S10.08 - Finance

S10.08 ? Finance

flowchart LR
    A[Finance context:
credit explanation
adverse-action notices
fraud triage] --> B[RAIDT
run-level evidence framework]
    A2[Traditional limitation:
policy claims without reconstructable run evidence] --> B
    B --> C[[Finance
sector playbook for explanation,
traceability, and contestability]]
    C --> D[Run-level evidence pack]
    C --> E[RAIDT score profile]
    C --> F[Reviewer reconstruction]
    D --> G[Audit readiness]
    E --> G
    F --> G
    H[Practical examples:
lending workflows
reason-code mapping
reviewer approval
compliance support] --> C

? Star S10 - Empirical Programme, Domains and Sector Playbooks

Star context: Explains how RAIDT is tested, calibrated and applied across domains and sector-specific playbooks. In finance, the framework shows how high-stakes GenAI use can be made reviewable through run-level evidence, explanation quality, and contestable decision support.

Academic picture

Definition / background

In RAIDT, finance refers to the domain-specific application of the framework to financial services and financial decision-support settings, especially where GenAI outputs may shape explanations, recommendations, workflows, customer communication, or documentation around consequential decisions. Typical examples include credit explanations, counterfactual reasoning, adverse-action notices, fraud case support, affordability commentary, policy interpretation, and internal analyst assistance.

The concept matters because finance is a domain where explanation, traceability, and contestability are not optional extras. Financial decisions often affect access, affordability, risk exposure, and customer treatment. As a result, organisations need more than a general claim that an AI system is responsible or explainable. They need evidence showing what was done in a specific run, under what configuration, with what criteria, and how the output can be checked or challenged.

This makes finance a strong RAIDT domain because the framework is centred on the run as the unit of governance. Rather than treating governance as a static property of a model or a policy, RAIDT asks whether one concrete use of a system can be reconstructed, reviewed, and defended. In finance, that run-level orientation is particularly valuable because similar tasks may produce very different governance risks depending on the data used, the instructions given, the review path, and the consequences of the output.

Finance therefore belongs inside RAIDT as both a domain of application and a test case for rigorous evidence production. It links directly to run-level evidence packs, because these can record the grounds on which an explanation or recommendation was generated, and to the five-pillar score profile, because financial uses often expose tensions across Responsibility, Auditability, Interpretability, Dependability, and Traceability.

Why this concept matters

Finance matters in RAIDT because it provides a clear example of why high-level governance principles are insufficient on their own. A financial institution may say that its AI use is fair, explainable, or accountable, but unless it can show evidence from a specific run, those claims remain difficult to test.

The concept also avoids a common confusion between model governance and use governance. A model may have documentation, validation reports, and internal approvals, yet a problematic run can still occur if the prompt, context, user instructions, or review process are weak. RAIDT helps separate these layers by showing what happened in one concrete use and whether that use produced evidence robust enough for explanation, challenge, and review.

If this concept is missing, organisations risk treating finance AI deployment as compliant merely because a policy exists or a tool has been approved. That can produce brittle explanations, weak adverse-action records, inconsistent reviewer practice, and poor organisational learning from disputed cases. RAIDT addresses this by making finance operational through run-level evidence rather than principle statements alone.

Key idea: Finance matters in RAIDT because high-stakes financial uses of GenAI require reconstructable, contestable, run-level evidence rather than generic assurances about good governance.

What this item captures

The financial domain as a sector playbook in which RAIDT is applied to high-stakes organisational work.
The need for explanations that can be linked back to recorded criteria, evidence, and review actions.
The importance of contestability when customers, reviewers, or regulators may challenge an AI-supported outcome.
The difference between broad AI governance claims and evidence about one specific financial run.
The practical connection between finance use cases and RAIDT outputs such as evidence packs and score profiles.
The way financial contexts stress-test all five RAIDT pillars, especially Interpretability, Auditability, and Traceability.

Practical example / likely audience question

Audience question

How does RAIDT work in finance if the underlying issue is not only the model, but also the explanation and review process around a decision?

Answer

The concern behind this question is that many governance approaches focus on the model in the abstract, while financial accountability often depends on how one specific decision-support event was produced, explained, and reviewed. In finance, the practical problem is rarely solved by saying only that a model passed validation. What matters is whether a concrete output can be linked to the criteria used, the evidence considered, and the reviewer actions taken.

RAIDT answers this by treating the run as the unit of governance. If a GenAI system drafts an explanation for a declined credit application, RAIDT asks for evidence of the configured task, the prompt or instruction set, the source criteria, the output produced, any reason-code mapping, reviewer edits, approval steps, and the final recorded explanation. That makes it possible to assess not only whether an explanation exists, but whether it is grounded, reviewable, and contestable.

A practical example is an adverse-action support workflow in which staff use GenAI to draft a customer-facing explanation. A generic AI governance approach may stop at policy compliance, user training, and model documentation. RAIDT goes further by preserving the evidence needed to reconstruct the actual run, compare it against policy or recorded criteria, and judge whether the explanation supports defensible organisational practice.

Practical example in RAIDT terms

Consider a retail lending team using a GenAI assistant to draft an explanation for a declined loan application. The use case is not fully automated decision-making; instead, the system helps a case officer prepare a customer-facing explanation and an internal case note.

The run-level issue is that the generated explanation may sound coherent while drifting away from the actual grounds recorded in the case file. It may overstate certainty, omit a relevant affordability factor, or produce language that cannot be traced cleanly to the internal decision criteria.

The evidence needed includes the run timestamp, user role, configured task description, prompt or wrapper instruction, model/version, the criteria available to the system, relevant case data snapshot, output text, reviewer amendments, reason-code mapping, escalation notes, and final approval record. These materials belong in the run-level evidence pack.

The most affected RAIDT pillars are Interpretability, because the explanation must be intelligible and grounded; Traceability, because the output must be linked back to recorded criteria and review actions; Auditability, because a reviewer should be able to reconstruct the run; Responsibility, because a named human or function must own approval; and Dependability, because repeated runs should not produce unstable or misleading explanations under similar conditions.

This improves governance readiness because the organisation can show not only that it has a governance policy, but that it can evidence how one financial explanation was generated, checked, and justified in practice.

Detailed link to RAIDT

Finance links to RAIDT in four ways.

First, it links to RAIDT's core idea that governance should be based on evidence from real organisational use rather than on abstract assurance alone.
Second, it links directly to the run, because financial accountability often depends on whether one concrete instance of AI-supported work can be reconstructed and challenged.
Third, it links to the evidence pack and score profile, because financial uses generate inspectable materials that can be reviewed against the five pillars.
Fourth, it links to reviewability, contestability, audit readiness, and organisational learning, since disputed or high-stakes cases require defensible records and feedback into better practice.

Finance ? Run-level evidence ? Evidence pack ? RAIDT score profile ? Governance readiness

In this sense, finance is not just an example domain. It is a demanding test environment for whether RAIDT can operationalise responsible GenAI governance where explanations matter and records must stand up to scrutiny.

Link to the five RAIDT pillars

Responsibility

Finance strongly affects Responsibility because financial uses often require explicit ownership for approval, escalation, and customer-facing communication. RAIDT makes clear who initiated the run, who reviewed it, and who accepted the final output into the workflow.