The benefits of AI in regulated industries, and why explainability comes first

Regulated industries are often described as slow to adopt AI. In practice, most are adopting it carefully, and with good reason. In insurance, banking and healthcare, a decision has to be explained to the customer, justified to the regulator and, sometimes years later, reproduced exactly.

These requirements do not prevent the use of AI. They shape how it should be designed, and the firms that plan for them from the start are the ones seeing results.

The benefits

For regulated firms, AI tends to deliver value in four areas:

• Speed to market. Products and changes that used to take months can be built and tested in days.
• Consistency. A well-tested automated process applies the same rules every time and reduces variation in manual work.
• Capacity. Specialists spend less time on data entry and document review and more time on decisions.
• Better records. Systems built for AI usually log, version and test more than older processes, which makes audits easier.

The last point is often overlooked. A well-designed AI process can leave a firm with clearer records of how decisions are made than it had before.

What regulators expect

Expectations vary by market, but they share common themes:

• In the UK, the FCA's Consumer Duty requires firms to deliver good outcomes and fair value. The FCA's general insurance pricing rules, in force since January 2022, stopped firms charging renewing home and motor customers more than equivalent new customers. Firms need to be able to show how their prices are set.
• In the EU, the AI Act treats AI systems used for risk assessment and pricing of individuals in life and health insurance as high-risk. That brings obligations covering risk management, data governance, documentation, transparency and human oversight.
• In the US, the NAIC's model bulletin on the use of AI systems by insurers, adopted by a growing number of states, expects insurers to have a written AI governance programme and to show that decisions supported by AI comply with laws on unfair discrimination and other insurance rules.

Taken together, the message is that firms can use AI as long as they can show how each decision was reached.

Building explainability into the system

Some firms try to meet these expectations after the fact, by producing explanations for models that were never designed to be explained. This approach is fragile. It is more reliable to make explainability part of the system:

1. Readable logic. Write rules and rates as readable configuration rather than program code, so the logic serves as its own documentation.
2. Full traces. Make every decision reproducible, with the inputs it used, the rules it checked and the value produced at each step.
3. Versioning. Create a new version for every change and record which version made each decision.
4. Tests as evidence. Run a set of known cases with expected outcomes on every change to show the system behaves as intended.
5. Human approval. Let AI draft and propose changes, and require a named person to approve them before they go live.

AI drafts, people decide

The pattern that meets both business and regulatory needs is straightforward. AI handles drafting, extraction and checking quickly, and accountable people approve the results. The AI speeds up the work and the design of the system makes it defensible.

Ptally applies this to insurance pricing. Models are readable configuration with tests included. Every quote can return a full debug trace showing each input, lookup, formula and rule. AI agents can edit draft versions, but only people can approve and publish them, and every status change is recorded.

Summary

Regulated firms do not have to choose between using AI and meeting their obligations. Explainability, testing and human oversight are what allow an AI project to go live, and firms that plan for them early tend to move faster.