A Practical Guide to Using AI Under Consumer Duty

Quick Answer

Financial advisers can meet Consumer Duty with AI when they keep ownership of every customer outcome. AI can draft, sort, summarise, and flag issues. However, people must set the rules, review material output, and monitor results. The firm remains accountable, even when a supplier provides the technology.

What This Guide Covers

• How Consumer Duty applies to AI use in advice firms
• Why existing FCA rules already cover AI-related risks
• Where AI can affect customer outcomes
• How to set proportionate controls for a small firm
• What evidence to retain and monitor
• How to build a governed AI workflow without creating needless admin
• How LaunchLemonade can support controlled AI use across a team

What Does Consumer Duty Require From AI-Using Advisers?

Consumer Duty requires firms to deliver good outcomes for retail customers. Therefore, an AI-assisted process must meet the same standard as a fully manual process.

The Duty has applied to open products and services since 31 July 2023. It has also applied to closed products and services since 31 July 2024. Its central question is simple: are customers receiving good outcomes in practice?

What Are the Four Consumer Duty Outcomes?

The four outcomes give advisers a useful way to assess every client-facing AI use case. Specifically, consider whether the use affects:

• Products and services
• Price and value
• Consumer understanding
• Consumer support

In addition, firms must act in good faith. They must avoid foreseeable harm. They must also support clients in pursuing their financial objectives.

Why Does the Duty Focus on Outcomes?

The Duty does not prescribe one approved technology stack. Instead, it asks firms to show that their choices produce good customer outcomes.

Consequently, AI adoption does not need to wait for a special permission process. A firm can test useful tools now. Yet it must put evidence and oversight around them from the start.

This is often more practical for smaller advice firms. You do not need a vast compliance programme before trying a low-risk process. However, you do need a clear decision, a safe boundary, and a record that shows why the process remains suitable.

Suggested Visual: A four-part Consumer Duty wheel, with AI controls mapped to understanding, support, foreseeable harm, and accountability.

When Does AI Become a Consumer Duty Issue?

AI becomes a Consumer Duty issue when it can influence a retail customer’s experience or result. This can happen even where the tool never speaks directly to the client.

For example, a meeting summary might shape a suitability letter. Similarly, an email triage tool might decide how quickly a vulnerable client receives help. An internal research assistant might affect the information an adviser uses in a recommendation.

Therefore, the key test is impact, not visibility. If a use case can change what the client receives, when they receive it, or how well they understand it, it needs a Consumer Duty assessment.

Why Is Senior Accountability Still Important?

Senior accountability does not disappear when AI enters a process. Instead, a named person remains responsible for the relevant business activity and outcomes.

That means a supplier’s security statement or compliance claim can inform your assessment. However, it cannot replace it. Your firm still needs to decide whether the tool, workflow, controls, and monitoring are appropriate for the work.

How Can Firms Meet Consumer Duty With AI?

Firms meet Consumer Duty with AI by treating AI as delegated work, not delegated accountability. Therefore, the safest model combines clear limits, human judgement, and ongoing testing.

Start With an AI Use-Case Inventory

First, list every AI use case in your firm. Do not limit the list to client-facing chatbots. Include tools that draft, research, summarise, classify, or automate internal steps.

A simple inventory should capture:

• The use case and business purpose
• The customer journey it may affect
• The information entered into the tool
• The person who owns the process
• The required review point
• The main foreseeable risks
• The monitoring method and review date

  AI Use Case	Potential Client Effect	Risk Level	Minimum Control
  Drafting review letters	Client may receive unclear or incorrect information	Medium	Adviser checks every final draft
  Meeting transcription	File notes may omit or misstate key facts	Medium	Compare summary with recording or notes
  Query triage	A client may face delay or poor routing	Medium	Escalation rules and support monitoring
  Internal research support	Advice may rely on an inaccurate statement	Medium	Validate against approved sources
  Client-facing automated response	Client may act on incorrect guidance	High	Strict scope and rapid human hand-off

Assess the Customer Outcome, Not Just the Model

Next, ask what could go wrong for the customer. A model’s technical score is useful. However, it does not answer the Consumer Duty question on its own.

For instance, an AI-written letter may read well in a test. Yet it can still fail if it confuses a particular customer, omits a limitation, or does not fit their circumstances.

Accordingly, assess the whole workflow:

• What triggers the AI?
• What data does it use?
• What decision or draft does it produce?
• Who checks the result?
• What happens if the tool is wrong?
• How can the client reach a person?
• How will the firm know if outcomes worsen?

Set Boundaries Before Staff Start Using AI

Clear boundaries reduce risk and speed up adoption. Without them, each person builds their own version of an AI process. That creates inconsistent standards and weak evidence.

Your policy does not need to be long. However, it should state:

• Approved tools
• Permitted business uses
• Prohibited uses
• Data-handling rules
• Required human checks
• Escalation routes
• Monitoring and review dates

For example, you may permit AI to create a first draft of a client letter. However, you may ban it from sending the letter automatically. This boundary is simple, clear, and easy to test.

Keep the Human Review Meaningful

A review only works when the reviewer can genuinely challenge the output. Therefore, do not ask staff to rubber-stamp long AI drafts under time pressure.

Instead, give reviewers a short checklist. Ask them to check factual accuracy, client context, tone, disclosures, risks, and plain language. In addition, require them to correct any output before it reaches the client.

The goal is not to retype the work. Rather, the adviser should apply the professional judgement that AI cannot hold.

Suggested Visual: A workflow showing AI draft, adviser review, client delivery, monitoring, and improvement loop.

Where Can AI Affect Consumer Understanding and Support?

AI can help clients understand information more easily. However, it can also create polished confusion when the firm does not test the result.

How Can AI Improve Consumer Understanding?

AI can simplify jargon, summarise dense documents, and flag undefined terms. As a result, it can help advisers make communications easier to follow.

For example, an adviser can ask an AI assistant to identify terms that may confuse a non-specialist reader. The adviser can then revise the language before sending the final letter.

Nevertheless, the client’s actual needs must guide the final communication. Clear language for one reader may still be unclear for another. This matters especially where the client has vulnerability characteristics or needs extra support.

What Can Go Wrong With AI-Written Communications?

AI can make an inaccurate statement sound persuasive. It can also omit key context while producing a neat, confident summary.

Common problems include:

• Incorrect figures or dates
• Missing warnings or limitations
• Generic wording that does not fit the client
• Unexplained technical language
• Overstated certainty
• A tone that feels dismissive or unclear

Therefore, never judge a draft only by its fluency. Check whether it is accurate, complete, fair, and appropriate for that client.

How Should AI Support Vulnerable Customers?

AI should never become a barrier to support. Instead, it should help the firm spot when a person may need more time, a different format, or direct human help.

For example, triage rules can flag messages that mention bereavement, financial distress, confusion, or a complaint. A trained person should then take over promptly.

Importantly, AI can assist with routing. However, it should not make unsupported assumptions about vulnerability. People must remain able to assess the context and respond with care.

Why Is Human Hand-Off Essential?

Clients must be able to reach a person without unreasonable friction. Consequently, a bot or automated response must never become a dead end.

A sensible process gives clients a visible route to human help. It also sets escalation rules for complaints, urgent matters, money movement, distress signals, and complex advice questions.

Why Does Foreseeable Harm Matter for AI?

Foreseeable harm matters because known AI failures are no longer surprising. Therefore, firms should design controls around errors they can reasonably anticipate.

What AI Errors Are Foreseeable?

Language models can produce wrong information. They can also miss nuance, follow unclear instructions poorly, or reflect weaknesses in their input data.

For an advice firm, foreseeable issues include:

• A draft letter that misstates a charge
• A meeting summary that misses a client objective
• A research output that presents an unsupported claim
• A triage tool that delays a vulnerable customer
• An automated response that offers unsuitable reassurance

These risks do not mean firms should avoid AI. Instead, they show where review and monitoring need to focus.

How Should Firms Match Controls to Risk?

Not every AI use needs the same control set. A private first draft of an internal agenda needs less oversight than a client-facing response about a financial decision.

A proportionate approach considers:

• Customer impact
• Decision significance
• Data sensitivity
• Automation level
• Ability to reverse an error
• Availability of human review

The higher the potential customer impact, the stronger the checks should be. This also makes your governance easier to explain to staff, boards, and supervisors.

What Should Never Run Without Review?

Any process that creates, changes, or sends material client information should have a clear human control. The same applies to work involving complaints, vulnerability, money movement, or tailored financial guidance.

That does not mean every internal task needs a committee. Rather, it means the firm should reserve automation for tasks where an error is contained and easy to correct.

Why Is Testing Better Than Assumption?

A firm cannot prove good outcomes by saying its tool is reputable. Instead, it should test real outputs from its own process.

For example, sample ten AI-assisted client letters each quarter. Check for factual accuracy, suitability, clarity, corrections, and repeat issues. Then record what you found and what you changed.

This evidence is stronger than a general promise. It shows that the firm is looking at customer outcomes in its own context.

What Does Consumer Duty AI Governance Look Like?

A Consumer Duty AI governance framework should be short, practical, and repeatable. Consequently, a smaller firm can build useful control without copying a large bank’s structure.

Who Should Own AI Governance?

Name one senior owner for the overall AI approach. Then name a process owner for each material use case.

The senior owner should oversee policy, risk appetite, monitoring, and material issues. Meanwhile, the process owner should maintain prompts, training, approval steps, and evidence.

This split keeps accountability clear. It also prevents AI from becoming everybody’s interest but nobody’s responsibility.

What Should an AI Policy Include?

A useful policy should tell staff what good use looks like. It should also show them when to stop and ask for help.

At a minimum, include:

• Approved tools and access rules
• Permitted and prohibited use cases
• Client data and document rules
• Mandatory review steps
• Escalation paths
• Incident-reporting rules
• Monitoring responsibilities
• Change-control requirements

Keep the document clear. Staff should be able to use it during a normal working day, not only during an annual training session.

How Can a Small Firm Keep Governance Proportionate?

AI compliance for financial advisers does not need to start with expensive software or a large committee. Initially, an AI register, reviewer checklist, sample log, and calendar reminder can create a workable baseline.

However, manual controls can become hard to maintain as use grows. At that stage, centralising permissions, approvals, and records can reduce the chance that evidence becomes scattered across inboxes and personal tools.

When Should You Reassess a Use Case?

Review each use case when something material changes. This could include a new model, a revised prompt, a fresh data source, a new integration, or a greater level of automation.

Similarly, review after an incident, complaint, repeated correction, or customer outcome concern. A quarterly scheduled review is also sensible for client-impacting workflows.

The point is not to freeze innovation. Instead, it is to make sure the controls still match the process you are actually running.

Suggested Visual: A simple AI governance cycle: approve, use, review, monitor, improve, and reapprove.

What Evidence Should Firms Keep to Meet Consumer Duty With AI?

To meet Consumer Duty with AI, record how the firm knows outcomes remain good. Therefore, evidence should cover decisions, reviews, results, and improvements.

What Is the Minimum Evidence Pack?

A good evidence pack does not need to be complicated. However, it needs to be consistent and easy to retrieve.

Keep:

• An AI use-case inventory
• Risk and customer outcome assessments
• Named owners and reviewers
• Approved policies and staff guidance
• Sampled outputs and review findings
• Error, complaint, and incident records
• Monitoring results
• Decisions and actions taken after reviews

How Should Firms Monitor Outcomes?

Start with measures you already track. For example, monitor complaints, repeat contacts, response times, client confusion, corrections, and quality-assurance findings.

Then compare trends before and after the AI process begins. If errors or repeat questions rise, investigate the workflow rather than assuming the staff member made an isolated mistake.

How Does This Fit the Board Report?

AI-assisted processes form part of how customer outcomes are produced. Therefore, material findings should feed into Consumer Duty governance and the annual board report on outcomes.

A board does not need every prompt and output. However, it should see the key use cases, controls, monitoring results, incidents, decisions, and planned improvements.

This makes AI oversight part of normal Consumer Duty governance. It should not sit in a separate technology folder that compliance never reviews.

What Should Happen After an Error?

First, contain the issue and assess who may be affected. Then correct the customer impact, record the cause, and decide whether wider remediation is needed.

Afterward, update the workflow. You may need a better prompt, tighter access, more review, a lower automation level, or a different use case boundary.

The most important point is to learn from the finding. A recorded correction shows active oversight. An ignored pattern suggests the control is not working.

How Can LaunchLemonade Support a Governed AI Workflow?

A governed AI workflow makes safe behaviour easier to repeat. Therefore, the right platform should help the firm apply its agreed controls in day-to-day work.

Why Does Central Governance Matter?

When staff use disconnected consumer AI tools, records, settings, and prompts can spread across many accounts. As a result, a firm may struggle to see what AI is doing or who approved it.

LaunchLemonade is an AI agent platform built for small and medium-sized businesses that need safe, secure AI agents. It supports work across meetings, research, client onboarding, and reporting. Crucially, it provides audit trails, role-based access controls, approval workflows, PII detection, and governance dashboards.

How Can Approval Workflows Support Human Review?

On Team and Enterprise plans, administrators can flag agent actions that need human review before execution. For example, a firm can require approval before an agent sends a client email, finalises a compliance report, or pushes data into a connected system.

This structure supports a simple Consumer Duty principle: AI may prepare the work, but a competent person approves the material action. You can explore the LaunchLemonade platform for teams to see how team governance can support that operating model.

How Can Audit Trails Improve Evidence?

LaunchLemonade logs every input and output for audit on Professional plans and above. Moreover, Team and Enterprise plans add governance and reporting dashboards that help administrators review that activity.

This can make evidence easier to retrieve when you need to review a client-impacting workflow. It does not replace judgement or Consumer Duty monitoring. However, it can make the evidence trail a built-in part of the work.

Can Advisers Build Their Own Controlled AI Agents?

Yes. LaunchLemonade’s no-code agent builder lets domain experts create and customise agents without engineering support. Teams can use ready-made agents, adapt them with their own documents and workflows, or build an agent from scratch.

Professional and Team users can access more than 300 large language models, including major frontier and open-source options. However, model choice should always follow the use case, risk level, and review controls. To explore a practical build approach, visit the no-code AI agent builder for financial services teams.

For a walkthrough of how governance features could fit your firm’s processes, you can book a LaunchLemonade demo.

What Is the Best First AI Use Case for an Advice Firm?

The best first use case is useful, low risk, and easy to review. Therefore, start with a process where a person can check the result before it affects a client.

Why Should You Start With Drafting and Summaries?

Drafting and summarisation can save time without requiring unsupervised decision-making. For example, AI can produce a first draft of meeting notes, an internal agenda, or a client letter.

The adviser then checks the work against the source material. This creates immediate value while preserving professional judgement.

Which Use Cases Need More Caution?

Treat direct client interaction, financial guidance, complaint handling, vulnerability assessment, and money movement as higher-risk areas. These uses may still be possible. However, they need strict boundaries, rapid human hand-off, and closer monitoring.

Do not confuse a polished answer with a safe answer. The higher the customer impact, the more the firm needs to validate the whole process.

How Should You Run a Pilot?

Set a clear scope. For example, run a four-week pilot for AI-assisted meeting summaries used only by two advisers.

Before the pilot begins:

• Define the purpose and output standard
• Choose the approved tool
• Set data-handling rules
• Require reviewer checks
• Decide what evidence to collect
• Set success and stop criteria

At the end, assess quality, time saved, corrections, staff feedback, and customer impact. Then decide whether to stop, improve, or scale.

What Does Good Adoption Look Like?

Good adoption is controlled, not slow. Staff know which tool to use, what they can enter, when they need approval, and how to report a problem.

That clarity reduces shadow AI use. It also lets the firm build confidence from real evidence rather than from guesswork.

Key Takeaways

• Consumer Duty applies to AI-assisted activities when they affect retail customer outcomes.
• The FCA’s approach remains principles-based and outcome-focused, rather than based on a separate AI rulebook.
• Your firm can delegate tasks to AI. However, it cannot delegate accountability for the outcome.
• Client-facing output needs meaningful human review, especially where it includes personalised or material information.
• AI can support consumer understanding and customer support when firms design clear boundaries and escalation routes.
• Known AI errors can create foreseeable harm. Therefore, firms should test, sample, monitor, and improve their workflows.
• A proportionate AI governance framework starts with clear ownership, an inventory, use-case limits, and evidence.
• LaunchLemonade can help firms centralise controls through audit trails, access controls, approvals, PII detection, and governance dashboards.

Conclusion

Financial advisers can use AI under Consumer Duty without waiting for a separate regulatory regime. However, the firm must retain control of customer outcomes. That means setting clear limits, keeping human judgement in material processes, and monitoring real results. In short, delegate the task, but never delegate the outcome.

If you want governance to be part of your team’s daily AI workflow, rather than a promise in a policy document, book a LaunchLemonade demo. You can see how controlled AI agents, approval paths, audit records, and role-based access can support a safer rollout.

Frequently Asked Questions