Can a model spot an AI-written review?
A detector that catches 86% of AI reviews while wrongly flagging 1% of genuine ones, the blind spot that makes older detectors useless, and a live demo you can try.
Read the case studyIs renting really “dead money”? A decision model tested against 21 years of history
A month-by-month simulation of a six-figure decision, a 2005–2026 backtest, sensitivity to rates and prices, and a calculator anyone can use.
Read the case studyIs AI taking jobs? Testing the claim on 722 occupations
Linking AI usage data to employment, controlling for confounders, and a placebo test that shows the headline “effect” was there before ChatGPT.
Read the case studyYour A/B test “winner” may be noise: what 100,000 simulated tests show
A Monte Carlo study of the most common A/B testing mistake, and a sequential stopping rule that lets you look every day without fooling yourself.
Read the case studyWhat these projects show we can do for you
The same methods apply to your data, whether you sell online, lend money, run experiments or plan a workforce.
Machine learning and NLP
Classifiers that hold up on data they have never seen, explained in plain English and deployed where they are needed.
See it in action →Causal inference
Separating cause from coincidence with controls, fixed effects and placebo tests before anyone acts on a correlation.
See it in action →Simulation and Monte Carlo
Thousands of simulated scenarios to show how often a process fails and what fixes it.
See it in action →Decision and financial models
Month-by-month models of big decisions, with sensitivity analysis and a historical backtest.
See it in action →Experimentation
A/B test design, sample sizes and stopping rules that keep false winners at the rate you signed up for.
See it in action →Tools people can use
Calculators and lookup tools that run in the browser, checked against the analysis code to the decimal.
See it in action →How a project works
Fixed scope, a written plan before any modelling, and results you can audit.
Frame the question
A 30-minute call to pin down the decision you need to make and what a useful answer looks like.
Audit the data
What you have, what is missing and what public data can fill the gaps, with a fixed quote.
Model and stress-test
Methods chosen for your question, then tested on data the model has not seen, with the limits stated.
Deliver and hand over
A plain-English report, the code, and a tool or dashboard if you need one, with a walkthrough for your team.
These case studies are independent projects built on public data to show our methods. They are not client engagements, and no client data was used.