Great Rail Journeys case study
AI & DataEngineeringCloud

Great Rail Journeys

AI-Powered Fare Prediction

Machine learning that outperformed manual fare estimation in 15 of 16 categories, proving the commercial case for AI-driven pricing in travel.

15/16
Categories beat manual
ML pipeline
POC to production API
Azure
Cloud-native deployment
Near-zero
Overall mean delta

The Challenge

Great Rail Journeys' customers book up to two years before travel. Rail tickets aren't released for sale until much closer to departure, forcing the business to price packages long before they know their actual costs. Previously this relied entirely on manual estimation. This was time-consuming, prone to missing sudden shifts in inflation, interest rates, and fuel prices, and dependent on key individuals. Accurate estimation is critical: it determines whether customers get a fair, competitive price and whether tours are profitable.

Two-Year Pricing Gap

Customers book years in advance, but rail fares aren't available until close to travel, creating a pricing estimation challenge that directly impacts profitability and competitiveness.

Manual Estimation Risk

Pricing depended on manual processes and key individuals, creating single-point-of-failure risk and inconsistency across the product range.

External Volatility

Inflation, interest rates, fuel prices. These external factors that affect fare prices were difficult to account for systematically using manual approaches.

Our Approach

We partnered with Great Rail Journeys on a proof of concept to determine whether AI/ML could match or improve upon manual fare estimation. We trained multiple models on historical booking data, iterated through data quality challenges in close collaboration with GRJ's data team, and validated predictions against real-world outcomes.

The service was deployed as a cloud-native application on Azure Container Apps with full infrastructure-as-code, creating a production-ready foundation for continued AI adoption.

Technical Solution

ML Ensemble Models

CatBoost, Random Forest, and XGBoost combination producing highest-scoring predictions

Python ML Pipeline

Repeatable pipeline predicting prices across different dates, routes, and product ranges

Azure Container Apps

FastAPI service on Azure with Terraform IaC, VNet security, and HTTPS endpoints

Model Validation

Go-based analysis tool comparing predictions against actual fares using Mean Delta, MAE, and RMSE

Security

Private container registry, IP whitelisting, VNet integration, HTTPS-only access

Data Engineering

Overcame low data volumes, data evolution anomalies, product classification gaps, and recent data challenges

The Impact

15/16
Categories where ML outperformed manual estimation
Model beat human pricing across tens of thousands of bookings
Near-zero
Overall mean delta
Wins and losses cancel almost exactly across the full product range
13.95
MAE on Classic range
Despite being the hardest range to predict due to unseen ticket class differences
Production
Cloud-native API deployed on Azure
Full infrastructure-as-code with Terraform

The model outperformed manual prediction in 15 of 16 categories across tens of thousands of bookings. The overall mean delta was almost zero, meaning wins and losses cancelled each other out almost exactly across the full product range. The Classic product range, notoriously hard to predict due to ticket class differences the model couldn't see, still achieved a MAE of 13.95, significantly better than manual.

The POC proved that AI/ML can not only match but improve upon manual fare estimation, unlocking fair and competitive customer pricing, reduced risk exposure to price spikes, more accurate P&L forecasting on tours, and the ability to redirect manual effort toward higher-value activities like new route development.

Long-Term Value

AI-Driven Pricing

Great Rail Journeys now has proven AI capability that transforms how they approach pricing across their product range.

  • ML pipeline from training data to deployed prediction API
  • Model validated across tens of thousands of bookings
  • Business case proven: AI outperforms manual estimation consistently

Strategic Foundation

The success of the pricing POC opened the door to broader AI adoption across the business.

  • Production-ready Azure infrastructure with Terraform IaC
  • Proposed Tour Design Companion platform for AI-assisted product creation
  • Safeguarded against single-point-of-failure pricing dependency

Ready to prove the value of AI?

We build machine learning pipelines that deliver measurable commercial impact, from proof of concept to production.

Let's talk