Improve AI efficiency by reducing input variation

Jack Welch, the former CEO of General Electric (GE) was famous for his use of the expression, “variation is evil”.

We are not sure if he coined the term, but it was his mantra, right across GE’s diverse range of businesses, as they introduced Lean Six Sigma with exceptional results.

Dr William Deming, the father of Lean Six Sigma, said it was all about getting the process right. Process was more important than people. We know now that both people and processes must be employed to full effect to get the best out of the modern insurance business.

Nevertheless, Welch’s mantra about variation rings true with the advent of AI, as minimising the variation of inputs at all steps of associated processes is important, especially at the beginning.

Removing the effort from processes

The fewer the variations, the easier it is for AI to work efficiently to streamline processes.

To explain using a simple example, think of every email you receive being delivered using a consistent web format with multiple choice options. In such a scenario, an AI solution could easily perform most of the initial processing.

The more consistent the inputs, the quicker and more effectively the AI solutions can remove both labour costs and mundane tasks from your processes to allow human resources to focus more time on customer care and new business opportunities. 

The challenge of standardisation

Developing the algorithms and "training" the machine to learn what is happening and how to handle a larger number of exceptions is arguably the easiest part, because these are essentially the “nuts and bolts” of an AI solution.

The hardest part of implementing AI is the change management required to make sure that the various data inputs into an insurance product life cycle are harmonised a.k.a. reducing variation.

We all know how difficult it is to get things standardised in insurance. Everyone has their unique set of underwriting questions, proposal inclusions and ways to send or share simple requests or documentation, such as quotes between stakeholders including brokers, underwriting agencies and insurers.

Given that it is also unlikely that insurers will collaborate to standardise datasets, formats and inputs, implementing an AI strategy in an insurance environment appears difficult but by no means impossible.

Starting simple in claims

From a claims perspective, general insurance inputs can be multiple and highly varied.

However it is interesting to note that AI solutions have most easily been introduced to automotive insurance claims processes, as they are more straight forward than other types of claims, and have minimal input variations.

Along with making AI solutions significantly more effective, reducing the variation of inputs has a further direct bearing on efficiency by decreasing the amount of rework required, which is one of the main challenges facing insurance businesses trying to implement technology solutions.

Granting space for development

We need to keep reminding ourselves that we are still at the beginning of the AI journey and the way forward may well continue for some time.

As a suggestion, you could start an organisation-wide implementation using the 80/20 rule, meaning a focus on solving the majority of issues and realising most of the automation opportunities.

Over time, more and more exceptions will be effectively handled by AI, and the speed at which that occurs will depend on how well we manage to reduce the variations in process inputs.

Employing digital collaborations

One solution being used by a progressive insurance technology provider, is to develop a digital laboratory in partnership with an underwriting agency client. The aim is to explore and test better processes that support AI and digital tools to improve efficiency and customer outcomes.

It is expected that developing AI tools using this kind of direct, collaborative approach will enable new technology to be tested and implemented far more quickly than traditional models which involved developing technology solutions prior to client testing.

As a result, efficiency gains should be achieved sooner through proven technology solutions that keep the insurance industry at the forefront of a rapidly growing, competitive digital landscape.