Leveraging Frontier Models in Valuation Control Functions

The application of frontier models in valuation control functions is moving quickly from experimentation to practical deployment. While the potential to accelerate analysis and improve consistency is clear, these capabilities must be introduced within robust control frameworks to avoid introducing new risks.

This blog explores how frontier models can support valuation workflows, the guardrails required to govern their use and practical applications in root cause analysis and trade data validation.

AI adoption

Financial institutions and service providers have used automation and analytics in parts of the valuation lifecycle for several years. The emergence of frontier models (including LLMs) has accelerated industry interest in where else AI can be applied.

While the benefits of using AI are obvious, adopting intelligent systems and workflows purely for cost reduction is a narrow objective if it overlooks the operational, model and governance risks that advanced AI can introduce. AI can deliver significant value, but only when embedded into workflows with robust governance, testing and human oversight.

Controls and guardrails

Prudential valuation frameworks, including Model Risk Additional Valuation Adjustments (Article 11 for the European Union) and the Prudential Regulation Authority’s SS1/23 guidance in the United Kingdom, highlight the need to identify and manage uncertainty and risks when AI or machine learning is used in valuation processes.

To mitigate the risks of using frontier models in business processes, organizations should treat them like any other model or system. This includes:

• Establishing clear governance and accountability
• Ring-fencing usage through use case tiering, such as decision support versus automated decisions
• Requiring thorough independent sign-off prior to rollout

Material outcomes should remain subject to human review. Robust process controls should include continuous monitoring and audit trails across inputs, outputs and prompts, supported by disciplined change management.

Adoption should also be reinforced through user training, clear operational guidance and fallback plans so that critical processes can revert to traditional methods if model performance degrades or market or regulatory conditions change.

The SS&C approach

At SS&C, AI does not replace approved valuation models or valuation control review. Instead, it is deployed to support valuation professionals by standardizing triage (i.e., determining the optimal waterfall approach to resolving valuation exceptions), flagging exceptions and improving evidence quality. The intelligence layer is integrated into our process, which includes:

• Automated market data checks to flag anomalies, including stale data points and surface inconsistencies across over the counter (OTC) datasets.
• Automated root cause analysis of valuation variances.
• Human oversight of exceptions prior to resolution, closure or client communication.
• Back testing and validation of model outputs.

Use case examples

1. Valuation reconciliation analysis
   
   Valuation exception analysis for OTC derivatives is not a trivial exercise. Reviewing valuation breaks for even the simplest derivatives requires reconciling multiple drivers, such as (but not limited to):

• • Trade booking
  • Market data
  • Static data setup
  • Model or methodology issues
  • Curve construction nuances
  • Valuation configuration setups
  • Timing of valuations issues, among others

Without expert judgement, determining where to begin is itself complex.

Frontier models can help by synthesizing structured and unstructured evidence and proposing candidate hypotheses for root cause. These hypotheses are then validated through deterministic controls and valuation control review.

What we observed in proof-of-concept:

• • For moderate complexity scenarios, Large Language Models (LLMs) can propose plausible hypotheses using trade dynamics, Greeks and other market variables.
  • For more complex instruments/issues, models can hallucinate and must not be relied on without strict controls.

When implemented within a governed framework, AI-enabled workflows can support faster resolution, improved accuracy of net asset value (NAV), profit and loss (P&L), clearer explanations and improved NAV delivery timelines.

As part of our governance and oversight framework, we would be controlling the scope of instruments use cases to employ AI and build tasks around how we unify and standardize the resolution of the output from such workflows. This will systematize the controls globally, reduce the analyst time in identifying issues and give clients increased accuracy of NAV deliverables while mitigating risk.

2. Trade economics anomaly detection

   Trade booking issues are a significant valuation discrepancy driver. Embedding frontier models in the trade capture process can help reduce valuation variances due to booking inconsistencies. For example, SS&C is embedding the use of statistical outlier detection plus an LLM grounded in market conventions to flag non-conforming attributes (e.g., unusual strikes/tenors, missing schedules, inconsistent settlement flags) as an enhancement to our trade capture process. Any anomalies detected are output as “warnings” with suggested remediation steps designed to be actionable.

Balancing innovation with control

The benefits of leveraging AI in workflows are well documented. For valuations, AI can help standardize and optimize the valuation review process by providing a waterfall mechanism for addressing the root cause of valuation discrepancies in a holistic global manner.

However, these should augment, not replace, pricing models or human judgement. Over‑reliance on AI, especially for highly technical workstreams such as valuation control, can introduce additional risk.

With clearly defined use case boundaries, robust testing and mandatory human oversight, AI can help teams focus time on the most complex valuation issues while maintaining strong control environments.

Download our brochure to learn more about how our internal analytical libraries and pricing applications can support your valuation requirements.