What does the market really think about AI?

AI is the hot topic of the moment – but what is the market really saying…

August is here, giving me a well-deserved break from the conference circuit. Hey, it’s not all drinks receptions and schmoozing you know! The first half of the year has been dominated by one topic: AI, which has all but replaced previous years’ panel discussions on embedded insurance, the IoT and, before that, GDPR.

The official ‘company lines’ I hear from clients speaking at events are often very similar: ‘We’re just starting our AI journey, cautiously, but we know it’s the future.’ But everyone knows the real conversations happen over a pint afterwards!

Here’s what I’m hearing…

AI replacing humans in the workforce is a long way off!

Taking the human element out of even quite basic tasks is not something you can do overnight. Initially, expert human oversight will be essential to validate and refine AI-generated outputs. Humans need to act as quality controllers, verifying the AI-generated results and adjusting rules and process as needed.

It makes sense to run AI in parallel with human testing to assess and verify accuracy. When the AI consistently performs well, you can reduce human intervention. But periodic quality checks will remain essential to make sure you’re meeting your audit requirement and internal quality assurance benchmarks.

Data quality is vital – poor data means poorly trained AI

If the training your new hires receive is rubbish, so will be the output from their work. The same goes for AI. It’s common sense really! The data fed into an AI system needs to be of high quality. Poor data quality will lead to inaccurate outputs, regardless of the AI’s intrinsic capabilities.

AI can only be relied upon fully when you can be sure of something approaching 100% data accuracy (say in the 90s at least, depending on the task), and when you have similarly high levels of clarity and confidence in your process and the rules you’ve implemented – or are seeking to implement. Only at that point, can you achieve an equivalent output to what you’d expect from an experienced insurance professional.

Some clients are replicating existing processes…

AI can be highly effective in scenarios where a process is thoroughly understood. This allows for precise rule-setting, which AI can follow to deliver accurate results.

AI can significantly speed-up processes when used appropriately, making it particularly useful for mundane time-consuming tasks where the inputs and rules are of sufficiently high quality. It’s ideally suited to mundane time-consuming tasks.

…others are using it to take on tasks no human can manage

Using AI to review large swaths of information, or data points, or the meta data that sits behind files is something no human can do en masse. The potential of AI lies, not simply in replicating familiar human activities, but also in taking on tasks we humans just don’t have the bandwidth to handle. AI can tell you things about your business and the processes within it that human analysis can’t move fast enough to pick up. The only limit is your imagination – and your ability to ask the right questions.

Iterative improvements are key

Using AI effectively requires making iterative improvements. Start by using AI on small well-known activities, validate its performance through human oversight as you gradually enhance both data quality and rule precision. One thing seems clear to me in all the conversations I’ve had is that, if AI is to fulfil its promise of freeing human experts from routine tasks to focus on activities that add real value, there is still much work to be done, and, so far, everyone seems to be proceeding with caution.