In the pricing engine market, most actuaries and pricing analysts face an impossible choice: use no-code flow diagrams that become unmanageable at scale, or write custom Python code that forces you to sacrifice either real-time performance or batch processing efficiency.

We knew that entering this crowded market required genuine differentiation. Over the last couple of months, we've built deployment functionality that solves these fundamental trade-offs.

The Last Mover Advantage

Being the last mover gave us a unique perspective. For years, the industry has shifted from coding solutions towards no-code solutions like flow diagrams. Whilst this works beautifully for managing 10 models, it becomes an absolute nightmare to source control and maintain when you're working with 65 models.

Meanwhile, Large Language Models have proven extraordinarily effective in coding environments. As early adopters of Cursor, we've experienced first-hand how AI excels at fixing bugs, making code changes, and explaining code. It's transforming how software is developed globally, making code-based approaches more accessible than ever before.

Beyond Open Source: Solving the Speed-Flexibility Paradox

You might ask: why not simply use open source alongside an IDE like Cursor? The critical limitation lies in languages like Python, where you must choose between scalar operations or vector operations. Scalar operations are faster for single real-time predictions; vector operations excel with larger batches.

For instance, a fast single prediction in Python using a scalar operation takes approximately 1ms. Executing these in a loop means you can only process 1,000 predictions per second, or 60,000 per minute. Whilst this can be done in a distributed fashion across multiple machines, that approach comes with significant infrastructure costs and complexity. With vectorisation, however, you could execute a million predictions in under 40 seconds—a 25× speed increase, with no GPUs needed.

Our Solution: Excel-Inspired Syntax with Intelligent Trade-offs

To overcome this limitation, we created our own syntax inspired by Excel, a language actuaries and pricing analysts already know intimately. This familiar foundation makes the learning curve minimal whilst giving us complete control over performance optimisation. The trade-off is introducing approximately 30ms of computational overhead per API call. An API call that would typically take 100ms now takes 130ms.

But here's what that 30ms buys you: the ability to simulate millions of quotes in a matter of minutes. Where a simple million-record scenario might take 40 seconds, more complex multi-variable simulations can process several million records in under 10 minutes, all on a single machine.

This has a transformative effect when building new insurance rates. Instead of sampling 10% of your portfolio data due to memory constraints, you can analyse the full dataset and catch edge cases that would otherwise slip through. The result: faster iteration cycles and greater confidence in your pricing decisions.

The simplicity of our syntax means LLMs can easily debug and improve your code, accelerating iteration dramatically. Due to this simplicity, a large complex rating structure can be represented in less than 120 lines of code. LLMs also generate visual diagrams of your code dependencies in real-time, delivering the same visual debugging experience as no-code solutions but with the power and complexity of coding. Our code editor includes built-in source control akin to Git, making it effortless to manage changes across iterations.

Performance That Transforms What's Possible

Our proprietary converter compiles our models into PMML files that are 35× smaller and run 4× faster than LightGBM's PMML output. In our proprietary format, they achieve up to 10× faster execution with models 700× smaller, whilst keeping accuracy very close to LightGBM at a fraction of the size and speed.

Why does this matter? These aren't just impressive numbers—they translate directly to business value. Smaller models mean lower memory requirements, reduced cloud computing costs, and the ability to deploy more sophisticated models within the same infrastructure budget. Faster execution means you can include more complex models in real-time quote processes without breaching the critical 2-second threshold required by price comparison websites.

We've spoken to many users working with older platforms who complain about having to strip down their models (removing interaction terms, reducing tree depths, eliminating variables) just to meet production performance requirements. They know they're not delivering the best possible price, but their tools force the compromise. In a marketplace where you're essentially selling a commodity, that disadvantage compounds quickly.

These same users can only backtest on 5-10% of their data due to memory and compute limitations. That means potentially missing important patterns in low-frequency events or specific customer segments.

AI That Actually Helps

We didn't just want to slap an LLM chatbot onto our product and call it AI (something all too common in the market today). We invested considerable time thinking about how AI can genuinely improve the lives of actuaries and pricing analysts working with these tools.

However, we firmly believe that in pricing, the human in the loop remains essential. Machines cannot be held accountable, and in a regulated industry where pricing decisions carry legal, financial, and reputational consequences, human oversight and approval are non-negotiable. Our approach augments human expertise rather than attempting to replace it.

Here's a concrete example: when you reference a variable that doesn't exist, our AI doesn't just throw an error. It examines your entire codebase, identifies variables with similar names or purposes, and suggests what you likely meant based on naming patterns, data types, and usage context. If you're calculating total_premium but accidentally typed totalpremium, it instantly recognises the intent and offers to fix it.

The AI excels at detecting downstream variable dependencies—something classic debuggers can only do one level ahead. Need to know every calculation that would be affected by changing how you calculate driver age? The AI traces the entire dependency tree instantly.

We've constrained the AI to focus exclusively on business logic (the pricing rules, calculations, and model implementations). Infrastructure and security are handled by our platform, not the AI. This ensures users can't inadvertently introduce security vulnerabilities or break critical infrastructure through AI-suggested changes. You maintain full control over your pricing logic whilst the AI handles the tedious parts of writing and debugging that logic, safe in the knowledge that the underlying system remains secure and stable.

Unprecedented Performance

Our proprietary converter compiles models to PMML files that are 35× smaller and execute 4× faster than standard implementations. In our proprietary format, performance jumps to 10× faster with models 700× smaller.

Why does this matter? These aren't just impressive numbers—they translate directly to business value. Smaller models mean lower memory requirements, reduced cloud computing costs, and the ability to deploy more sophisticated models within the same infrastructure budget. Faster execution means you can include more complex models in real-time quote processes without breaching the critical 2-second threshold required by price comparison websites.

We've spoken to many users working with legacy tools who complain about having to strip down their models—removing interaction terms, reducing tree depths, eliminating variables—just to meet production performance requirements. They know they're not delivering the best possible price, but the tools force the compromise. In a marketplace where you're essentially selling a commodity, that disadvantage compounds quickly.

These same users can only backtest on 5-10% of their data due to the memory and compute constraints of yesterday's tools. That means potentially missing important patterns in low-frequency events or specific customer segments.

Flexibility and Compatibility

Our Excel-like syntax supports operations such as VLOOKUPs, nested conditionals, and complex lookups, giving you the flexibility to implement sophisticated pricing logic without wrestling with programming paradigms.

We also ensure backwards compatibility for models fitted on legacy systems. Migrating from another platform doesn't mean rebuilding everything from scratch. Once you import your CSV-based rating tables or model files, they're automatically compiled into our optimised format—immediately gaining the performance benefits without any rework. This dramatically reduces migration risk and accelerates time-to-value.

Our entire deployment solution can be accessed via a REST API, which makes it easy to integrate into your current policy administration system. Whether you're using modern cloud-based platforms or legacy on-premise systems, the API provides a clean, standardised interface for real-time rating and batch processing.

Our deployment was built to integrate seamlessly with our simulation feature on the platform, creating a continuous workflow from model development through testing to production deployment.

Built for Pricing Developers

We designed our deployment functionality with pricing developer happiness as the North Star. We built our solution with LLMs deeply embedded to assist our users, and we optimised our process to deliver the best of both worlds: the speed and flexibility of code with the accessibility and visual clarity of no-code platforms.

The result is a platform where you can build better models, test them more thoroughly, deploy them faster, and maintain them more easily—without the compromises that have constrained the industry for years.

Ready to see how it works? Request a demo to see our deployment functionality in action, or explore our documentation to learn how we can transform your pricing workflow.