Coppelia · Custom parts for data science

The problem

The pieces are in place, but it isn't working

You have a data science team, the data and the tools. Yet the project isn't producing the promised result.

A part is missing, or doesn't fit the problem.

The solution

A bespoke component built by specialists to fit your problem

The textbooks, the blog posts, and the LLMs trained on them, all solve vanilla problems. Your problem is unique. You need a custom part, whether that's an unusual model, a custom-made algorithm or a predictor that works on problematic data.

How it works

From finding the gap to fitting the part

Step 1
Find the gap

We immerse ourselves in your business, define the problem and identify the missing part.
Step 2
Specify the fit

We agree on how the part will connect to your systems, and specify the inputs and outputs so it slots right into place.
Step 3
Build it

AI has made building components fast and affordable. You pay for the thinking, the expertise and the science, not for hours of coding.
Step 4
Deliver it

We make installing the part hassle-free by delivering containers, code packages or self-installers.
Step 5
Bed it in

We support and train your in-house data scientists so the part keeps working after we've gone.

Parts we've built

Some custom parts from previous projects

Part 01A vehicle insurer

Driving style detector

The problem: Judging driving from the sensors in drivers' own phones, where the signals are noisy and the events that matter are rare.
The part: A model that combines accelerometer, gyroscope and GPS readings to detect driving events and score each journey.

Part 02A global media agency

Audience sizing engine

The problem: How do you work out the size of a campaign audience from minimal information without bias?
The part: A module that uses symbolic logic, Bayesian statistics and information theory to produce the best available estimate.

Part 03A credit comparison platform

Expert-trained ranking model

The problem: Card rankings traded eligibility against APR and offers using weights set by trial and error, undocumented and untested.
The part: A choice model in an unusual setting: experts choose between card profiles, and the model recovers the weights they apply to drive the ranking.

Part 04A fashion research agency

Fashion trend forecaster

The problem: Predicting how fashion trends will rise and fall in retail, from short, seasonal histories and catwalk signals that land late and unevenly.
The part: A forecasting package and app that models the delayed catwalk effect and picks and back-tests the best model for each trend overnight.

Part 05An anti-extremism research charity

Radicalisation pathway mapper

The problem: Finding the accounts that draw people from fringe conspiracy groups into extremist networks, hidden in huge volumes of social-media data.
The part: A graph database and network model that classifies each account by its role, so researchers can query for pivotal accounts and pathways.

Part 06An online fashion marketplace

Bayesian A/B testing module

The problem: Every product team ran A/B tests its own way, so results were hard to compare or trust.
The part: A Python package, notebook templates and a shared knowledge base so every test is designed, monitored and read the same Bayesian way.

Part 07An NHS research team

Hospital outbreak simulator

The problem: Working out how testing and triage policies would affect the spread of covid on wards, before trying them on patients.
The part: An agent-based simulation of a hospital, from admission to ward, that compares testing strategies at different levels of community infection.

Part 08A health-technology company

Biomarker change detector

The problem: Could noisy daily biomarker readings reliably reveal a change in someone's physiology? The answer was needed before building a device.
The part: A simulator that generates realistic readings and stress-tests change-detection algorithms, from simple thresholds to Bayesian changepoint models.

Part 09A broadcaster

Viewer impression map

The problem: Understanding what different viewer segments want from TV, in their own words.
The part: A model that maps open-ended viewer responses into an interactive space showing which impressions cluster together and matter most to each segment.

A fair question

Can't we just use AI?

You can try, and there is no denying that AI can innovate, but it's trained on textbooks, blog posts, and repositories that solve standard problems.

The growing consensus is that effective AI is a partnership and a dialogue between experts and agents – which is exactly how we build our custom components.

Why us

Hard problems are what we've always done

Coppelia was founded in 2014 by Simon Raper, an RSS accredited statistician with more than twenty-five years in data science, statistics and machine learning. We have spent that time on the difficult parts of projects: the forecasts, simulations, choice models and language systems that don't come off the shelf.

Our clients range from Google, The Economist and ITV to Wikipedia, Sainsbury's and Hope not Hate. We also train and mentor in-house teams, which is why it matters to us that what we build can be understood and kept running.

More about the team

2014: Coppelia founded
25+: years in data science, statistics and machine learning
55: clients across media, retail, finance, research and charities
RSS: accredited statistician, and a regular writer for Significance

Clients

Who we've worked with

What clients have said

In their words

Working with Simon helped us look at our data in a new way. He quickly grasped our issue and found an elegant methodology and a practical implementation that we can keep building on. Simon also explained his approach and tooling well. It built on tested technologies and each step was clearly explained which was essential in allowing us to implement the tooling and expand on it ourselves. It will now form a central part in our ongoing work.
Patrik Hermansson, Senior Researcher, Hope Not Hate
Simon has the skills and the expertise to get straight to the heart of the problem and his solutions are always original and exciting. He is the best person I have worked with in his field and I would not hesitate to recommend Coppelia to anyone needing analytics or just some clear thinking about data and what to do with it.
Juliet Bauer, CCO, DC Thomson Family History
Our work with Simon has accelerated our business into a space I couldn’t imagine a year ago. And working with Simon is such a pleasure. Not only is he able to bring structure and meaning to the most complex data he does so from a deeply commercial and pragmatic place. He dials down ‘the esoteric’ because he knows that humans have to use his models and tools. And perhaps most importantly for me he has a great sense of humour.
Dominic Payling, Planning Director, MSL Group
Simon was excellent throughout: he grasped the brief immediately, quickly converted hypothetical discussions into an achievable plan of work. Simon far exceeded what we hoped might be possible within the time and budget available. All of this was delivered with great warmth and with ideas expressed in terms that everyone could grasp. I couldn’t recommend Simon any more highly.
Andy Hamflett, Co-Founder and Director, AAM Associates
Working with Simon has been a game-changer for WGSN’s data capabilities, taking our data science journey to the next level. Simon’s technical expertise combined with his ability to understand business goals have been crucial in building a brilliant solution, which is not only future-proof but also aligned with our company values and mission. His work has always been of the highest standard, impressing both technical and non-technical stakeholders. All in all, I couldn’t recommend him enough and would love to continue to work with him in the future.
Sara Gaspar, Head of Data Analysis, WGSN
Simon has worked with us for many years now. Due to his great breadth of knowledge and experience he’s been able to advise us on all our machine learning projects, suggest alternative directions to take, and discuss the latest trends in AI with us. He is a great help and full of ideas.
Ben Stockermans, Senior Developer, Beauhurst
To put it simply, Simon has been the best thing that has happened to me since I started working at TotallyMoney five years ago. He has immensely helped me develop both my soft and hard skills, contributing critically to my growth as a data scientist.
Adria Luz, Senior Data Scientist, TotallyMoney
Last but not least, Simon is also a great person and a delight to work with, so if you ever have the chance of working with him – take it!
Adria Luz, Senior Data Scientist, TotallyMoney
Simon has played a key part helping me expand the role and acceptance of data science within the company, thanks to his formidable ability to understand business challenges and translate them into statistical problems that can be solved mathematically.
Adria Luz, Senior Data Scientist, TotallyMoney
Simon has been preparing excellent training sessions for our team for a number of years now. Despite the varied experience within the team of statistical models and machine learning, Simon leads insightful sessions for novices and experts alike.
Alex Henman, Head of Product Engineering, Beauhurst
I am fortunate to have been trained and mentored by Simon at both Annalect and ITV over the past few years. Simon has not only taught me the theoretical side of statistics and data science but also the practical application; such as writing up a project briefs, through to what kind of statistical model I could use. I have found Simon’s training invaluable, and frequently look back at my notes or previous jupyter notebooks from Simons sessions when kicking off a project.
Annie Baldwin, Senior Data Scientist, ITV
From the first time I met Simon, I was impressed and fascinated about the passion he conveyed and the way he was able to frame a story to make it relevant for the audience.
Clemens Zaucher, Senior Data Scientist, Sainsburys
In the regular conversations I had with Simon, I was able to discuss things I wanted to learn more about in a more technical sense, like for example data visualisation and D3.js, but also more soft skills like presenting results to a non-technical audience, or how to approach certain meetings. One of the most valuable ways he helped me in my day-to-day job was discussing a project, the approach I was taking, and the results of analyses. To have an experienced person looking over your shoulders, providing pointers, and highlighting common pitfalls is invaluable, especially when you are starting out.
Clemens Zaucher, Senior Data Scientist, Sainsburys
We had training in groups on topics like Spark and MapReduce, web scraping, or simulation which helped not only to know how to use an API of a certain framework, but also to understand the concepts behind them. I cannot recommend Simon enough, his mentorship has helped me so much and continues to do so to this day.
Clemens Zaucher, Senior Data Scientist, Sainsburys
Simon’s mentoring was very useful in helping me to expand the range of statistical techniques I could apply to my day to day work, and helped to give me confidence that the analysis and results I’m producing are statistically sound.
Danny Flintoft, Data Analyst, Unidays
We have worked with Coppelia since the beginning. They have been instrumental in helping us move our analytics proposition forward into a contemporary and advanced space. With Coppelia’s help we have upskilled all our analysts and successfully created a more innovative and creative analyst community
Jenny Bullis, Managing Director, Annalect Solutions UK
Simon is a very skilled teacher who would explain in detail the things that I had not yet understood but who was also able to recognize when my existing knowledge was already sufficient enough for him to skip ahead. Without his guidance, some of the projects I worked on during my time with him would not have been as successful as they did.
Laurens Geffert, Senior Data Scientist, Sainsburys
“We don’t always have a smoking gun. Instead, we are like lawyers, presenting a case for or against a claim.” This piece of advice has changed the way I work. It has also helped with a bit of imposter syndrome where I have often assumed there was some magic way of answering a question, that I wasn’t smart enough to know about. When in fact there probably wasn’t, and there are often some more simple, sensible analysis to be done instead, that can be just as powerful.
Molly Devine, Data Analyst, The Economist
It was very useful to get advice and brainstorm how to best approach various tasks. Not only was this helpful from a statistical and methodology perspective, but also a business perspective and I learned many useful tips on how to respond to stakeholders.
Molly Devine, Data Analyst, The Economist
Since the mentoring, I’ve become a lot more comfortable in python, data presentation and visualisation and statistical methodologies/approaches to data. I’ve also become more comfortable in discussing statistical analysis with stakeholders and understanding what is possible and what is not.
Molly Devine, Data Analyst, The Economist
Simon has a tremendous breadth of knowledge, which allows solutions to be found for a wide range of problems. The training and mentoring sessions have really allowed me to extend my skillset which has enabled me to analyse and interpret data using a much wider range of approaches than I previously could. The training sessions provided me with a greater appreciation of testing methodologies and alerted me to some common misconceptions and misinterpretations.
Neil Burden, Senior Web Analyst, Unidays
The one-on-one mentoring sessions were invaluable in providing bespoke solutions to genuine work questions, and to provide approaches that I could follow for future projects. The sessions have enabled me to utilise Jupyter Notebooks to carry out advanced analysis, whilst improving my SQL and Python skills. It has also opened upon a whole range of different data visualisations to employ to better demonstrate the outcomes of my analysis.
Neil Burden, Senior Web Analyst, Unidays

01 / 22

Melt

Coppelia is part of Melt

Melt is a group of independent data science professionals who work together on larger projects. When a part needs more hands or another specialism, we draw on the collective.