Bringing custom analytics into daily risk decisions at Numerix’s Taipei user group

A risk calculation becomes useful when teams can bring it into the decisions they make each day. In his session at Numerix’s 2026 Taipei user group, Wes Laih, Pre-Sales Engineer at Numerix, examined the practical work behind that transition, from connecting existing systems and giving users purpose-built interfaces, to matching computing resources to the calculation. His examples covered an insurer’s derivatives overlay, a bank’s portfolio risk calculations, and a commodities trading workflow before turning to AI-assisted trade novation analysis. 

NxCore Workspaces provided the development environment connecting these examples. Laih described how teams can use it to build applications for requirements that standardized systems leave unmet. The session’s emphasis stayed close to implementation, including where existing infrastructure remains in place and where financial judgment still belongs with people. 

Connecting analytics to the insurer’s workflow

For an Asia-based insurer using a derivatives overlay to help manage assets and liabilities, the difficulty lay partly in bringing information together. Positions sat in a core booking system, Bloomberg supplied pricing for callable instruments and options, actuarial teams provided liabilities analysis, and Excel combined the results. That fragmented process constrained scenario and stress analysis, which the insurer had historically performed monthly. 

The approach Laih described retained the core system as the book of record. Numerix Oneview introduced complex pricing and scenario analysis, while NxCore Workspaces supplied custom integration and interfaces, including support for asset-liability hedging reports. Users could examine interest-rate scenarios and assess how the duration overlay compared with the underlying asset-liability positions. 

The division of responsibilities matters for institutions with established systems. The booking platform continues to record positions, and the custom application brings analytics into the insurer’s own working process. Moving spreadsheet logic into a more systematic workflow also reduces the risk of users overwriting formulas or leaving them out.  

NxCore Workspaces supports several ways to put applications like this one into use. Teams can run calculations directly, expose functionality through an application programming interface (API), or deploy a web application as a standalone tool or embedded within Oneview. Numerix hosts the environment on Amazon Web Services (AWS) and handles part of its operation and maintenance. Quant teams gain a route from development to deployment, with less dependence on an internal IT queue for each custom requirement. 

Matching compute to the trading workload 

A European bank example addressed a different constraint. The bank needed significant resources to calculate risk across a sizable structured notes portfolio, with a workload that covered thousands of trades, including equity baskets with two to six underlying assets, and calculations using tens of thousands of Monte Carlo paths. With volumes expected to roughly double in one to two years, the bank was approaching the limits of its on-premises computing ability. Intraday delta and gamma calculations supported risk assessment alongside hedging needs, while overnight runs addressed broader portfolio analysis and scenarios. 

Laih described using serverless execution for intraday requests and grid-based scaling for larger end-of-day (EOD) workloads. Keeping the existing API schema reduced the integration changes required of callers, as they could change the endpoint URL without redesigning the interface. Continuous integration and delivery helped the team manage development and deployment with limited IT support. 

Preliminary proof-of-concept results gave an indication of the potential improvement for that workload. Laih reported that EOD portfolio Greeks calculations fell from two hours to half an hour at comparable cost, a fourfold speed improvement. Those figures describe the bank example presented in the session and are not a benchmark for every portfolio or deployment. 

Scale is not the only reason calculations slow down. A commodities customer working with energy trading illustrated how calculation speed and application design can affect the same desk. Only two or three quants supported its traders. Excel calculations took 50 minutes, and traders could damage the workbook when changing parameters. 

The quants developed a front end through Workspaces and used parallel computation in the back end. According to Laih, the calculation time fell to 10 minutes. The interface also gave traders a way to adjust inputs without editing the underlying workbook. For a small quant team, that separation helps keep routine parameter changes from becoming spreadsheet repair work. 

Using AI to assess novation choices 

The session’s AI discussion centered on a public-sector financial institution use case involving valuation adjustments (XVA) and potential future exposure (PFE). Novation transfers a trade from one counterparty to another. A proposed transaction can look attractive on price while its XVA charges or exposure-limit implications make the overall economics less appealing. 

Laih walked through an architecture developed with the customer to investigate those choices. The agentic workflow collected relevant data, inferred counterparty funding spreads from quoted XVA, and examined exposures. It then identified and ranked possible novations, including combinations of trades. An orchestration layer coordinated the work, while a recommendation agent helped assess exposure shapes and possible combinations across hundreds of trades. 

The allocation of work was explicit. Numerix libraries calculated the funding-spread inversion, exposures, and scenario results. Agents coordinated tasks and helped evaluate candidates. AI was also used to develop the calculation engine and the tools deployed in Workspaces, leaving only the core numerical calculations to Numerix libraries. Reusing precomputed exposures could reduce the amount of computation required when exploring alternatives. 

Laih also described how Model Context Protocol (MCP) servers and API wrappers can let agents access Numerix products and libraries or invoke custom tools. In the novation workflow, a trader or quant could ask about moving specified trades from counterparty A to counterparty B, then use the calculated results to assess the consequences. 

Human experts retained the decision to execute, and a novation proceeded only if every party involved stood to benefit enough to justify the legal process. 

Where analytics meet the workflow 

Across the examples explored during our Taipei user group, the practical gains depended on how the calculation reached its user. The insurer needed connected analysis that gathered inputs from multiple systems and teams. The trading teams needed computation sized to their workloads and interfaces that put the analytics in front of the people using them. The novation workflow added agent-assisted exploration while keeping numerical calculations in Numerix libraries and execution decisions with experts. Laih’s session gave concrete implementation choices to consider as teams bring custom analytics into daily risk management. 

To learn more about Numerix’s user groups and find an event in your region, contact us.