Intended Audience⚑

• senior actuarial managers
• chief actuaries
• senior insurance leaders

Key Takeaways⚑

• Programming is essential in actuarial and other analytical work, not an afterthought
• Actuarial teams need to be supported through training, access to tools, access to experts and adjusted hiring policies

Programming Model and Workflows⚑

Project Managgement and Communication⚑

• short development cycles - sprints, agile
• issue tracking, pipeline management
• Ditching email e.g. Slack, MS Teams, projection management systems,

Version Control⚑

• Version Control leads to:
  • clear version history
  • backup, easy to revert if necessary
  • understanding why things went wrong
  • easier collaboration (both within teams and across teams), a process to merge changes by team members
  • being able to view changes
  • code reviews
  • pull requests (links to project management and CI)
  • no more accidental changes
  • no more Jan 10 with RAS change final v3 RAS draft final v2.xslx

Continuous Integration⚑

• automated checks of changes
• automated builds
• automated deployments

Don't Repeat Yourself (DRY)⚑

• reduces chances of error
• better reuse of work
• modularisation

Project and Code Documentation⚑

Testing⚑

• test driven development
• unit testing
• batch testing
• reconciliation testing
• speed tests
• Aim to avoid backward steps, no regressions, speed slow downs
• Documentation coverage

Containerised Environments⚑

• reproducible code
• guarantees analysis/process works the same for everyone
• avoid deployment issues and manual work
• more responsibility for the developer

Quality⚑

A Recipe for a Team⚑

• Decide on a stack (but don't make it too rigid as new tech is coming in all the time) - Which base languages (R, Python, Julia, Scala, Matlab, C++, etc) - Which version control tool (Git, Murcurial, SVN) - Which version control server (GitHub, BitBucket, GitLab) - Which continuous integration and testing tool (Jenkins, CircleCI)
  • Coding style
    • For Python - PEP8, docstring standards
    • What for R
  • Documentation tools:
    • For Python: Sphynix, readthedocs
    • For R: Vignettes, etc
  • Learn Command Line, a base language and Git at the same time
  • Start with some

Standard Actuarial Team Model⚑

• 'Get some data', explore an idea
• some rough work
• gets developed
• becomes messy, move on to the next problem
• needs to be maintained indefinitely
• data stored in silos
• lots of spreadsheets and links
• copies of logic
• easy for errors to slip in
• difficult to test
• how to version control
• how to refactor
• poor data visualisation options
• cannot extend to more advanced statistical methods
• difficult to follow - complex array formulas, nested if statements, formula in name definitions
• struggles with data changing size
• spreadsheets create dimension problems - restricted to 2 dimensions (how do you model a process that varies by class, underwriting and calendar year)

Path Towards Adopting Best Practices⚑

1. Engage with IT

   • Don't let IT own the conversation - just because you are talking in their language doesn't give them a right of refusal.
   • 'Spec it up and IT will implement it' is not often the solution
   • Explain what you need (simple but will develop over time):
     • access to software - almost everything is free (but you should try to contribute)
     • figure out how analysts can experiment with new tools and approaches in work - staff shouldn't have to figure this out in their own time or on their own machines first
     • websites
     • hardware (down the line)
     • try to find IT staff interested (they should be for their own development)

2. Training:

   • Add to objectives
   • loads of stuff for free
   • Lots of online courses use Freemium model (e.g. Datacamp, etc) - provide subscriptions
   • Go on training courses
   • In house training / lunch-and-learns (give people time to run these)
   • Aim for a mix of skills in the team
     • actuarial teams tend to be full of actuaries
     • analysts in other industries might mix data architects, data scientists, PhDs alongside domain experts
   • Who is supporting teams in projects
     • consider external consultants on a retainer basis if expertise not available in house

3. Start simple

   • Give people time / avoid the temptation to rush for results
   • Show off progress to others in the team
   • Examples:
     • Try a new reserving method for a class of business
     • Validation of capital model output
     • An ETL process
     • A dashboard
     • NLP for claims
     • Clustering on a class of business

4. Get Version Control into projects as soon as possible

Get some tools (luckily)