Head of Data Science at Paddle in UK

Job description

• Application
• We are looking for a Head of Data Science to build Paddle's data science capability from the ground up. Our Merchant of Record model gives us a data position no PSP or billing provider has: subscription and pricing context alongside granular payment-outcome data, across thousands of software businesses. This role exists to turn that into economic value by putting machine learning and agentic systems into production.
• The mandate is deliberately narrow and deliberately ambitious: data science at Paddle owns automated decisioning inside the product — traditional machine learning and agentic systems alike — not decision-support analytics.
• We have a long list of candidate opportunities than we can fund, spanning payment performance and revenue recovery, sales and marketing operations, growth and monetisation intelligence, and risk, trust and compliance. We have a working hypothesis about which of these pays back first, and we'll share it — but part of the job in your first quarter is to pressure-test it, size the alternatives yourself, and tell us where to start.
• This is a founding role, and for the first few quarters it is a building role more than a managing one. You'll be the only person in the function: doing the analysis, engineering the features, training and evaluating the models or agents, taking the first system live with our engineering teams — and then operating it, answerable for its uptime, its drift and its numbers. Once the first use cases are proving out, you'll hire and lead a hub-and-spoke team of data scientists and machine learning engineers embedded across our highest-value business areas. You'll report into the VP of Data and work in close partnership with Product, Payments, Engineering, Risk and Finance.

Responsibilities

• Prioritise which opportunities have the biggest impact.. Build a value-based use-case backlog across payment performance and revenue recovery, sales and marketing operations, growth and monetisation intelligence, and risk, trust and compliance; size the leading candidates properly; and make the call on sequencing with the relevant Product, Risk, RevOps and Finance stakeholders. This gets refreshed quarterly.
• Personally deliver the first system end to end: the analysis and back-test, the features, the model, policy or agent, the deployment, and the shadow and A/B tests that prove it works. Not a spec handed to someone else to build.
• Operate what you deploy. Own monitoring, retraining, drift response, incident handling and rollback for live decisioning, alongside the engineering teams whose services call it — and set the expectation that the function runs its systems rather than shipping them.
• Work across both traditional ML and agentic systems, and be honest about which a problem actually needs: a propensity or uplift model, a policy of rules, or an agent with tools, evals against golden answer sets and trace-level observability. Several of our strongest candidate use cases point each way.
• Build and lead the team — hire senior data scientists embedded in value areas and machine learning engineers in the hub, and set the professional standards, shared methods and reusable components the function runs on.
• Establish the production stack alongside Data Platform and Engineering: reproducible training data, code and artefacts; a model registry; inference services with real latency, availability and rollback requirements; historically accurate features where decisions need backdated reconstruction; eval harnesses and trace observability for agentic workflows; and monitoring for data quality, drift, model performance and economic outcomes. Start ad hoc where that's sufficient and platformise once the first use cases have shown what's actually needed.
• Own value capture end to end. Shadow-test and A/B test every deployment against the incumbent strategy, translate metric movement into a financial number on a methodology co-signed by Finance, and publish a quarterly report on realised value.
• Set the governance model for automated decisioning — proportionate risk assessment, clear ownership, latency and availability requirements, human escalation and rollback — working with Legal, Privacy, Compliance and Risk on GDPR, EU AI Act and payments obligations, and producing the evidence early enough to shape the design.
• Define the boundaries and the working relationship with Product Science, Analytics Engineering, Data Platform and AI Enablement, so accountability for decision support versus automated decisioning stays unambiguous.
• Deliver cross-functionally: embed in delivery groups with product owners, domain experts and platform engineers rather than handing models over the wall.

Requirements

• Experienced leading data science or ML teams that own systems in production, with deployments that moved a commercial metric and kept running afterwards. Proofs of concept and dashboards are not what we're hiring for.
• Hands-on now, not formerly. Your first quarters are spent writing SQL and Python, engineering features, evaluating models and agents, and doing the work that gets a system live — not reviewing someone else's.
• Experienced across both traditional ML and agentic systems, and clear about how they differ in practice: propensity and uplift models, feature pipelines and drift on one side; tool and context design, prompt and retrieval iteration, evals against golden answer sets and trace observability on the other.
• Practised at running live systems rather than just launching them — monitoring, retraining, incident response, rollback, and the on-call reality of a decision the business depends on.
• Pragmatic about method: happy to argue for rules or heuristics where they capture most of the value, and to reserve models and agents for where they genuinely earn their keep.
• Comfortable with an undefined starting point. There's a long list of candidate use cases and a strong internal hypothesis, but no fixed roadmap — you'll be expected to form your own view, back it with numbers, and defend it to an executive audience.
• Strong on measurement — experimentation, uplift modelling, back-testing — and willing to hold your own work to a financial number that Finance will scrutinise.
• Fluent in the engineering side of production ML and agents (training pipelines, model registries, inference services, feature stores, drift detection, trace observability) and able to hold your own with senior engineers on latency, availability, observability and rollback.
• Able to hire, level and develop senior data scientists and ML engineers, and to set standards for a function that doesn't exist yet.
• Credible with executives and commercial stakeholders: you can take a use case from business problem to value estimate to prioritisation decision, and say no to low-value work.
• Experienced in payments, fintech, subscriptions, high-volume commercial operations or another regulated transactional domain — enough to get to a credible value estimate in a new area quickly.
• Comfortable treating regulated ML as a design constraint rather than paperwork — model risk assessment, DPIAs, auditability, traceability of model and policy versions to production decision logic.

Skills

Required

• SQL
• Python
• Machine learning
• Agentic systems
• Propensity and uplift models
• Feature pipelines
• Drift detection
• Tool and context design
• Prompt and retrieval iteration
• Evals against golden answer sets
• Trace observability

Benefits

• Why you’ll love working at Paddle
• We are a diverse, growing group of Paddlers across the globe who pride ourselves on our transparent, collaborative and respectful culture.
• We are a ‘digital-first’ company, which means you can work remotely, from one of our stylish hubs, or even a bit of both! We offer all team members unlimited holidays and 4 months paid family leave regardless of gender. We invest in learning and will help you with your personal development via constant exposure to new challenges, an annual learning fund, and regular internal and external training.