SME e-trade volumes are exploding. It’s making the market far more efficient. But this success has led to a flood of SME business being referred to underwriters. This causes a real dilemma for insurers – how do you wade through all these requests whilst keeping service levels high?
The land of the blind
All too often the answer is based on little more than luck and a gut feeling. The gut of an experienced underwriter can be a powerful thing, but even the most experienced underwriter can only make a call on the referrals they actually get to look at. Given the complexity of SME insurance needs, the number of cases that require highly trained input can be overwhelming, and our competitive market makes wasted effort costly.
The status quo frustrates insurers and MGAs, given the considerable effort each case demands. As seen in the recent Insurance Times survey, it equally frustrates brokers with poor turnaround times and sporadic responses. This is not to mention poorer customer outcomes with fewer viable options for cover and slower service for quotes.
It is clear that things could be, and should be, better.
Where there’s data, there’s insight
With Acturis at the forefront of pushing digital trading, we have seen first-hand the benefits data can have. Our e-trade marketplace uniquely allows us to see the problem from both the insurer and broker perspectives. With our extensive experience enabling e-trade, we drew on our underwriting and broking knowledge to investigate further. As we investigated the referrals problem, the scale of the issue became clear.
Just looking at bound commercial business on the Acturis broker market last year, just over 20% of quotes required referral to the underwriter. Worse, the majority of referrals came from business that wasn’t bound. In the meantime, average referral response times have almost doubled since last summer.
The data confirmed the problem and got us thinking how we could improve outcomes for everyone.
The future: AI assistance
Machine Learning has the power to learn from the past. At its heart it uses various statistical models and algorithms to learn from data and apply that learning to new situations. The better quality your data and the more data you have to feed into the models, the better AI you get. Time and again you will hear that AI is only as good as the data behind it.
The insurance industry should be leading the practical application of AI. Insurance has always, from its earliest days, used observational data and statistical modelling to manage risk. This industry DNA should help with adoption; in effect, machine learning is nothing new — it is a natural extension to the way actuarial science has been applied.
With this, it is no wonder that the future of insurance is moving towards the integration of AI tools and services. This ranges from already established applications in identifying fraud, to analysing images of damage to make claims valuation more efficient, to streamlining and improving customer interactions, and lots more in between.
Looking to the broker SME market, there is the obvious answer: let well-trained AI assist underwriters. This will remove manual review for simpler cases and augment underwriters’ decision making for complex ones by providing insight into the risk. This will lead to a more engaged, efficient and effective workforce.
For all these reasons, we saw that AI could have an immediate impact by being used in the referral response process. Training an AI assistant across years of past referrals means it can direct underwriters to those referrals where they are most competitive.
Everyone wins: insurers will get better return on investment for their underwriter’s time, brokers get the best quotes back from referral quicker and the customer gets more competitive quotes for their needs.
Referrals done right
Having analysed the data and defined the problem, we started to apply machine learning to the data. A referral AI began to take shape.
With our newly trained AI assistant, we needed to find out how effective it could be, and the impact it could have. To measure this, we split the referrals that insurers had attempted to clear in the last year into two groups; those that the AI assistant would have recommended an underwriter target first and those identified as lower priority. We then looked at the performance of those two groups.
Behind the scenes
AI is only as good as the data behind it. It is therefore no surprise how much work is involved mining, cleansing and preparing the data before any machine learning takes place.
The Acturis Data Science team has performed extensive feature engineering (data transformation) to get the best inputs for the machine learning referral AI assistant.
One technique commonly used is one-hot encoding, turning categorical data (e.g. someone’s trade) into a series of Yes/No answers (e.g. Plumber? Yes!). This makes the data easier to understand for the machine learning algorithms and results in better AI models.
The results show a glimpse of the power of AI. The AI-recommended priority referrals had conversion rates 35% higher on average compared to referrals the AI flagged as lower priority. In reality, we are seeing similar performance of the AI with our customers who are using it live.
This clearly shows the Acturis AI model can help underwriters spot the referrals that will be competitive when cleared and can make the entire referral chain more productive.
For the Acturis Data Science team, this is only the beginning. We know that AI has the potential to make the insurance industry work better for everyone, and we are excited by the prospect of harnessing AI to enhance how our customers do business. Referrals are just the start!
If you would like to talk to me about referrals or AI and machine learning at Acturis, please contact me at firstname.lastname@example.org.
Gordon Jenkins is the Acturis Data Science Product Manager and has been working with data at Acturis since he joined us back in May 2008.