Financial institutions are expanding their use of AI across credit assessment, customer service and everyday banking. DBS deployed agentic AI to about 1,500 relationship managers and credit risk managers, while Absa reported adoption of its chatbot and Starling added task-specific workflows to its banking assistant. Outside banks, AI infrastructure advanced across forecasting, model routing and payments. Ant International expanded its FX and cash-flow forecasting model among global banks, Stripe agreed to acquire OpenRouter and AWS made AgentCore payments generally available with Coinbase and Stripe Privy wallets. Read more on the week’s key developments: 1. DBS moves agentic AI into corporate credit assessment DBS announced on 19 August that it had rolled out an agentic AI credit-assessment capability to about 1,500 relationship managers and credit risk managers globally, following a pilot with 150 users. Specialised agents complete more than 70 tasks and combine annual reports, industry research and internal records into a review-ready first draft of a credit memo for large and mid-sized corporate clients. Credit-memo preparation can absorb up to 40% of a relationship manager’s time, with DBS targeting a reduction of at least 30%. Bankers and credit risk managers remain responsible for refining the draft using client knowledge, industry judgement and business context. DBS does not describe the agents as making the credit decision. The rollout follows agentic AI upgrades to DBS’s customer assistants in Singapore, Hong Kong and Taiwan and extends the technology from customer service into an internal credit-assessment workflow. 2. Ant expands AI FX forecasting tool among global banks Ant International announced on 20 August that Barclays, Citi, Deutsche Bank and Standard Chartered had adopted FalconTST 2.0 for cash-flow and foreign-exchange forecasting. Ant said the upgraded time-series model achieved a Mean Absolute Scaled Error score of 0.666, placing it at the top of the public benchmark leaderboard cited by the company. The banks are applying the model through different treasury and FX services. Barclays has integrated it into BARX NetFX, Citi has combined it with its Fixed FX Rates offering and Standard Chartered uses it alongside SCALE. These relationships pre-date the upgraded model, including a Citi pilot for airline FX exposures and Standard Chartered’s earlier work forecasting Ant International’s exposures. Ant reports accuracy above 93% across the deployments but does not provide results by bank or disclose how accuracy was measured consistently across the different use cases. 3. Stripe brings OpenRouter’s model routing into its infrastructure stack Stripe agreed on 19 August to acquire OpenRouter, an AI gateway that routes workloads according to model capability, price, speed and reliability. OpenRouter provides access to more than 400 models from over 80 providers and says it processes more than 10 trillion tokens daily. The companies did not disclose the transaction value, which was reportedly more than $7 billion. Stripe already provides OpenRouter with payments, invoicing, tax, fraud and usage-billing infrastructure. In January, the companies said OpenRouter served more than five million developers, compared with the more than 10 million developers and companies it reports today. Subject to completion, the acquisition would add model selection and routing to Stripe’s infrastructure for tracking usage, applying prices and billing customers. This extends Stripe’s position from monetising AI consumption into managing how that consumption is distributed among model providers. 4. Flagstar selects Finxact to consolidate three legacy banking environments Flagstar Bank selected Fiserv’s Finxact platform on 17 August as its next-generation system of record and transaction-processing engine. The US regional bank plans to replace legacy technology with a single consolidated core through phased conversions. Finxact provides cloud-native, API-first and real-time transaction processing and will form part of Flagstar’s S2 technology programme. The core project addresses complexity accumulated through the combination of Flagstar Bank, New York Community Bank and parts of Signature Bank. Flagstar said in June that S2 would consolidate three legacy banking environments, six data centres and disparate technology stacks. Finxact establishes the target core architecture, but Flagstar has not disclosed a timetable for completing the core conversions or data-centre consolidation. 5. Amazon equips Bedrock AI agents with Coinbase and Stripe payment infrastructure Amazon Web Services made AgentCore payments generally available on 18 August, adding transaction capabilities to AI agents built on Amazon Bedrock AgentCore. Agents can discover and pay for APIs, Model Context Protocol servers and digital content using Coinbase or Stripe Privy wallets. The service supports stablecoin microtransactions, the x402 and Machine Payments Protocol standards and infrastructure-level controls over spending, with records of agent transactions. AWS introduced the capability in preview in May with Coinbase and Stripe. Coinbase provides wallet provisioning and access to x402-enabled services, while Stripe’s Privy supplies a separate wallet connection. General availability embeds wallet authentication, payment orchestration and spending controls within the infrastructure used to run AI agents. The initial use cases centre on agents paying for machine-consumed services, including APIs, model inference and digital content. 6. Absa connects chatbot adoption with wider AI implementation Africa’s Absa Group disclosed on 19 August that its agentic AI chatbot handles more than 100,000 queries a month from about 1.6 million users. The service interprets natural-language requests, asks clarifying questions and provides personalised responses. Its Business Banking version supports South Africa’s 11 official languages, with Absa planning to extend the multilingual capability elsewhere. The chatbot sits within a 7% increase in IT-related spending to ZAR8.8 billion ($520 million). Absa has also consolidated customer information from 32 systems into an AI-enabled master-data platform. Around 30,000 employees use Microsoft Copilot monthly and more than 3,800 employee-built agents are in production. Absa does not say whether the chatbot’s 1.6 million users are registered, cumulative or active users, limiting direct comparison with its monthly query volume. 7. Ping An reports AI deployment across service, sales and claims Ping An reported on 20 August that AI agents supported CNY57.3 billion ($8 billion) in sales during the first half of 2026. AI service representatives handled about 939 million interactions, equal to 81% of total customer-service volume, while average daily token use increased from 30 billion in December 2025 to more than 120 billion in June. The group serves 253 million retail customers, while monthly online active customers peaked at about 90 million during the 12 months ended June. Its AI-enabled Express Service supports inquiry, consultation and processing across 88% of Ping An’s business scenarios. Separately, 59% of life-insurance claims were settled through its quick-claim service, while Ping An P&C reported CNY7.11 billion ($1 billion) in claims savings from smart fraud detection. Ping An says AI agents “helped realise” the CNY57.3 billion in sales, indicating technology-assisted sales without attributing incremental revenue to AI. 8. Monzo activates its independently hosted backup bank during an outage Monzo activated Monzo Stand-in on 19 August after identifying technical problems affecting its main banking services. The bank said customers could continue making card payments, withdrawing cash, freezing cards and sending or receiving bank transfers while it restored the primary platform. Monzo did not publish transaction-success or disclose the technical cause of the incident. Stand-in runs on Google Cloud Platform, separately from Monzo’s primary platform on Amazon Web Services. It uses 18 dedicated services, compared with around 3,000 on the main platform, and costs roughly 1% as much to keep running. Monzo first enabled all Stand-in components for all customers during a one-hour outage in August 2024, making the latest activation another live use of an established resilience system. The smaller platform uses eventually consistent data and limited app functionality to support critical services, while the primary platform remains Monzo’s system of record. 9. Alipay makes merchant services accessible to external AI agents Alipay launched its full-stack agentic-commerce platform on 17 August, allowing merchants to convert existing pages, products and service workflows into Skills and Model Context Protocol tools that AI agents can invoke. Merchants with their own AI capabilities can use the platform to build and coordinate agents, while Alipay supplies payment, identity-verification, risk-management and order-fulfilment infrastructure. The platform builds on Alipay’s expansion of agentic commerce during 2026. AI Pay processed more than 120 million transactions in one week in February, while AI Wallet and Token Pay added wallet controls and payments for AI services in May. Its Ah Bao consumer agent connects users with more than 10,000 services and has integrations with five smartphone brands covering over 70% of China’s smartphone market and 16 automakers. The merchant platform links these consumer entry points with agent-readable business processes across discovery, payment and fulfilment. 10. Starling turns its banking assistant into task-specific tools Starling Bank launched Smart tools on 20 August, adding preconfigured actions to Starling Assistant. The initial tools help business customers identify transactions subject to tax and sweep a percentage into a savings Space, assess Making Tax Digital requirements and help retail customers review spending or plan a student budget. Starling plans to release new tools weekly for the rest of 2026. Starling Assistant can already set up transfers, create Spaces and analyse spending through natural-language or voice prompts. Smart tools package those capabilities into defined workflows with pre-programmed prompts. The planned weekly releases turn the assistant into an expanding catalogue of bank-configured actions, while customer-built tools remain a future capability. Starling is seeking ideas from its five million retail and SME customers. Financial Technology Weekly covers key technology and AI developments transforming banking and financial services globally. Subscribe via LinkedIn.
