Simon
Thornewill von Essen

At Deliveroo I've worked across three main areas: Fraud Detection/Trust, Logistics, and Restaurant Operations.

In fraud, my work has centred on protecting the business from financial loss while keeping the customer experience intact, using causal inference and Bayesian modelling to cut through noisy signals and make high-stakes policy decisions on limited data.

On the logistics side, I've designed and analysed experiments on dispatcher algorithm features, contributing to improvements in delivery time and restaurant wait-time estimation.

In the Restaurant Operations team, I helped to make sure that our orders were delivered "perfectly every time".

I've also worked across the full project lifecycle, from framing ambiguous problems with senior stakeholders to mentoring junior colleagues and delivering internal training to over 150 employees.

Highlights

• Blocked a £5M/year cost risk by designing an experiment that exposed a 3DS fraud algorithm's 100%+ chargeback increase, far outweighing its 0.5% order volume gain.
• Reduced missing item order inaccuracy by ~15% (projected saving of ~£500K/year) through a banner experiment that surfaced customer misunderstanding of who bears the cost of inaccuracy claims.
• Delivered ~£300K in short-term savings by revamping the top-customer fraud exemption policy, optimising across multiple outcomes during a period of rapidly rising costs.
• I quantified and reduced uncertainty for item fraud from 5–70% down to 15–20% using Bayesian modelling on camera-labelled data, enabling more confident project prioritisation.
• Reduced average delivery times by ~3 seconds per order by designing experiments to validate dispatcher algorithm features enabling dynamic route optimisation.
• Reduced large-order misclassification by 3.5% by validating an LLM-based classification model and running the production readiness experiment.

At Goodgame Studios, a mobile gaming company, I worked embedded in the marketing team, helping turn it into a centre of excellence within the Stillfront group. My work spanned marketing analytics, A/B testing infrastructure, data quality monitoring, and stakeholder-facing reporting.

I built and maintained the dashboards and tooling that gave the team visibility into campaign performance and testing outcomes, and helped scale the team's influence to cover marketing for multiple group companies.

I also maintained a cohort ROI forecasting model used to guide marketing spend decisions.

Highlights

• Supported the marketing team's rise to a centre of excellence within the Stillfront group by building and maintaining 10 Tableau dashboards covering A/B testing and multi-channel marketing performance.
• Saved ~1.5 hours of manual investigation per week and caught ~2 data quality issues weekly by building a Python automation tool that ran nightly checks and emailed alerts on dashboard divergences.
• Enabled a smooth Flash-to-HTML5 engine rollout by aggregating and matching cross-platform metrics in an A/B testing dashboard, surfacing multiple bugs before full release.
• Raised stakeholder awareness of A/B testing pitfalls by presenting on Bayesian statistics at a local data science meetup.

A pair of project-heavy nanodegrees covering the full data science workflow, from data wrangling and visualisation through to machine learning and big data.

Projects spanned supervised learning, recommendation systems, neural networks, and large-scale feature engineering in Spark.

Projects

• Image Classification Neural Network — developed and evaluated neural networks for image classification in PyTorch.
• Recommendation System for IBM Article Platform — rank-based and collaborative filtering methods with evaluation.
• Churn Prediction Using Spark — big data feature engineering and model evaluation at scale.
• Predicting Burn Areas for Forest Fires — regression with feature selection and hyperparameter tuning.

An integrated master's in Chemistry, combining rigorous quantitative training, statistics, thermodynamics, computational modelling, with hands-on laboratory and research experience.

The degree built strong foundations in experimental design, data analysis, and scientific communication that have transferred directly into data science work.

Projects

• Final-Year Project: Sustainable Activation of Boronic Esters; literature review, experimental planning, and analysis.
• Systematic Review: Conversion of Cellulose to Biofuels and Chemical Feedstocks; evidence synthesis and scientific writing.
• Group Project: Computational Design of Novel Statins; molecular modelling using Autodock Vina.