The complementary role of Artificial Intelligence in Insurance CPD

Artificial Intelligence (AI) is seldom out of the headlines. Some suggest it promises a utopian future, whereas others predict far darker consequences. What is clear, however, is that AI is already having a transformative impact on our economy and society and we cannot ignore it. This includes the effect of AI on how we work and how we learn. This article explores how generative AI can complement workplace learning, with specific reference to how insurance professionals can use the technology effectively.

From the outset, it must be emphasised that education within the insurance industry should remain grounded in credible and authoritative sources. AI does not replace CII qualifications, textbooks or e-learning platforms. These provide structured, often expert-curated learning programmes aligned to industry needs and should remain the foundation for professional study. Workplace practice, mentoring, regulatory publications and the industry press also remain important in achieving professional competence.

Generative AI can, however, play a complementary role in developing knowledge and improving understanding. Applied carefully and purposefully, it can question, challenge and explain, simulate situations and provide real-time feedback. Many of these uses also reflect established principles and approaches within learning theory.

AI has attracted legitimate criticism in education. It can generate unverified learning materials containing errors, invented information or dubious citations that learners then rely upon. These are often referred to as ‘hallucinations’. Learners may also use AI to generate plausible answers to assignments without properly engaging with the questions. Irrespective of the veracity of the output, generating an answer is not the same as employing the knowledge and reasoning required to reach it independently. Used in this way, AI may reduce the cognitive effort involved in learning.

The educational value of AI therefore depends heavily on how it is used. A useful principle is to use it in ways that make the learner think

Most of us have likely used a Large Language Model (LLM) such as ChatGPT, Gemini, Copilot or Claude. Useful outputs depend on effective prompts, entered through text or increasingly voice.

When prompting, think about what you want to achieve. It is useful to give the LLM a role, such as that of a supportive tutor. Give it clear instructions, relevant context and any source material it should use. You can also add constraints or parameters, such as telling it not to reveal an answer before you have attempted the question, or give it examples of how the output should be structured.

Suppose you wanted to assess your knowledge of a particular topic. You could ask the LLM to question you and provide feedback. Someone new to insurance could use the following prompt:

Assume the role of a tutor providing training to insurance professionals. I am new to the insurance industry and work as an apprentice in a general insurance brokerage in the UK.

Ask me ten questions relating to the fundamental principles of insurance. Ask one question at a time, allow me to respond, provide feedback and then move on. Do not reveal the answer before I respond. Then assess my overall knowledge and identify any areas in which I need to improve.

This provides immediate feedback while requiring the learner to recall information from memory. Reading course resources or completing e-learning materials can create familiarity with a subject, but familiarity does not necessarily mean that knowledge can subsequently be recalled or applied.

This utilises retrieval practice, where learners actively recall information independently rather than having to refer to notes or learning materials. The feedback also provides useful formative assessment for the learner, helping them identify strengths, knowledge gaps, and misunderstandings while learning is taking place.

AI also allows the difficulty to increase at the learner’s own pace. You could move from simple questions to scenarios requiring application. The AI can be prompted to provide ‘scaffolding’ by offering hints or explanations when needed, then gradually reducing that support as the learner becomes more confident and capable of applying the knowledge independently.

Another way AI can support understanding is through its ability to explain concepts in different ways. This is particularly useful in insurance, where legal terminology, policy language and industry jargon are frequently used.

Consider the following statement:

Where an insured under a non-consumer insurance contract has failed to make a fair presentation of the risk, the insurer will have a remedy only if it can show that, but for the qualifying breach, it would have declined the risk or accepted it on different terms. The remedy depends on the nature of the breach and, where it was not deliberate or reckless, what the insurer would have done had a fair presentation been made, in accordance with the Insurance Act 2015.

This may be clear to an experienced practitioner, but less so to someone early in their career. Repeatedly asking a human tutor or colleague for clarification can cause social discomfort but using AI allows this to be done multiple times in a virtual environment, where learners can ask for repeated clarification without fear of embarrassment or judgement.

A learner could prompt:

Explain the following technical insurance information in plain English for someone who is new to the insurance industry. Explain technical terms clearly, but do not omit important qualifications or change the legal meaning.

The learner can ask for a different explanation if the first one does not help and continue asking questions until they understand the concept.

This is consistent with cognitive load theory, which recognises the limited capacity of working memory. Complex subjects can be easier to understand when broken into smaller components and introduced progressively.

The learner can then go further and explain the concept back to the AI and ask it to identify anything missed or misunderstood. This self-explanation requires the learner to organise and articulate their understanding before progressing with the original learning material.

AI can also support learning about a particular insurance product or service. Care is needed when uploading information to an LLM. Personal, confidential or commercially sensitive information should not be uploaded to a public LLM. Documents should be uploaded only where your organisation permits it, to AI systems authorised by your employer, and in accordance with its information security, data protection and AI policies.

In insurance, where permitted, you could provide approved policy documentation and ask the LLM to simulate a prospective customer interaction:

Using only the policy documentation I have provided, assume the role of a customer considering this insurance product. Ask me questions about the cover, significant exclusions and important conditions.

After each answer, compare my explanation with the documentation provided. Tell me if I have omitted an important point and identify the relevant section of the document.

This can provide coaching for client interactions in a controlled and low-risk setting. Relevant FCA rules or guidance could also be provided, or signposted, to the LLM and used to prompt review of aspects of the interaction relating to Consumer Duty outcomes, such as consumer understanding or consumer support. However, any regulatory interpretation produced by the AI should still be checked against the FCA Handbook, FCA guidance and the firm’s own compliance procedures.

Using AI to simulate potential client conversations can allow learners to practise realistic situations in a safe environment. They can participate in a mock client conversation, receive feedback, consider what they could improve and repeat the exercise to develop their performance. Using different customers and circumstances also requires learners to adapt their approach. This reflects experiential learning, where skills are developed through practice in realistic contexts.

AI can also encourage learners to reflect on their own thinking. After a scenario, it could ask why you chose a particular response, how confident you were or what further information you would need before advising the customer. This helps learners identify the limits of their knowledge and examine their reasoning.

There is, however, an important qualification to these examples. An AI system can give incorrect content or feedback. It may tell a learner that a wrong answer is correct, oversimplify a technical principle or produce a confident but inaccurate regulatory interpretation. As such, AI works best as a tool to complement and enhance traditional learning resources and should not be used as a substitute. Where accuracy is important, the learner should return to the authoritative sources and the relevant learning materials to verify the position.

The educational potential of generative AI lies particularly in its ability to interact with individual learners. Used carefully, it can challenge understanding, simulate practical situations and provide immediate, non-judgemental feedback. It can also encourage learners to reflect on their reasoning and engage more actively with authoritative knowledge and established CPD methods.

Professional qualifications, authoritative sources and workplace expertise should remain the foundations of insurance learning. Generative AI can complement them by helping learners make sense of complex concepts, retrieve knowledge, explain their reasoning, apply what they have learned to realistic situations and adapt their responses. Its educational value is greatest when it supports understanding while requiring the learner to think.