Ron describes it as a “replacement cognitive layer” that can be inserted into business processes, particularly where language, documents, and human judgement are central. That matters because insurance is not manufacturing. It is a promise-based industry, built largely in natural language: policy schedules, endorsements, claim forms, broker correspondence, underwriting notes, complaint logs, regulatory submissions, and customer communications. This is the domain where large language models are at their best.

A practical decision framework for businesses - For many insurers, especially smaller businesses, the question is not whether AI will matter, but where to start, and how to do it without wasting money.

Ron’s framework begins with a simple distinction:

1. Do you need specialised quantitative modelling, or do you need language-based cognition?
   If the task is deeply quantitative (for example, asset allocation optimisation), you need specialist tools and models. But if the task involves understanding documents, extracting information, detecting errors, or generating responses, large language models can be a strong fit.

2. Test capability quickly, before building big projects.
   One of the advantages of generative AI is that you can assess usefulness immediately. Using a secure, controlled chat interface, you can trial real business documents (with the right precautions to prevent training on sensitive data) and evaluate output quality. If it works on day one, you have a clear signal.

3. Then wrap process, controls, and automation around what works.
   Once the outputs are reliable, the next step is putting governance around it, repeatability, monitoring, review, and safety controls. That’s where real value is created, without exposing customers or the business to avoidable risk.

This approach is especially relevant for firms still cautious after years of expensive and painful technology implementations. Ron acknowledges that insurers have been “bitten” before, large, multi-million-rand system projects that ran over time, over budget, or failed outright. But he believes generative AI differs in one important way: you can see quickly whether it solves your use case before you commit to the heavy build.

Where the biggest gains are coming - Looking across the insurance value chain, Ron sees the most immediate upside in any process involving high volumes of documents:

• claims documents and claim form validation

• policy schedules and endorsements

• broker email correspondence and information requests

• comparative quoting where data must be pulled from multiple sources

• payment processing and routine claim handling with built-in safeguards

He also highlights a quieter but powerful frontier: personalised policyholder communication. Instead of generic letters, insurers can tailor communication based on context, claims history, risk behaviour, policy changes, or engagement patterns, in a way that feels genuinely human, not templated.

Beyond that, there is a strategic layer: improving the work of scarce, high-value specialists. Actuaries are expensive and rare, and their decisions influence pricing, reserves, and capital. Tools that help them produce more up-to-date insights and faster scenario analysis can become a meaningful competitive advantage, not just a cost-saving exercise.

Choose wisely and execute carefully - The temptation with AI is to try everything. The disciplined reality is that no organisation has infinite bandwidth. Ron believes the industry will increasingly separate AI projects into two categories:

• Efficiency projects: streamlining claims, communications, document processing, and internal workflows

• Optimisation projects: improving pricing, risk selection, reserves, and capital decision-making

Both matter. But leadership is about sequencing, selecting projects that can be executed properly and will deliver clear value, while managing the organisation’s capacity for change. Because ultimately, AI is not just a technology decision. It is a change decision.

And in insurance, change only becomes progress when it is implemented with clarity, governance, and a sharp focus on where the value is real.