Project Overview
The Problem We Saw in Nigeria
Nigeria's insurance sector loses an estimated 20 billion naira every year to fraud. Traditional claims reserving methods assume the world is stable, but climate change, economic volatility, and changing consumer behavior break those assumptions.
Insurance penetration in Nigeria is below 1% of GDP, one of the lowest globally. Trust is low. Actuarial capacity is scarce. Meanwhile, the National Insurance Commission (NAICOM) has introduced the first formal insurtech regulatory framework, opening the door for technology-driven solutions.
We are building a system that addresses this gap.
What We Are Building
Three modules, each addressing a specific insurance pain point, all powered by the same pipeline engine.
ReserveAI
Claims reserving and loss forecasting using LSTM time-series models. Predicts per-claim payment development and catastrophe-exposed reserves with 15 to 20 percent accuracy improvement over traditional chain-ladder methods.
FraudShield
Claims fraud detection using CNN-LSTM hybrid models achieving 98.5 percent accuracy in published research. Cross-insurer duplicate detection, document forensics, and real-time NIID motor policy verification.
PriceLens
Dynamic risk pricing using telematics, IoT data, and climate risk scores. Enables individualized premium recommendations based on actual behavior rather than broad demographic categories.
Why Nigeria First
A fast-growing market with urgent problems and a new regulatory framework that welcomes innovation.
₦2.3T
Gross Premium Written
Q4 2025, 47% YoY growth
₦20B
Annual Fraud Loss
Estimated across industry
<1%
Insurance Penetration
Of GDP, one of lowest globally
₦724B
Claims Paid Q4
31.5% loss ratio
Regulatory tailwinds
NAICOM published the first formal insurtech guidelines in August 2025. The Insurance Industry Reform Act was signed into law. A new Partnering Insurtech license category creates an accessible entry point for technology companies.
Data infrastructure exists
The Nigeria Insurance Industry Database (NIID) provides real-time motor policy verification. NAICOM publishes quarterly market data. NiMet offers weather data for catastrophe risk modeling. The building blocks are available.
Fraud has immediate ROI
With 20 billion naira lost annually to fraud, AI-powered fraud detection pays for itself from the first claim scored. The business case does not require speculative future scenarios.
Actuarial capacity gap
Nigeria has few qualified actuaries. AI-powered reserving and risk scoring fills a structural capacity gap that traditional hiring cannot solve quickly enough.
Technology Foundation
LSTM neural networks, proven in academic research for insurance claims processing.
LSTM for claims reserving
Long Short-Term Memory networks model time-series patterns in claims development. Published research shows 15 to 20 percent accuracy improvement over traditional methods, especially during catastrophe years where old assumptions break down.
CNN-LSTM for fraud detection
Combining convolutional neural networks with LSTM achieves 98.5 percent accuracy in claims risk classification. The model detects patterns humans cannot see across document images, claim histories, and behavioral signals.
Climate-enriched models
Integrating weather and catastrophe data into LSTM models improves reserve accuracy for disaster-exposed portfolios. NiMet and NOAA data feed directly into the prediction pipeline.
Human-in-the-loop
NAICOM regulations prohibit unsupervised AI claims rejection. The system recommends. Humans decide. Every decision is logged with full provenance, model version, and confidence score for audit and regulatory reporting.
Pipeline Architecture
The same pipeline architecture that powers our procurement intelligence platform, adapted to insurance claims processing. Same engine, different data domain.
1
Source. Ingests claims from insurer systems, NIID policy lookups, and HMO platforms.
2
Hydrate. Enriches claims with policy details, claimant history, medical records, weather data, and telematics signals.
3
Filter. Detects duplicates, verifies documents through OCR and forensics, checks NIID, flags potential fraud.
4
Score. LSTM and CNN-LSTM models score each claim: legitimate versus fraudulent, reserve amount, risk category.
5
Decide. Generates reserve recommendation. Low-risk claims route to fast payout. High-risk claims route to human review. Every decision is logged.
Design Principles
Five principles that shape every decision in the system.
AI recommends, humans decide. No unsupervised AI claims rejection. Every high-stakes decision requires human approval.
We own intelligence, not insurance risk. The system sits between claims data and actuarial teams. We never touch funds or hold underwriting risk.
Compliance first. NAICOM guidelines, NDPR, and the Insurance Industry Reform Act shape the architecture from day one.
Explainability is non-negotiable. Every score comes with attention weights, feature importance, and a full audit trail. Black-box predictions are not acceptable in regulated insurance.
Evidence quality outranks model sophistication. A simple model on reliable data beats a complex model on weak data. We benchmark against traditional methods before deploying.
Plan Moving Forward
Where This Goes Next
The research is validated. The next phase is pilot and deployment.
01
Pilot with a Nigerian insurer. Process 500 historical claims through the pipeline. Deliver a fraud and reserving report at no cost. Prove the value on real data.
02
NIID integration. Connect to the Nigeria Insurance Industry Database for real-time motor policy verification. This is the fastest path to measurable fraud reduction.
03
Partnering Insurtech license. Apply for the new NAICOM Partnering Insurtech license once a pilot partnership is signed. The entry capital requirement is accessible.
04
Scale across product lines. Move from motor insurance into health, property, and general accident. Each line adds training data, which improves model accuracy across the portfolio.