Skip to content
Building Bulb

How we do data science at Bulb

Ashwin, a data scientist, working at his desk

The data science team at Bulb is growing. We wanted to share how we work and what we're learning along the way with the wider data science community.

Our mission at Bulb is to reduce our members' bills and their carbon emissions. Our 18-strong data science team is still focussed on that mission while we work remotely during the Covid-19 pandemic. As we all pause from business-as-usual, we wanted to take some time to reflect on our normal working practice and share the things that have served our growing data science team well.

1. We welcome generalists

Today, most of the data science team are generalists. This means we look after the whole data pipeline, from extracting raw information from databases to explaining the performance of machine learning models we build to the wider company.

We believe this makes us a stronger team because it gives us full autonomy over what we deliver. By owning the whole data pipeline, we can add new features to our feature store where necessary, or we can deploy a new machine learning model within a sprint. We can do these things without depending on another team for our success.

We've also found that this generalist approach has some drawbacks. For example, without data engineers there is no sole owner of our data warehouse pipeline. This means that knowledge of our data warehouse is distributed across teams. So far, we've been happy to make this trade-off as we move quickly and adopt a minimum viable product (MVP) mentality. We want to improve on these areas as we grow and we expect there to be more room for specialism. We’re now starting to hire for specialised roles that cover both analytics and data engineering. We think introducing more focus will help us to achieve more from those areas.

2. We focus on results

Owning the whole data pipeline means it's easy for us to see which work delivers the biggest business impact.

That means we use high-tech tools like BERT, a deep-learning based language encoder, to sort through and organise the thousands of emails we receive from our members every week. We also use more hands-on methods like linear regression to go back and explore the variables which add up to a member wanting to leave Bulb, so we can work out how to serve our members better in the future.

Because of this, most of our ideas come from our own team. That includes innovations like using computer vision to read energy meter photos from members, or optimising the route our engineers take to install smart meters. As data scientists, we're able to imagine new ways to move Bulb's mission forward, and build a quick MVP model to validate our assumptions.

For example, recently we suggested to our Member Experience team that we could build a tool to automatically respond to simpler queries from our members, helping to provide service out of hours. Before jumping straight into a complex solution, we built a proof-of-concept model based on simple business rules to deliver impact straight away. The tool went live and helped us to speedily serve more members, more of the time. We're now working to improve the model to make it more helpful for members and the team.

3. We democratise data science

As a company, we perform best when everyone is clued up on how to use data and understands when data can be misleading. We don't want data scientists to be the only keyholders for data at Bulb.

That's why we run data training for the whole company. This starts with the basics of SQL and progresses onto statistical topics, such as Simpson's paradox.

The data science team giving a talk at Bulb HQ
Data training at Bulb HQ

This makes it possible for everyone to contribute ideas that help the company on its mission. And, crucially, to prioritise work according to the results we might expect to achieve.

At the same time, we've chosen to build a Bayesian framework for experimentation that helps us communicate the results of experiments in plain English.

We also democratise data science within the data team itself. We hold weekly meetings and regular Journal Clubs to ensure we're all up to date with the latest on what's happening in the industry, from deep learning to data protection and ethics.

We're learning all the time

This is a little bit about how we do data science at Bulb today. We're constantly challenging our assumptions about what works for us and what we need to throw in the bin.

There are lots of opportunities to learn, progress, challenge and be challenged as part of the data science team at Bulb. Why not join us?


We're still hiring while our team is working remotely. You can check out our job openings for data scientists on our careers page, or reach out to us directly at careers@bulb.co.uk. We'd love to hear from you.

Bulb founder, Hayden is giving regular updates about how we're staying focussed on our mission during the Covid-19 pandemic.