Credit where it's due: using AI to enhance underwriter performance

The financial services industry is opening its eyes to the vast possibilities offered by Artificial Intelligence (AI) and one of its offshoots, machine learning.

Many financial firms and other sector bodies are already exploring and implementing AI for a range of reasons. These include: boosting employee efficiency, making the use of consumer data more compliant and transparent, and analysing that data to build products and services that people will truly value.

One of the perceived drawbacks of AI, however, is that if robots can out-think humans, there may follow tens of thousands of redundancies.

Having pioneered a new machine learning-based model for a front-line underwriting function we discovered that although AI can build in the efficiencies that CROs crave and the transparency consumers demand, humans will always have a crucial role to play in decision-making.

Ultimately the objective is to automate decisions that take a few minutes for a person; 60% to 80% of the overall queue in credit underwriting, and let underwriters focus on the specialised decisions they have been hired for.


Removing inconsistency from decisions

For decades the approach to underwriting has remained relatively unchanged; relying on tried and trusted processes such as scorecards, policy rules, and the expert judgment of the individual underwriter. And, to be fair, these traditional techniques must have been successful, or they wouldn’t have survived for so long.

But driven by the financial services industry’s exposure to the rise of AI, there is now a clamour for new ways of thinking. In underwriting, this means finding innovative solutions to address key recurrent issues such as:

  • Improving overall efficiency - what can machines do better than underwriters, and how does this affect individual workloads?

  • Providing consistent and accurate decisions to all credit applicants

  • Delivering a reduction in natural human bias, which can affect judgement relating to individual cases

  • Ensuring transparency of the decision-making process with fair explanations recorded

Knowing how far to take AI

4most was commissioned by a lender to consider how AI could be used to tackle these very issues.

Although AI can build in the efficiencies that CROs crave and transparency that consumers demand, humans will always play a crucial role in decision-making.

Choosing the right model for your business is vital. The complexities of underwriting demand a novel approach that isn’t overly simplistic in order to accurately represent the intricacies of manual decisions. Yet it must also be sufficiently transparent to understand the reasons for decisions. To address these twin objectives, we developed a model that uses the random forest technique - and it’s paying dividends.

Firstly, the solution harnesses machine learning to standardise rules that have always been true in the underwriting world. In other words, we’ve set in stone the decision-making process.

Secondly, a key benefit of using machine learning for ‘simpler’ underwriting tasks is to free humans from dull administrative jobs, enabling them to focus on other aspects of the role that add value.

Thirdly, a huge number of cases can be processed in minutes by machines, while humans get on with the more complex tasks mentioned above.

Finally, by following a set of rules that remove human error and bias from the process, the solution automatically builds fairness, consistency and transparency into decisions. There’s a comprehensive audit trail available for the CRO and the regulator to check the outcome of applications should they wish to do so. We know from the client feedback that this approach is already providing a clear line of sight to managers, that wasn’t previously possible.

It’s greatly to their credit

We set the algorithm to work as part of a control test that pitted the model against the manual decisions of an underwriting team. This was achieved by processing a large volume of applications through the solutions that had already been assessed by the underwriters.

The model produced an accuracy rate of 95%; but outstripping the performance of the human underwriters, was the fact that the solution was proven to be more consistent in the decision process.

So, what does this mean for the future of underwriting and for those who currently hold a role in the profession? Should they be looking for something else to do?

Well, it’s not a case of everyone packing up their desk or needing a side hustle straight away. There are many trickier tasks that will always require human intervention and ingenuity.

It’s the transfer of simpler tasks where the biggest prize of AI lies at present; a win-win for CROs and their underwriting teams but also for applicants, who can receive clear, honest - and in most cases - speedier decisions.

People and machines working in harmony, using new techniques for better results: that really would be a great credit to the underwriting industry. There are exciting times ahead.


For further information on how AI and Machine Learning can benefit your business, please contact our Head of Data Science, Fabrizio Russo: