The first is the ability to fully understand the impact of a potential change in this industry where every change has a corresponding cost. The second is the ability to thoroughly explain the impact of innovation to non-technical stakeholders by addressing the key issues they care about. It's hard to change actuarial processes. It's easy to say we should do things differently, but it's difficult to come up with a good way to be innovative.
New actuaries are often the most creative. They come in with fresh perspectives and ideas, but are often rightfully shot down. Sometimes these new perspectives are interesting, but they just don't work in practice. So, a lot of new actuaries tend to get their ideas squashed because they don’t fully understand the reasoning behind the way things have always been done. This means that another important feature is persistence; this ability to accept when a suggested change is a bad one, and then keep going anyway. Having courage to share their ideas, accept when it's a bad idea, learn from it, and then keep sharing until you find an innovation that makes sense for everyone.
Co-workers and supervisors are usually quite comfortable telling you when a recommendation won’t work, and asking why it won’t work is a good idea. When I was doing a rate review early in my career, my manager didn’t like one of my selections, but I was convinced that deviating from the “normal” way was correct. My manager allowed me to present my selection so that I could experience first-hand how stakeholders reacted to it and what questions they asked. They ultimately shot down my idea, but they explained why they rejected it in their own words and in terms of issues that they cared about. I quickly learned that the actuarial sphere is a small part of the insurance ecosystem, and our decisions have wide-reaching consequences. It’s valuable to learn that an idea might make sense in a vacuum, but not make sense for other parts of an organization.
Ask questions when people tell you an idea isn’t a good fit. Learn why it's a bad idea, and in doing so, you'll understand more about the organization, your company, and the industry as a whole.
Actuaries need the opportunity to fail, to have disagreements with everyone at all stages of their career, and the ability and freedom to disagree constructively. When your manager gives you the opportunity to be wrong, that's fostering a culture of innovation. What actuaries need is leadership that encourages innovation in a way that welcomes good ideas, comfortably shuts down bad ideas, and encourages the development of ideas that are somewhere in-between. At a later stage of your career these ideas can be broad strategies discussed with other leaders.
It's important for leadership to acknowledge there may be a better way to do things, to be open to uncomfortable change, and to give innovation a chance. Innovation flourishes when leadership listens to ideas and provides constructive and transparent feedback to help develop employees.
People often talk about the move to R or Python instead of Excel, and while they are right to do so, we're going to have to make some tough decisions about what the actuarial coding skill set should include. Learning a programming language is a no-brainer - that's something that will absolutely be valuable. But I have seen some actuaries start taking on roles that are closer to that of a computer programmer or a cloud engineer than an actuary. They learned how to code and they're now standing up an entire platform for their department.
Do we want to use actuarial talent to build an actuarial automation script? Yes, we want to do that.
Do we want to use actuarial talent to troubleshoot and update that automation script for every new update to R as the packages used are updated or become unsupported? Probably, but the best use of an actuary’s time is not updating code.
Do we want to use actuarial talent on upkeep of a robust cloud-based automation tool? Now we're squarely in the role of a software or cloud engineer and this work is better done by non-actuarial partners.
Drawing the line on coding is going to be difficult. Are we going to include defensive coding in our syllabus? That’s not a key actuarial skill, but it is a good skill. Where does that fit in? How about robust code review? That is an important topic because actuarial analysis has to be reviewed, but it's difficult to figure out exactly where code review strategies fit into our learning and credentialing process. Where we allocate our actuarial coding capacity is going to be key.
For example, let’s take Power BI. A gross oversimplification of Power BI is that it creates graphs and charts. Given enough time, actuaries can create these graphs and charts in R or Python. However, that's a bad use of your time when tools already exist that automate increasingly complex - but universal - processes and allow actuaries to spend their coding capital on things that are most meaningful and unique to the business. There are going to be a lot of useful ad hoc coding projects, but they'll be more removed from this software engineering, cloud-engineering type of coding.
The actuary’s job will soon become an 80/20 split of reviewing/doing as opposed to 80/20 doing/reviewing. Steve Armstrong and Jamie Mills recently wrote an article on how to automate actuarial processes. They talked about the feedback loop of automating a process, reviewing it, then running the automated process again with updated inputs or assumptions. Data comes from a notoriously fickle real world, so any type of actuarial automation needs to have eyes on it. This review will become a significant portion of the actuarial work as opposed to putting together a script or dragging down a column in Excel. Actuarial work will move towards reviewing final calculations that are done, not manually creating them. Understanding the assumptions of a technique is going to be way more valuable than the ability to code it out.
Right now, many actuaries are using R and Python essentially as an advanced Excel. We need to take advantage of automation in new ways instead of automating the old ways. Why not automate to get a range of 30 indications to review instead of automating our current process for a single indication?
The exact definition of what is an effective use of the actuarial skillset as it relates to coding will be very important in the evolution of the actuarial profession.
Whenever anyone says, “I could never understand that,” take it as a personal challenge to be able to explain it to them. Learn to talk to the level of your audience in a language that they understand and relate to. When you're explaining a model, you're very rarely explaining it to another modeler. Many times, you're explaining it to someone who has an absolute wealth of industry knowledge but doesn't know how a GLM works. You need to be able to explain it fully in a way that they can relate to and then make the best business decision based on your explanation. Your ability to communicate will directly correlate to your ability to be an influential actuary.
Whenever you learn a statistical technique, think about all the ways that real world data can cause it to misbehave. When you're learning mathematical formulas, don’t just think, “I know the formula, I can apply it.” Think of the ways that it can break. Think of the assumptions behind it that could be voided by real world data. This is the same reason that actual work is not going to be fully automated. Blind application of modeling and statistics to real-world data is a recipe for trouble.
A lot of the problems come from the difference in expectations of what material is and the expected outcome of the work. I find that a lot of data scientists love getting 95% of the way with 10% of the work. Actuaries might say, “Well, you have to do the other 90% of the work to get this last 5% and be the most correct.” We often don't have the same view of what a finished product is, and we don’t always agree on the consequences of an imperfection. In a pure data science project, it might be okay to have a model that is correct 95% of the time. That’s a pretty darn good model, right? In actuarial science, that 5% might be a huge opportunity for adverse selection and a recipe for unprofitability. In order to get data science and actuarial science to collaborate more effectively, the two groups need to meet before a project begins to align on the expected result, the materiality of different types of errors, and specific success criteria.
There's always the answer that innovation opens new opportunities and can improve your business and improve your combined ratio. Of course that is what innovation can do, but innovation also brings up a sense of achievement and ownership within a process and a sense of collaboration between departments. Within a department, innovation creates buy-in. “I've developed this thing. I own it. I am proud of it, and I'm going to do my best to make this work.” When employees are allowed to innovate, they're more invested, and that really helps the company as a whole.
Additionally, actuarial innovation does not often stay within the actuarial department. It can influence an entire organization. Explaining your innovation to non-actuarial business partners also helps foster collaboration across the organization. Innovation helps build those connections and break down the silos between departments that might not normally talk to each other. The more impactful an innovation is, the more it reaches across departments, the more often people are reaching out to other departments, the greater the internal alignment of a company.
I was going to say machine learning and AI, but these would not be adopted without corresponding innovations in data, visualizations, and communication. I think the single most influential hour of my career was seeing a presentation on the “How” of presenting. What is a good or bad presentation style? How do you keep your audience's attention? That session helped me communicate ideas and emphasize the points that I wanted to make in a way that was most applicable to my audience.
New innovations in communication allow machine learning and AI projects to exist. New innovations in data visualizations allow the machine learning techniques to be understood and implemented. Everything starts with communication. So, no matter how good your product is or how good your model is, if you can't explain it, it's never going to be used. As machine learning gets more complicated, the communication has to get simpler. Communication is leading, and complex machine learning is following.
Actuaries are adopting a lot of changes right now. They are learning to code, and the syllabus is being updated with advanced techniques and machine learning. I'm going to reemphasize the importance of communication as we make this journey. As we adopt R and Python and machine learning techniques, we need to explain increasingly complex topics at the same level of understandability.
The change that I would emphasize in the actuarial world is this: as we make the methodology changes, as we make the technical and machine learning changes, we need to keep pace with innovations in clear communication and explanations. People love machine learning, but, without that ability to communicate it's not going to go anywhere.
Actuaries, as they're looking to be innovative, should look for any opportunity they can to work with people outside of their department. Learning the upstream and downstream impacts of a change is how you get to ideas that can be implemented and how you get to innovations that make sense for the organization as a whole. That's what I would say to any actuary that wants to be innovative. Learn about non-actuarial functions and learn what the other divisions do.
