↩ Work / EXPERIENCE

Head of Data.Employment

Period: Oct 2022 - Mar 2025
Company: GOZEM ↗

Owned the data function for Gozem's Sub-Saharan Africa group: governance, team scaling from 4 to 14+, FinOps, and fintech-grade analytics.

Built and mentored the data science and analytics team from 4 to 14+ across Sub-Saharan Africa, establishing data catalog practices and a domain-oriented data governance framework with domain leads.
Led FinOps initiatives delivering a 42.57% reduction in Google Cloud costs in a single quarter through data model redesign, BigQuery warehouse schema optimization, orchestration optimization, cost monitoring, and ETL/ELT pipeline tuning, while maintaining SLA on data freshness and integrity.
Led the development of ML-based fraud screening and financial transaction monitoring pipelines, reducing field agent payment processing time by 7× and delivering a Group Financial Closing Database for enterprise-wide performance monitoring.
Co-developed a domain-oriented data governance framework with the team, delivering a KPI framework with definitions, data source documentation, and table descriptions across the group.
Co-chartered the marketing analytics stack (CAC, LTV, cohort, and multi-touch attribution embedded into weekly Growth reviews), owning the end-to-end supporting data pipelines from campaign platforms.
Integrated analytics and ML features into mobile apps and internal platforms with Product and Business teams, including dynamic surge pricing for transportation and an extension of the agent incentive automation use case.

[•] PROJECTS & USE CASES

8 ENTRIES

Dataform modernization and BigQuery FinOps

Result Reduced Google Cloud costs by 42.57% in one quarter while maintaining data freshness and integrity expectations.

Problem / Need

Scheduled BigQuery queries had grown into a costly transformation estate with limited dependency control, repeated computation, and weak visibility into which workloads drove spend.

Solution

Led the migration to production Dataform modeling with explicit dependencies, incremental models, shared intermediate marts, partition-aware MERGE logic, and scoped assertions. Analyzed BigQuery job metadata, separated ingestion, transformation, reporting, and direct-analysis workloads, then targeted the models, schedules, and queries responsible for avoidable cost.

Tools

BigQuery·Dataform·SQL·BigQuery job metadata·Incremental modeling

Production orchestration on self-hosted Airflow

Result Replaced cron-based orchestration with a production scheduler supporting dependency-aware execution and operational monitoring.

Problem / Need

Cron jobs could trigger recurring work, but did not provide the dependency management, recovery controls, and shared operational view required as the pipeline estate expanded.

Solution

Designed and deployed self-hosted Apache Airflow on AWS EC2 as the production orchestrator, migrated scheduled workloads from cron, and used DAG-level dependencies, retries, and monitoring to operate data workflows consistently.

Tools

Apache Airflow·AWS EC2·Python·Cron

Lower-cost analytical serving with ClickHouse

Result Created a bounded-history serving path for recurring dashboards, reducing direct analytical pressure on BigQuery.

Problem / Need

Interactive Metabase dashboards repeatedly querying the central warehouse increased BigQuery demand even when each use case needed only a limited history window.

Solution

Deployed open-source Metabase and ClickHouse on AWS EC2. Built Airflow replication workflows that moved selected BigQuery tables or partitions into ClickHouse with use-case-specific history windows ranging from three to eighteen months.

Tools

ClickHouse·Metabase·Apache Airflow·BigQuery·AWS EC2

Data team capacity expansion

Result 12 hires onboarded end-to-end from 50+ evaluated candidates, with 30/60/90-day ramp milestones.

Problem / Need

The data team capacity needed to scale across African operations to support growing operational and BI needs, requiring 2 senior hires alongside a data engineer owning pipelines and platform.

Solution

Authored the job scopes, technical requirements, and business cases justifying the senior hires to the leadership. Designed and ran technical interviews and assessments, evaluated 50+ candidates, negotiated offers end-to-end, and coordinated onboarding for 12 total hires with tooling access, team integration, and 30/60/90-day ramp milestones.

Tools

Team capacity planning·Hiring pipeline management·Team productivity tools·Candidate evaluation·Onboarding

Data science use-case knowledge base

Result 100+ use cases catalogued; top 15 adapted with Gozem-specific feasibility and outcome assessments.

Problem / Need

The data team needed a curated reference of proven use cases to prioritize roadmaps, drawing inspiration from how Gojek, Grab, and Uber operate at scale and adapting to Gozem’s Sub-Saharan markets.

Solution

Catalogued 100+ use cases across demand forecasting, driver scoring, fraud detection, surge pricing, and route optimization. Selected the top 15 and adapted each with Gozem-specific feasibility, data availability, and applicability assessments. Defined success metrics, technical challenges, infrastructure requirements, and estimated business value, structured as a living reference for the data and product teams.

Tools

Research & development·Market analysis·Use-case design·Proof of concept development

Ebanking analytics and reporting infrastructure

Result ~150 incidents surfaced per month; flagged a fraud worth more than 25M XOF; fintech warehouse powering regulatory reporting.

Problem / Need

Gozem Money (fintech vertical) needed real-time fraud monitoring, a fintech-grade data warehouse, and a reporting layer to support mobile payment operations and regulatory needs across African markets.

Solution

Designed and supervised deployment of a micro-batch processing pipeline monitoring anomalous patterns, fraud signals, and threshold breaches. The system surfaces ~150 incidents per month and flagged a major fraud worth more than 25M XOF. Built the fintech data warehouse for Togo operations, powered by ETL/ELT pipelines and data models as the foundational layer for financial analytics, regulatory submissions, and stakeholder dashboards (transaction volumes, product adoption, revenue, cohorts). Established data quality pipelines and documented data models, reporting glossary, and metric definitions.

Tools

Apache Airflow·Python·MongoDB·BigQuery·Dataflow·Cloud Storage

Self-service no-code data exploration platform

Result 12+ active manager users; routine query load on the central data team halved.

Problem / Need

Non-technical users across operations, finance, product, and customer support depended on the central data team for routine reporting and ad-hoc queries, creating a bottleneck.

Solution

Assessed data literacy and exploration needs across all Gozem markets. Coordinated configuration of a no-code query and exploration layer (Metabase) integrated with Gozem’s central data warehouse. Adoption reached 12+ active manager users, halving routine query volume on the central data team.

Tools

Metabase·BigQuery·Looker Studio

Delivery agent incentive automation

Result Incentive-related ticket volume cut nearly in half; support disputes resolved within 24h.

Problem / Need

Delivery-agent positioning across high-demand zones was unmanaged, and incentive payments required manual finance intervention plus heavy customer-support load to resolve disputes.

Solution

Modeled demand by zone, time-of-day, and event type to identify underserved areas. Designed a performance framework (acceptance rate, completion rate, zone presence, quality scores) and end-to-end payment automation from incentive calculation through disbursement instructions. Built a justification database with pre-computed data points enabling support to resolve cases within 24h. Cut incentive-related ticket volume nearly in half through pre-emptive payment automation.

Tools

BigQuery·Python·Airflow·Google Sheets

← Work