How we do data science at Bolt
The structure of the Data Science team at Bolt
The Data Science team at Bolt consists of data scientists and machine learning engineers responsible for advanced analytics methodologies, production machine learning models, and the experimentation engine powering Bolt’s decision-making.
At Bolt, data scientists own their models and optimisation algorithms.
This is no easy feat and requires some software engineering skills. Hence, data scientists are a part of the Engineering division, and we’re also looking for people who can write the necessary code and own and deliver the projects.
This approach may differ from other large tech companies where data scientists focus primarily on analytics and experimentation-related work. At Bolt, this role is carried out by product data analysts who keep a finger on the pulse of Bolt’s global expansion and operations.
Work and personal growth
Our week-to-week work happens in cross-functional teams with software engineers, product managers and data analysts, optimising for common KPIs. However, the nature of data science work is quite iterative and experimental and different from how regular software teams operate.
The process usually starts from ideation with product management, exploratory data analysis, and first proofs of concepts on historical data. After this, the work continues by training, deploying, and monitoring the models globally across all the markets.
Due to the iterative and experimental nature of the work, we encourage our colleagues to never stop learning:
- We organise regular discussion sessions about the newest projects, papers, or methodologies;
- We promote feedback and reviews (design docs, pull requests, workshops) between people working in different product teams;
- We host a learning club; in the past, sessions were about operations research, reinforcement learning, causal inference, and deep learning.
Dedicated Model Lifecycle engineers have built a platform for training and deploying ML models. Thanks to this, it’s possible to take a well-structured Jupyter notebook’s code into production in a couple of days (removing Pandas from the inference part first, obviously).
Products
Bolt operates globally with more than 100M customers in over 500 cities across Europe and Africa and with over 3M drivers and couriers. We have evolved from a single business into a global mobility provider in:
- Rides: ride-hailing service to get from A to B;
- Food: food delivery service;
- Market: fast and convenient grocery delivery from Bolt’s own stores;
- Micromobility: e-scooter and e-bike service to alleviate the “last mile” problem;
- Drive: short-term or long-term on-demand car-sharing service;
- Platform: support for all the business lines by building internal solutions and managing APIs to external solutions, e.g. maps, verification platform, fraud detection, and customer support tooling.
Teams
Marketplace (Rides)
The main objective of the team is optimising ride-hailing marketplace efficiency. Our flagship product is dynamic pricing which balances demand and supply. This requires working on problems such as price/ETA sensitivity estimation, conversion modelling, demand/supply forecasting, etc.
Another focus area of our team is order dispatching. These algorithms maximise order completion rates over different rider and driver matches. The main complexity here is to solve that problem at an operational scale with constraints regarding the number of available drivers, riders’ willingness to wait, expected driver earnings, and other factors.
We also work on research topics such as agent-based marketplace simulation, structural economic modelling, and causal inference that help us test, validate, and explore new optimisation areas.
Campaigns (Rides)
Ride-hailing campaigns boil down to rider discounts and driver bonuses — a widely used tool in the industry to boost market growth.
Our Data Scientists work on optimising global campaign budgets at micro (user) and macro (market) levels.
Micro-level optimisation means working on LTV, retention, churn modelling, and developing novel campaign targeting methods. Long-term market modelling for investment strategy determines the macro level spend optimisation. Here, we deal a lot with multivariate time series forecasting of various marketplace and financial indicators.
Routing (Platform)
We’re optimising for the accuracy of duration and route predictions powering the dispatching engine used in ride-hailing and food delivery. To do this, we use a combination of a traditional graph-based router and a machine learning layer.
We work on various projects, including travel time prediction, traffic modelling, inferring missing map elements, and finding route-optimal pick-up/drop-off coordinate perturbations.
Places (Platform)
Our main goal is to improve the experience of our ride-hailing users during the pick-up/drop-off selection.
To achieve this, we use an NLP model for correcting user queries and a Learning to rank (LTR) model for providing the most relevant results given the context. We also have an additional model for detecting potential user frustration in search.
We use unsupervised learning to see missing or erroneous places of interest, ensuring that our underlying data has good coverage and is up-to-date.
Map Improvements (Platform)
This team is responsible for researching ways of improving our internal maps.
Currently, we’re focused on inferring missing map elements, such as missing roads, one-ways, turn restrictions, and others, by using traffic data and street imagery. These improvements help us provide better routing and dispatching.
Delivery
The food delivery marketplace is even more complex than ride-hailing due to having an additional third party — the food providers.
Therefore, the dedicated Delivery team owns a wide array of projects that use some elements from ride-hailing, such as dynamic pricing, dispatch optimisation, travel and cooking time prediction, and user food recommendations.
Identity & Trust (Platform)
The main product is the Verification Platform, which we use to verify the identity of Bolt users and driver partners by checking the data from the documents they provide.
Afterwards, we match the face of the person to the document and make sure that both the face and the document are first-hand captures belonging to the real person.
We use vendors and in-house text extraction and liveness detection systems to provide state-of-the-art performance and scalability.
Tools & Workflows (Platform)
The Tools & Workflows (T&W) team is responsible for optimising different workflows inside Bolt.
Whether helping our Customer Support team handle incoming user requests more efficiently or enabling our mapping specialists to improve our routes and maps, the T&W team has the right tools for the job.
The ML projects inside the T&W team range from customer ticket classification and automation to workload forecasting and service level optimisation.
Rentals
As Bolt is a micromobility provider offering e-scooter and e-bike rental across Europe, this opens up a variety of interesting technical challenges.
We use levers such as pricing or deployment allocation to optimise fleet utilisation. Machine learning-assisted checks are in place to detect when multiple users are riding the same scooter.
At the end of the ride, we ask users to take a picture of a parked scooter. To ensure the scooter is parked correctly, the Rentals team has developed an AI model that analyses these images.
Due to these different problems, the diverse Rentals Data Science team consists of IoT specialists, mathematical optimisation gurus, and deep learning enthusiasts.
Fraud (Platform)
Capturing credit card fraud is an uphill battle of fraudsters constantly adapting and using stolen credit cards to find ways to get free rides.
However, our fraud rules engine and machine learning models can capture such instances quickly, keeping costs down for honest users.
Tech stack
A data scientist at Bolt will quickly get used to dockerising their Python code, orchestrating its execution via Airflow and Jenkins, managing model lineage through Amazon SageMaker and using a mix of Great Expectations, Grafana, and our in-house A/B test analysis engine to make sense of the model results. Most models are deployed as Python microservices, avoiding other teams’ integration friction.
Bolt’s data scientists and machine learning engineers are responsible for all stages of an applied project. They start with mapping a product problem to a technical solution, gathering the source data, developing a prototype (generally in a notebook-like environment), refactoring it and finally deploying it to production.
Most data science projects use machine learning methodologies, such as gradient boosted tree ensembles, time series forecasting, deep learning or reinforcement learning. However, the problems sometimes relate to causal inference from observational data or constrained optimisation.
Therefore our team consists of various profiles, such as mathematicians, computer scientists, AI specialists, physicists, signal processing engineers, and even economists.
Testimonials
Joonas, Machine Learning Engineer, Routing: My favourite thing at Bolt is that you can work on problems having a significant scale impact. The fact that some model endpoints are queried millions of times daily and used for downstream decision processes is an incredible feeling. It means that you may considerably impact the business by coming up with some clever feature engineering approach or finding a new informative feature.
Francesco, Senior Machine Learning Engineer, Identity & Trust: Several aspects of working at Bolt keep me as motivated as I am. The first is the team. I work with talented, genuinely helpful, and incredibly driven people. We set a common goal and work towards it together, leaving no one behind.
The second is the amount of knowledge you can absorb if you’re willing to. Want to dig into MLOps? Reach out to the data engineering team and ask them. Curious about databases and how their performance affects your application? Go for it.
The third is the speed at which we ship software. We strive to reduce at a maximum the feedback loop for a feature. Does it work? Good. Maybe not quite yet? Iterate until it does. Consistency and focus are key here. We keep projects on track to have the most significant impact with the slightest effort and time.
Théophile, Data Scientist, Delivery: At Bolt, I enjoy working with talented people, especially in data science. It amazes me how big of an impact small teams of data scientists and ML engineers have thanks to the scale to which algorithms are deployed. The high quality of the tech backbone allows us, the data scientists, to work serenely with data while focusing on the core data science part of projects. On top of that, the company provides many opportunities to learn many topics and perfect oneself in data science. Working remotely is almost seamless, as most teams work in a hybrid manner.
Elizaveta, Senior Data Scientist, Fraud: One thing that motivates me at Bolt is that data scientists are actively involved in the decision-making process. We don’t wait for someone to tell us what to do.
As data scientists, we continuously experiment with models and approaches, find the best solutions, and propose our ideas to the Product teams. Data scientists are part of the planning process and suggest new directions. Bolt values proactivity, taking control, and making things happen rather than just adjusting to a situation or waiting for something to happen. These principles align with my values and have kept me performing as part of the Data Science team at Bolt for more than four years.
Kiryl, Data Scientist, Marketplace: The thing I enjoy most about Bolt is the variety and complexity of the problems at hand. The scale of these problems is enormous, and even a 1% improvement translates into a vast and meaningful impact.
A home-grown ML platform allows you to abstract away a lot of time-consuming tasks and focus on what’s important. Top it up with a high bar of peers collaborating with you, and you end up in a perfect environment, i.e. one where you can grow as a specialist, pursue what’s meaningful, make an impact and have some fun along the way.
Mikk, Senior Data Scientist, Campaigns: Bolt is a company with an inspiring vision where my work can reach millions of people around the world. There are endless opportunities within the company across multiple verticals — all I need to do is show initiative, and I’m able to contribute. I can choose what I work on and how I work — as long as I deliver value to our customers.
Join Bolt!
Bolt is serving over 100 million customers in 45 countries across Europe and Africa.
You can bet that there are fascinating challenges involved with this kind of growth! If you’d like to join us in building the future of urban transport, visit our careers page.
