6.0. Production
Productionising code is a large step up from running scripts locally.
In local development, analysts typically run code manually, inspect outputs, and rerun if things fail. In production, code must run reliably, repeatedly, and automatically - often without human intervention. This requires:
- Strong programming skills to write robust, defensive code that handles edge cases gracefully.
- Familiarity with cloud platforms (e.g. Azure, AWS) and the range of services they provide.
- An understanding of how different systems interact (databases, APIs, pipelines).
- Experience at the coalface - knowing the operational challenges that arise when code is no longer just a one-off analysis, but part of a business-critical workflow.
Why Productionise Code?
Even if running a script seems trivial, moving to production offers huge benefits:
- Stay on top of complexity – production code can handle large, messy data and complex workflows without constant manual fixes.
- Reduce context switching – automated pipelines run without you having to “babysit” them.
- Avoid key-person dependency – processes don’t rely on one analyst manually running code; anyone can trigger or monitor them.
- Enable advanced workflows – APIs, scheduled jobs, dashboards, and monitoring are only possible in production environments.
- Deploy models and prices – core requirement in pricing; models must move from analysis to live rating engines where they influence quotes.
Example in pricing:
Instead of an analyst manually refreshing a claims frequency model in Excel every quarter, a productionised process could automatically pull claims data, retrain the model, validate its performance, and flag results for review before deployment.
Limitations and Roadblocks
Productionising code is not without challenges. Common issues include:
- Limited choice of tools – many insurers lock down environments, restricting libraries or software.
- Permissions – getting access to databases, APIs, or deployment environments can be bureaucratic and slow.
- Time constraints – analysts often have little time to set up production pipelines when project deadlines loom.
- Debugging complexity – failures in cloud environments are harder to trace than local errors. Logs, monitoring, and alerts are essential.
- Skills gap – most pricing analysts are comfortable in Excel or Python notebooks, but fewer have experience with DevOps, CI/CD, or cloud infrastructure.
The Shift in Mindset
Production is not just about writing working code - it’s about writing code that:
- Can run unattended.
- Can recover from failure.
- Can be understood, maintained, and extended by others.
- Integrates smoothly with wider IT systems.
This shift is often the biggest hurdle for pricing teams: moving from an analytical mindset ("does this work for my dataset?") to a production mindset ("will this work for all datasets, at scale, every time, and can someone else maintain it?").