams work together, moving away from slow sequential cutovers towards more concurrent coordination between business staff, developers and testers. In some retail finance cases this has reduced the launch cycle for standard products to one or two weeks, and for more complex features to two or three weeks.

The same logic sat behind the bank’s new core banking transformation. The bank explained that its old core could still have supported operations for several more years, but that waiting would have meant losing future opportunities. The new system, built in-house rather than on foreign core banking software, was driven by three things: security and stability, efficiency and scalability, and the need to support future business growth.

The migration took about 27 months and involved a team of more than 100 people at peak, including suppliers. Instead of running old and new systems in parallel for months, the bank chose a one-time cutover over a long holiday, with about seven hours of downtime mainly for data migration. The bank said the new core can now handle up to 240 million transactions per day, has improved transaction processing latency by 37%, reduced batch processing time by 77%, and achieved 99.999% transaction success availability.

Across the day, the strongest message was not that China has found one perfect model for AI, platforms or banking transformation. It was that institutions here are moving on several fronts at once. Enterprise AI players are trying to turn data infrastructure into agent-ready systems. Consumer platforms are turning search, behaviour and community content into commercial seeding engines. Traditional banks are rebuilding around demographic realities, platform partnerships and more agile internal delivery. For the delegation, the real value of the day was in seeing how those layers now interact. AI is no longer sitting outside the institution as an experiment. It is being pulled into the operating model, the distribution model and increasingly the strategic model itself.

Comments (0)
Cancel