Top 3 lessons actuaries can teach data scientists

This follows on from our previous post about the Top 3 things data science can teach actuaries.  This time, we flip it over to look at what actuaries can teach data scientists in return.

1. Making room for business judgement

Data science techniques have the benefit of being scientific, objective and rigorous. The results are what the data shows. Models represent what happened in the past. This is all very useful for insurers.

However, many insurance problems require a strong dose of business judgement overlaid upon the data science before they can be used in the real-world.

For example, machine-learning models might identify which customers would take up a new offer. It may also provide a clear business case for doing so. But in the real-world, business judgement needs to be carefully applied also: Is this appropriate for all or just some of these customers? How do we connect with these customers using our operations or distributors? What does regulation enable us to do? etc.

Actuaries have more skills in applying this holistic business judgement. The data science forms an important part of a longer chain of analysis and judgement that is needed to take the right action.

2. Actionable insights

Unfortunately too many data science projects still end up being interesting pieces of research that do not generate real impact in the business.

If done well, data science should enable staff to make better decisions day to day. Customer interactions will be more relevant and engaging. Executives will make more effective decisions.

Actuaries are well plumbed into the insurance operations and the organisation. They can focus the data science on areas where the insights have the best chance of being useful to the business. They can also help translate the data into a form that front-line staff can digest more easily and intuitively in their day to day work.

Models may also be actionable in unexpected ways. For example, models of customers’ investment portfolios can uncover propensities, risks and other characteristics. Standard commercial actions might use this to try to improve pricing or targeting of customers. But with a more actuarial lens, we might uncover more valuable actions, such as how to better mitigate risks in some pockets of the portfolio for the benefit of both customers and insurers at the same time.

3. Paradigm shifts

Being a heavily regulated industry, things tend not to change very quickly in insurance. Patterns of buying behaviour and product mix tend to drift along and change slowly over time. Insurers are rarely nimble, and they usually haven’t needed to be.

But paradigm shifts do happen, often triggered by regulation.  For example in the UK, the Retail Distribution Review, and annuity taxation, dramatically changed the behaviours in the life insurance and pensions market.

From a data science perspective, paradigm shifts are a problem. They undermine the value of the historical data to the point that the results are certainly less useful, and may be very misleading.  Data scientists are of course aware of this, but are constrained by the data they have available.

Actuaries often work in situations where they need to make judgements, but with more limited data. The profession’s training merges the mathematical, business and risk backgrounds to do this. In my experience it encourages an approach that results in a considered and balanced assessment of the facts as they present themselves.  This is very useful when paradigms shift, and is a valuable skillset from the actuarial toolkit that data scientists can benefit from.


Insurers are slowly becoming more sophisticated with data. But to make real breakthroughs, we believe that we need to combine the data science and actuarial toolkits much more effectively than today.