近日，全球知名财资媒体《Treasury Today》刊登了对PingPong 财资负责人 Neil Xu的专访《The language of business》。报道中，Neil Xu结合自身 20 年财资从业经历，分享了 AI 在跨境支付、反欺诈合规与企业财资管理中的实践与展望，并提出“AI正在成为商业的‘新通用语言’”。

以下为本次专访英文原文节选，附中文对照：

AI 正成为商业的“新通用语言”

Neil Xu’s 20-year career started with KPMG where he learned “accounting: the language of business.” Today, as Head of Treasury at PingPong – a payment service provider (PSP) that processes annualised $300bn+ for 750,000 merchants globally – Xu believes AI is set to become a new lingua franca.

Neil Xu 的 20 年职业生涯始于毕马威（KPMG），在那里他领会到“会计是商业的语言”。如今，作为 PingPong 的司库负责人——这家支付服务提供商（PSP）每年为全球 75 万家商户处理超过 3000 亿美元的年化交易——Xu 认为，人工智能将成为一种新的通用语言。

“Artificial intelligence via Large Language Models (LLMs) and other applications of the technology has advanced. It has made tremendous progress in recent years, particularly in its agentic form,” says Xu.

“借助大语言模型（LLM）及其他技术应用的人工智能已经向前迈进。近年来，它取得了巨大进展，尤其是在智能体形态方面。”Xu 说道。

 “But we are still in the early stages of AI. It has made tremendous progress but it’s still on information management end uses, such as improving data forecasting, visibility, aggregation, presentation, pattern spotting for anti-fraud purposes on payment flows, and so on.”

“但我们仍处在人工智能的早期阶段。它已经取得了巨大进展，但目前仍主要应用于信息管理类场景，例如改善数据预测、可见性、汇总和呈现，以及在支付流中识别反欺诈模式等等。”

AI 反欺诈与合规项目：双智能体架构

Over the last two years, PingPong has developed a new in-house proprietary AI-enabled risk architecture with two agents to solve the problem:

过去两年，PingPong 开发了一套全新的内部专有 AI 风险架构，包含两个智能体，以解决这一问题：

The first AI agent ingests data, such as KYC information, transactions records, invoices and so on to identify threats. The AI model spots common signatures of fraud activity, such as shared beneficial owner patterns, timings and payment tools chosen to refine and improve risk scores that require investigation. This has reduced false positives and improved efficiency, client protection and performance.

第一个 AI 智能体：摄取 KYC 信息、交易记录、发票等数据，以识别威胁。该 AI 模型能够识别欺诈活动的常见特征，例如多个主体共用同一实际受益人的模式、时间规律以及所选支付工具，从而细化并改进需要调查的风险评分。这减少了误报，并提升了效率、客户保护水平和业务表现。

The second AI agent drafts any possible threats into RFIs to bank partners like J.P. Morgan, which assisted in the rollout. The agent also compiles regulatory narratives to aid bank interactions and compliance reporting. A human still decides what to investigate.

第二个 AI 智能体：将任何潜在威胁整理成提交给摩根大通等银行合作伙伴的信息请求（RFI）；摩根大通协助了系统上线。该智能体还会编写监管说明材料，以协助银行沟通和合规报告。最终调查什么仍由人工决定。

“Bank RFI response times have consequently halved to five days, and our IT estate and customers are better protected than ever before. The platform is now being rolled out to Canada, then South-East Asia and latterly more risky emerging markets like Brazil where it’ll have to learn the particular risk factors there.”

“因此，银行 RFI 的响应时间缩短了一半，降至五天；我们的 IT 资产和客户也比以往任何时候都得到更好的保护。该平台目前正推广至加拿大，随后是东南亚，之后还会进入巴西等风险更高的新兴市场；在这些市场，平台需要学习当地特有的风险因素。”

数字化未来：人的"战略性角色"更加关键

 “Since the turn of the decade, AI is definitely having an impact,” continues Xu. “AI is empowering treasury management and will have a still bigger role in future. ”

“自这个十年开始以来，人工智能无疑正在产生影响。”Xu 继续说道，“人工智能正在增强司库管理能力，未来还将发挥更大的作用。”

 “However, as AI plugs into this increasingly digitised and data-centric world to improve the accuracy and analysis of the vast information flows available to it, the strategic role of the human becomes more important.”

然而，随着人工智能接入这个日益数字化、以数据为中心的世界，以提高对海量可用信息流的分析准确性和处理能力，人的战略作用反而变得更加重要。

In closing, Xu reflects that AI can automate and analyse things – but when it comes to yield and risk management, a human being is still better suited for seeing something out of the ordinary that doesn’t compute. “For spotting an unidentified opportunity or a black swan risk lurking in the background, experienced humans are pretty good,” he adds.

最后，Xu 认为，人工智能可以自动化处理和分析事务——但在收益与风险管理方面，人类仍然更擅长发现不合常理、无法用数据解释的异常。“在发现尚未识别的机会，或察觉潜伏在背景中的黑天鹅风险方面，经验丰富的人相当擅长。”他补充道。