Rapid Fireside Chat with Justin Xu

Table of Contents:

• What's the biggest misconception about AI in FX today?
• How can AI help corporates better anticipate and manage FX exposure before it becomes a problem?
• Will AI ever fully replace human decision-making in FX?
• How will AI change the way businesses think about currency risk over the next five years?

 

What's the biggest misconception about AI in FX today?

One of the misconceptions about AI is to think that it is a plug-and-play solution that will immediately solve all of the problems in FX markets.  

The success here really depends on matching the right problem to the right tools within the right ecosystem.

AI offers a powerful set of capabilities, however an FX solution doesn’t become effective just because we “add AI.”

The starting point consists of disciplined problem identification: What are we actually trying to solve? Is it a risk prediction challenge (e.g. identifying liquidity gaps), or an analytical/explanatory task (e.g., summarizing event risk)?

Only after the problem is clearly defined and properly evaluated should we select the appropriate AI tech stack: LMs, agentic workflows, RAG, knowledge graphs, etc. to solve it.

Equally important, technology choices must be evaluated within a broader operating ecosystem:

• Risk & controls: Model risk governance, monitoring, guardrails, and human-in-the-loop design.
• Regulatory alignment: Documentation, explainability, auditability, and data handling standards.
• Data foundations: Quality, lineage, latency, and coverage across structured and unstructured sources.
• Edge cases & resilience: Stress behavior, distribution shifts, and fail-safe modes.

When we “marry” problems to tools inside this ecosystem, AI becomes fit-for-purpose: reliable, explainable, and operationally sound. When we skip these steps, we get model theatre - impressive demos that don’t survive real FX markets. Therefore don’t treat “AI in FX” as a single product. Treat it as a design discipline: start with the problem, choose the technique that fits, and engineer for the full FX lifecycle: risk, regulation, data, edge cases, and integration. 

 

How can AI help corporates better anticipate and manage FX exposure before it becomes a problem?

AI can transform FX risk management from reactive firefighting into proactive control by connecting three layers: data assembly, insight generation, and personalised delivery, all under robust governance.

First, it consolidates fragmented information across enterprise resource planning and treasury systems, CSV and PDF files, as well as unstructured sources such as contracts, emails, and logistics updates. It does this by using entity recognition, retrieval augmented generation (RAG), knowledge graph to tag currency, tenor, counterparties, and clauses so the currency risk manager can see a live, line-item exposure picture rather than static balances.

Second, it turns the multi-modal data sources into forward-looking analytics, running currency risk scenarios and stress tests from macro, geopolitical, and liquidity shocks, detecting regime shifts and event risk, and recommending hedge ratios and tenors aligned to risk appetite, accounting rules, and liquidity constraints - while explaining the “why” behind each suggestion.

Third, it delivers the right action to the right stakeholder with role-aware dashboards, proactive alerts ("hedge trigger reached" etc), workflow integration to TMS/EMS, and natural-language queries that supports in depth analysis which can be used for downstream applications. 

In practice, AI might parse new client purchase orders, estimate net EUR exposure, flag a volatility uptick and recommend a staged hedge ladder with projected accounting impact. All of this must be anchored in strong guardrails: data lineage and quality controls, model-risk management with human-in-the-loop, auditability and documentation, and security by design, so that analytic insights become timely, explainable actions in currency risk management.

 

Will AI ever fully replace human decision-making in FX?

I don't think AI will fully replace human decision-making in FX market, at least in the foreseeable future. AI is best understood as an effective co-pilot - a partner that augments human judgment - rather than a replacement. It excels at processing vast data, spotting patterns, and generating timely recommendations that help treasurers and CFOs take proactive risk management decisions. 

The FX market is complex, path-dependent, and context-rich. Corporate exposures are idiosyncratic - shaped by contracts, cash-flow timing, pricing power, supply chains, and accounting treatments. Addressing these realities requires bespoke design: marrying the right data and models to the company’s specific policies, constraints, and risk appetite. That tailoring is where human domain expertise remains important and effective.

There is also the governance dimension. Regulatory, legal, and audit requires an explainable and transparent decision making process. AI can support these obligations, by tracing inputs, evidencing assumptions, and logging recommendations, but humans must own the ultimate choices, especially when trade-offs involve reputational and regulatory risk.

Practically, the winning model is human-in-the-loop: AI surfaces scenarios, highlights regime shifts, proposes hedges or tenor ladders, and quantifies impacts; humans validate, adapt, or override based on the company domain specific context and policy. This partnership delivers speed and consistency without compromising judgment and accountability.

 

How will AI change the way businesses think about currency risk over the next five years?

Over the next five years, AI will reshape currency risk management in two major ways.

It will make decision-making deeply context-rich. Instead of relying on generic market views or historical time-series alone, AI will break data silos and fuse firm-specific information: sales and revenue projections, hedge policies, and cash-flow timing, with external context such as event calendars, macroeconomic data, geopolitical signals, and liquidity conditions. This synthesis turns static exposure snapshots into live situational awareness tailored to each company’s unique context.

In the second instance, AI will drive on-demand, bespoke insights tailored to each stakeholder and use case. Instead of one-size-fits-all dashboards, treasurers and CFOs will receive targeted recommendations: hedge ratios, tenor ladders, trigger levels, and policy exceptions, delivered when and where they are needed. Natural-language queries, proactive alerts, and workflow integrations will make those insights instantly accessible and easy to act on, whether the task is approving a hedge, adjusting exposures, or revisiting currency risk management strategies. 

The net effect is a move from periodic, generic risk reviews to continuous, personalised risk intelligence - context-aware analytics that surface the right action at the right moment for the right user.

To learn more about how MillTech’s AI-driven FX solutions can help your business turn data into intelligent currency risk management, https://www.milltech.com/contact.