Hiring 2 ML Engineers: How a Growth-stage FinTech in London Scaled with Second Talent
Published April 25, 2026
At a Glance: A London-based FinTech company with $12M ARR hired two senior ML engineers through Second Talent to build a fraud detection and credit scoring pipeline. The AI-native engineers, based in the Philippines via EOR, delivered 96% fraud detection accuracy while reducing false positives by 70% and saving the company 65% compared to London-based ML talent.
The Challenge
The payments company was processing 180,000 transactions monthly by early 2025. Their rule-based fraud detection system flagged 22% of legitimate transactions as suspicious. Manual review teams spent 340 hours per month investigating false positives. Customer complaints about declined payments increased 45% quarter over quarter.
The VP of Engineering needed two senior ML engineers with specific expertise in feature engineering and MLOps. The budget allowed for $140,000 per engineer annually in London. At that rate, they could only afford mid-level talent without the 8+ years of production ML experience required. Local agencies quoted 12 to 16 weeks for placement with no guarantee of fraud detection domain knowledge.
The company had no existing ML infrastructure. They needed engineers who could build the entire pipeline from scratch. Data ingestion, feature stores, model training, A/B testing, and automated retraining all had to be production-ready within six months. The timeline was critical because a competitor had just launched an AI-powered fraud product.
Previous attempts to hire remotely through freelance platforms failed. Three contractors over eight months delivered models that could not scale beyond 10,000 transactions daily. None had experience with real-time inference or model monitoring. The CTO needed a partner who could source proven ML engineers and handle employment compliance across jurisdictions.
The Solution
Second Talent presented four candidates within 11 days. All had 8+ years of ML engineering experience. Two specialized in financial fraud detection with previous work at payment processors and digital banks. Both were based in the Philippines and available to start within three weeks through Second Talent’s EOR service.
The engineers were AI-native from day one. They used Claude and GPT-4 to accelerate feature engineering work. Code generation for data transformation pipelines reduced development time by 40%. They built 87 features in the first month compared to the 34 features the previous contractors delivered in three months. AI-assisted code review caught 23 potential bugs before production deployment.
Second Talent handled all employment logistics through EOR. Contracts, payroll, benefits, and tax compliance were managed without the company establishing a Philippine entity. The engineers received $6,200 and $6,500 monthly, well above the $5,000 minimum. Total loaded cost including EOR fees was $89,000 per engineer annually. This represented 65% savings compared to the $140,000 London budget.
The team integrated seamlessly with the London-based product and engineering teams. Daily standups at 9 AM GMT worked with the Philippine time zone. The ML engineers participated in architecture reviews and sprint planning. They documented every model decision in Notion and maintained a shared Slack channel for real-time questions. The CTO reported zero communication friction after the first two weeks.
The Results
The fraud detection model went live in production after 19 weeks. Initial accuracy reached 94% with a false positive rate of 8%. After two months of automated weekly retraining on new transaction data, accuracy improved to 96%. False positives dropped to 6.6%, representing a 70% reduction from the original 22% rate. Manual review hours decreased from 340 to 95 per month, freeing the operations team for higher-value work.
The credit scoring pipeline processed 45,000 applications in the first quarter of operation. Default prediction accuracy measured 89% compared to 71% with the previous rule-based system. The company approved 18% more applications while maintaining the same risk threshold. Revenue from newly approved customers added $780,000 in the first six months.
Model retraining became fully automated by month five. The MLOps pipeline pulled new transaction data weekly, retrained models, ran A/B tests against the production model, and deployed winners automatically. The system processed 240,000 transactions monthly by Q4 2025 with no degradation in accuracy. Infrastructure costs stayed flat at $3,200 monthly because the engineers optimized cloud spending through better resource allocation.
Key Outcomes
- Fraud Detection Performance: Achieved 96% accuracy with 70% reduction in false positives, cutting manual review time from 340 to 95 hours monthly and eliminating customer friction from declined legitimate transactions.
- Credit Scoring Improvement: Increased default prediction accuracy from 71% to 89%, enabling 18% more application approvals and generating $780,000 in additional revenue from previously rejected customers in six months.
- Cost Efficiency: Saved 65% on ML engineering costs compared to London market rates, paying $89,000 per engineer annually versus the $140,000 budgeted, while accessing senior talent with 8+ years production experience.
- Automated MLOps: Built fully automated weekly model retraining pipeline that handles 240,000 monthly transactions, runs A/B tests, and deploys improved models without manual intervention, maintaining flat infrastructure costs at $3,200 monthly.
“We needed ML engineers who could build production systems, not just train models in notebooks. Second Talent found us two engineers with real fraud detection experience who delivered 96% accuracy and saved us 65% compared to London hires. The EOR service meant we had them onboarded in three weeks instead of spending four months setting up a Philippine entity.”
CTO
Want results like these?
We source, vet, and manage senior AI-native engineers. $0 upfront. Matched in 24 hours.
Get Matched Now
