Indicina — AI-Driven Credit Scoring, Nigeria

Served as an analytics engineering consultant for Indicina, a credit-scoring startup based in Nigeria. The core deliverable was the “Decide” model — an AI-driven credit scoring system that evaluates small businesses using bank statement transaction data and merchant records. The model combines NLP-based transaction segmentation with predictive AI to generate forward-looking credit scores, addressing a critical gap in financial inclusion for underbanked small businesses.

1. Segment transactions using NLP models

Built NLP pipelines to ingest raw bank statement text and automatically categorize each transaction into meaningful segments — payroll, rent, utilities, merchant income, etc. Accurate segmentation is the foundation on which the entire scoring model depends.

# Example: Transaction segmentation with keyword-based NLP
import pandas as pd

CATEGORIES = {
    "income": ["salary", "payroll", "transfer in", "deposit", "revenue"],
    "rent": ["rent", "lease", "landlord"],
    "utilities": ["electric", "water", "internet", "airtime"],
    "merchant": ["pos", "merchant", "sale", "invoice"],
}

def segment_transactions(df: pd.DataFrame, desc_col="description") -> pd.DataFrame:
    def classify(text):
        text_lower = text.lower()
        for cat, keywords in CATEGORIES.items():
            if any(kw in text_lower for kw in keywords):
                return cat
        return "other"

    df["category"] = df[desc_col].apply(classify)
    return df

2. Build predictive scenario models for future transactions

After segmentation, trained AI models to project likely future transaction patterns for each business. These scenarios inform the credit score by estimating a borrower’s capacity to repay under various economic conditions — not just their historical behavior.

# Example: Simple transaction forecasting skeleton
from sklearn.linear_model import Ridge
import numpy as np

def forecast_monthly_income(history: np.ndarray, horizon: int = 3) -> np.ndarray:
    X = history[:-1].reshape(-1, 1)
    y = history[1:]
    model = Ridge(alpha=1.0)
    model.fit(X, y)
    forecasts = []
    last = history[-1]
    for _ in range(horizon):
        pred = model.predict([[last]])[0]
        forecasts.append(pred)
        last = pred
    return np.array(forecasts)

3. Generate composite credit scores for small businesses

Combined the segmented transaction features and forward-looking scenario outputs into a single interpretable credit score. The scoring engine was designed to be explainable — flagging the key factors driving each decision — so that both the platform and the end borrower understand the outcome.

4. Integrate with merchant transaction data

Extended the model beyond bank statements to incorporate merchant-side transaction records. This dual-source approach provided a richer and more reliable signal, particularly for businesses that operate primarily through point-of-sale or invoice-based revenue.