Between 2020 and 2025, Techcombank expanded its customer base from about 8 million to 18 million while keeping headcount broadly stable. Jens Lottner, chief executive officer of Techcombank, said employee growth averaged only 0.4% per year over the period, while operating income grew "probably two to three times." Over the period, Lottner said, Techcombank built a highly scalable platform that enabled rapid growth without a corresponding increase in headcount or physical infrastructure. The bank built that platform around a simple operating principle. Techcombank separated the intelligence that decides what should happen from the systems that execute transactions. Rather than making every application independently intelligent, Techcombank kept decision-making in a central layer and used operational systems as channels for execution. "Technology will not make the difference, but intelligent technology and the intelligent application of technology will really be the sustainable competitive advantage," Lottner said. In his view, AI only becomes a structural advantage once a bank has resolved the underlying questions of scale, data and decision-making. Building the platform before scaling AI Techcombank's transformation rests on three connected pillars: technology, data and talent. The technology pillar began with a cloud-first strategy which affords the bank more flexibility to scale systems and deploy new capabilities. Lottner said about 60% of the bank's compute now sits on Amazon Web Services (AWS), which provides the scalable foundation for its AI and data workload, a level he described as relatively high for a full-sized bank. The bank is also migrating its core banking system to the cloud, with technical completion expected by yearend and further functional upgrades to follow. The second pillar is data. Lottner said Techcombank decided it needed to "own the data", rather than treat data as a by-product of technology systems. The bank separated the chief data officer role from the chief information officer role and built its data infrastructure on AWS. It now stores around eight billion customer data points a day, profiles each customer using about 12,000 features and runs around 55 AI models to support decision-making. The most important design choice is where the intelligence sits. Lottner said Techcombank deliberately kept operational and vendor systems focused on execution rather than decision-making. Customer offers, channel choices, risk signals and next-best actions can be determined centrally, then sent into the relevant platforms for execution. The third pillar is talent, especially people who can bridge business and technology. Lottner said Techcombank had significantly strengthened its technology capabilities with the share of employees working in technology rising from around 5% in 2021 to more than 15% today, but the more difficult capability sits at "the intersection of business and technology." This includes "technologists who can talk business" and business people who understand the implications of technology. Techcombank has invested in product owner training, change management and HR support to close that gap. Techcombank's collaboration with AWS also supports its platform strategy. Lottner said AWS helps Techcombank connect with a broader ecosystem of technology partners and specialists enabling the bank to build its own capabilities and innovate faster than it could independently. "We work with somebody like AWS and their ability to pull talent and partners of their ecosystem into Vietnam, that's hugely valuable," he said. At the same time, Techcombank still builds internally where local context matters, such as Vietnamese-language processing and document interpretation. For Lottner, the value of partnership lies in helping the bank move faster, because time is the one resource that "cannot be replenished." Creating capabilities that did not exist before "If you just try to optimise existing processes, you will get some benefits, but you will never be able to rethink the bank or the way you actually do banking," Lottner said. The bank's credit offering for micro-businesses illustrates this shift in approach. Vietnam has about 6.5 million household businesses, many of which lack consistent financial records. Lottner said the traditional banking response to that uncertainty has often been simply not to lend. Techcombank's response was to create a new data set, rather than rely only on traditional financial records. Field representatives capture images of business premises, such as noodle shops, from different angles. Software is then used to analyse the images to assess activity levels, footfall and other criteria and benchmark them against comparable businesses in the area. Those signals feed into Techcombank's decision engine and help the bank assess risk in a segment that conventional credit models struggled to serve. The bank now serves about 2.5 million customers in this micro-business segment, with around one million pre-approved for credit. Lottner said many of these customers would likely not have qualified for credit two years earlier under conventional assessment methods. Governing AI before exposing it fully to customers The same structure that lets AI scale also increases the consequences of mistakes. When Lottner found more than 1,000 AI agents running across a 15,000-person organisation, it confirmed appetite for AI but raised two questions: whether the agents were generating value and whether the bank had sufficient control over what they were doing. Cost creates another constraint. Lottner said the bank quickly realised that AI economics matter because some use cases worked well but became too expensive to scale. Techcombank's cloud infrastructure, powered by AWS, gives it detailed visibility into model usage, token consumption and cloud capacity, which helps it identify where costs rise faster than value. The bank now evaluates AI deployment across cost, accuracy and response time. Accuracy also matters. Techcombank distinguishes between internal use cases, relationship manager support and direct customer interaction. Lottner said AI already supports software development, management reporting, market summaries, operations and relationship manager productivity, but the bank remains cautious about exposing customers directly to AI outputs. Techcombank is building a single platform to control how AI agents are created and used — tracking which agents are running, what data they consume and whether teams can reuse an existing agent rather than build another. Lottner said the platform should be in place around the third quarter, with broader guardrail capabilities to follow by year-end. The bank is also strengthening its knowledge layer, meaning the internal information AI needs in order to work properly. Internal memos, policies, legal documents, regulations and operating rules must be structured so agents can read, classify, analyse and apply them correctly. Regulation reinforces that discipline. Lottner said regulators are increasingly focused on AI models that affect customers in banking, where such models may influence risk, credit, investment and advice and will require validation, monitoring and explainability before being allowed to operate freely. "If you make a mistake, this mistake can replicate in no time," he said. Making AI useful for relationship managers and customers Techcombank's caution does not mean it is holding AI back from the business. Lottner said the bank is already embedding AI into relationship manager workflows, where the technology can improve productivity while human judgement remains accountable for the customer interaction. Relationship managers receive suggestions on which customers to contact, why the conversation matters and what message may be relevant, but they remain responsible for checking and sending it. Lottner described Techcombank's commercial ambition through its 2-2-4 strategy: double the customer base, double product penetration and generate four times total operating income. The bank continues to acquire about 2.5 million to 3 million customers a year, but Lottner said penetration is the harder challenge. He defines a main transaction banking relationship as more than 100 transactions per month and about 25% to 30% of customers currently meet that threshold. The target is 40% to 45%. Hyper-personalisation sits at the centre of that effort. The goal is not to push more messages through digital channels, but to present offers and interactions that are useful to the customer. Conversion rates have already increased by up to 200% to 300% in some areas, Lottner said, with click-through rates in certain cases rising from about 1% to 3%. The longer-term opportunity lies in continuous learning. Lottner said this depends on "constant data gathering" and feedback loops that allow the bank to adjust offers and messages based on how customers respond. The bank already runs AI models on next-best-offer and next-best-action, but he said it is "nowhere close" to where it should be. The reason, he said, is that there is a natural ceiling to what manual optimisation can achieve at scale. If models can learn from feedback and adjust on their own, "that I think is where we can really see the biggest difference." The next test is measurable execution Lottner identified four priority areas for broader AI adoption over the next 12 months: hyper-personalisation, relationship manager productivity, operations and software development. He aims to see productivity gains in applied use cases — not incremental improvements of 10% to 15%, but gains that can reach 50% to 60% where the technology works. Techcombank's ability to move more customers from occasional usage to main transaction banking will show whether the bank is deepening relevance, not just acquiring customers. Current and savings accounts, loyalty systems, transaction accounts and personalised engagement all connect to that objective. AI will not create the same advantage for every institution. Where data remains fragmented and decision logic is trapped inside individual systems, it will mostly produce scattered efficiency gains. By keeping intelligence central and enabling decisions to be governed at scale, Lottner argues, AI becomes part of how the bank competes.
