This case study examines how AI tools respond to regulatory questions relevant to Product & Business Development teams at Retail Banking firms operating in the United Kingdom. The analysis covers one regulation — Consumer Duty (PS22/9 + PRIN 2A), issued by the Financial Conduct Authority — and focuses on one aggregated question where AI assistants consistently produced incorrect or misleading answers. Across the AI tools assessed, the same question produced materially wrong outputs in both instances tested, indicating a pattern rather than an isolated failure.
The errors identified are not minor nuances: they misrepresent the FCA's stated expectations in ways that could directly affect how a firm structures its fair value assessment processes. Product & Business Development teams that rely on AI answers in this rule area without independent verification face meaningful regulatory and operational risk.
Product & Business Development teams at Retail Banking firms routinely turn to AI tools to accelerate regulatory mapping when scoping new products or repricing existing ones. A team designing a new savings account, revising a mortgage fee structure, or building out a packaged current account will typically ask AI assistants to summarise what the Consumer Duty requires of them — particularly around fair value. That question sits squarely in the finding documented here.
The same dynamic arises when a department is preparing regulatory training materials, drafting internal policy documents for sign-off by Compliance, or producing a regulatory rationale to accompany a product approval submission.
The corporate use-cases that depend on getting this answer right are substantial. Fair value frameworks, product approval documentation, customer communications, and commercial pricing rationales all flow from the team's understanding of what the FCA actually expects. If that understanding is sourced from an AI tool and the AI's answer is wrong, the error propagates downstream into every document, decision, and sign-off that relies on it.
A new product launched with a fair value assessment built on a misreading of FCA expectations may not satisfy Consumer Duty obligations — and the gap may not surface until a supervisory review, a thematic examination, or a complaint triggers scrutiny.
The firm bears the cost of these errors, not the individual employee who used the AI tool. If the FCA identifies that a firm's fair value process does not meet the standard set out in PS22/9, it can take supervisory action, require remediation, issue public censure, or impose financial penalties. The reputational consequences of a Consumer Duty failure in a product or pricing context extend beyond the regulatory sanction itself — public censure affects customer trust, damages relationships with distribution partners, and can trigger class-action-style redress obligations.
The Product & Business Development team, its leadership, and the firm's executive committee all sit in that line of exposure.
The finding in this case study reveals a consistent directional error: AI tools systematically upgrade the obligations the FCA places on firms, presenting the regulator's explicit statements as more demanding than the published text supports. Where the FCA has chosen not to require quantification — and has said so plainly — AI assistants reframe that position as a conditional permission or an implied expectation to go further. This type of error is particularly hazardous because it does not read as a hallucination in the conventional sense.
The AI responses sound authoritative, are framed in regulatory language, and arrive with citations attached. A reader without independent access to the source text has no obvious reason to distrust them.
All errors in this review cluster on a single regulation — Consumer Duty (PS22/9 + PRIN 2A) — and on a topic area that is directly load-bearing for Product & Business Development workflows: what a firm must do to demonstrate fair value. This is not an obscure corner of the rules. Fair value assessment sits at the centre of Consumer Duty compliance for retail product teams, and the FCA has been explicit about its expectations in that area.
The fact that two AI tools independently produced contradictory or inflated accounts of those expectations, both attaching citations that appear relevant but do not support the claims made, signals a structural weakness in how AI tools handle this regulation rather than an occasional retrieval failure.
The systemic risk to a Retail Banking firm is amplified by how Product & Business Development teams typically work. A single AI-generated regulatory summary may be shared across a product team, used as the basis for a training slide deck, folded into a product approval template, and referenced in a regulatory mapping document — all before anyone has checked it against the source. If the underlying AI answer is wrong in the way documented here, each of those downstream work-products carries the same error.
The cost of correcting them — if the error is caught at all — rises steeply with each layer of use. If it is not caught, the cost shifts to regulatory exposure.
1 finding in this case study. Click any to see its full evidence card.
The default position for any Product & Business Development team in a Retail Banking firm should be that AI tools are a starting point for regulatory research, not a primary source. This is particularly important for Consumer Duty obligations, where the FCA has published detailed policy statements and finalised guidance that are publicly available and authoritative. When an AI tool summarises what those documents require, the summary should be treated as a prompt for further verification, not as a reliable account of the rule.
Any AI output that describes what a regulator expects, permits, or prohibits in a specific area should be checked against the underlying source before it informs a firm work-product.
At the firm level, practical safeguards should be built into how the team uses AI tools. A regulatory verification policy should explicitly identify AI-generated content as requiring independent source-checking for any output that touches FCA rules or Consumer Duty obligations. Where AI output has influenced a document — a product approval note, a fair value assessment framework, a training slide deck — there should be an audit trail that records what the AI said, what source was checked against it, and who conducted the verification.
Sign-off requirements before AI-drafted regulatory content enters firm-wide use reduce the risk that an unchecked error propagates across multiple work-products. It is also worth distinguishing in templates and document metadata between content that was AI-drafted and content that was independently verified — that distinction matters if the firm is ever asked to demonstrate its Consumer Duty governance process.
AI tools do have a safe role in a Product & Business Development workflow. They are well suited to drafting non-regulatory marketing or product copy for human review, generating first-draft questions for a legal or compliance briefing, summarising lengthy documents where the team will subsequently check the key passages, and organising research the team has already gathered. The risk zone is narrow but important: wherever an AI answer would be used to determine what a rule says or what a regulator expects, verification against the source is not optional.
RegLeg publishes hallucination research as a free resource that Product & Business Development teams can use before acting on any AI answer in areas covered by this research. For Consumer Duty — and for the specific question of what the FCA expects from a fair value assessment — the published findings give teams a reference point: if the AI answer your team received resembles the patterns documented here, that is a signal to go back to the source before using the output.
The research is updated as new findings are aggregated, so the coverage expands over time to reflect the regulatory questions teams are actually putting to AI tools.
For firms that want to go further, RegLeg offers bespoke regulator deep-dives that map which AI-supported workflows within a Retail Banking firm carry the highest hallucination exposure. A Product & Business Development team that regularly uses AI tools for regulatory scoping, product approval documentation, or Consumer Duty governance work can benefit from understanding precisely where in that workflow the failure risk is concentrated — and how to build targeted verification steps that are proportionate to the risk, rather than defaulting to blanket prohibitions on AI use that reduce efficiency without improving accuracy.
RegLeg also offers a confidential review of a firm's existing AI-use policy against the failure-mode catalogue built from this research, with prioritised recommendations for where policy adjustments would deliver the most protection. For teams that need to build internal capability, RegLeg can provide training material and CPD-aligned content that explains the hallucination patterns documented here in terms that are practical for Product & Business Development professionals — not just for Compliance or Legal. The goal is to help teams use AI tools more effectively and with appropriate safeguards, rather than to discourage their use.