Data Scientist (AI-Native) at Umba
Job Description
Job Summary
We’re hiring a Data Scientist to own the two models that decide whether Umba grows profitably: how we
acquire customers, and how we underwrite them.
On the credit side, you’ll build and continuously improve the scoring systems that decide who we lend to and
on what terms — drawing on bank statement data, payments history, CRB (Credit Reference Bureau) data,
and the behavioural signals we collect across our app. The same scoring stack needs to serve both digitally
acquired customers and the customers our sales team brings in for underwriting, so you’ll design for both
flows.
On the growth side, you’ll optimize how we spend marketing budget to acquire those customers — ad
targeting, funnel conversion, channel attribution, and the experiments that tell us which levers actually move
CAC and LTV. You’ll own the loop from “who do we target” through “did they convert” through “did they
repay.”
This is not a traditional data science role.
We operate in an AI-native environment, where the team leverages Claude Code, Codex, and other LLMbased systems to accelerate analysis, generate model code, build pipelines, and iterate quickly. As a result,
the role increasingly focuses on:
- Defining clear problem specs that AI agents can execute against
- Reviewing, validating, and hardening AI-generated analyses and code
- Building feedback loops that let models improve automatically with new data
- Setting the quality bar — what “good” looks like for a model in production
- You’ll collaborate closely with Engineering, Product, and the Sales team to ship models that affect lending
- decisions on day one. This is a highly technical, in-office role in Nairobi. You’ll join a small, high-performing
- team where ownership is expected and impact is immediate.
Key Duties & Responsibilities
Credit & underwriting
- Build, deploy, and continuously improve credit scoring models using bank statement data, payment
- histories, CRB pulls, and in-app behavioural signals
- Design automated underwriting flows that serve both digitally acquired customers and salessourced applications
- Implement model retraining pipelines so scoring improves as we accumulate repayment outcomess – not as a quarterly project
- Own model performance monitoring, drift detection, and automated alerting
- Partner with Risk and Operations on policy thresholds, override rules, and the human-in-the-loop processes that wrap the models
Growth & marketing analytics
- Optimize ad targeting across our acquisition channels — audience selection, bid strategy, creative performance, lookalike construction
- Instrument and analyze the acquisition funnel end-to-end (impression → click → install → KYC → first loan → repayment)
- Design and run A/B tests on acquisition and product experiences; build the experimentation infrastructure so the team can run tests without you
- Build attribution and LTV/CAC models that the business can actually act on Cross-cutting
- Write clear technical specs that AI-assisted workflows can execute against
- Use AI tools (Claude Code, Codex, etc.) to move 10x faster on data wrangling, feature engineering, and analysis — while rigorously validating outputs
- Extend our data platform with new sources (third-party APIs, CRB providers, payment rails) when a model needs them
- Process, clean, and verify data integrity — especially for anything that touches lending decisions
- Present findings clearly to non-technical stakeholders; defend recommendations with data
Educational Qualifications, Experience, & Skills Required
- 4+ years of hands-on data science / applied ML in production environments
- Strong Python (pandas, scikit-learn, numpy) and SQL — you can go from raw data to deployed
- model without waiting on engineering
- Deep practical experience with classifier and regression modeling — feature engineering, model
- selection, calibration, evaluation under class imbalance
- Solid applied statistics: hypothesis testing, regression, experimental design, dealing with selection
- bias and censored outcomes
- Experience working with messy real-world financial data (transactional data, bank statements,
- payments, credit bureau data) — or strong evidence you can ramp on it quickly
- Comfort with relational databases (Postgres / MySQL) and modern data tools
- Strong written and verbal communication — you can explain a model’s behaviour to a credit officer,
- a marketer, and an engineer in the same week
