Background
Lebanon's Banque du Crédit Méditerranéen (BCM) — a mid-size commercial bank with 18 branches across Greater Beirut and Mount Lebanon — faced an existential challenge by late 2022: its entire credit scoring infrastructure had become non-functional. The Banque du Liban's credit registry had ceased meaningful operations, and the two private credit bureaus that had operated in Lebanon both suspended services following the 2019 financial crisis.
With a loan portfolio of approximately $340 million (USD-denominated) and growing pressure from depositors and regulators to resume retail lending, BCM needed a new approach to borrower risk assessment — one that did not depend on the formal credit infrastructure that no longer existed.
The Challenge
- No functioning credit registry or bureau data available for 94% of applicants.
- Existing internal scoring models trained on pre-crisis data were producing unreliable outputs.
- Loan officers reverting to purely relationship-based lending, creating concentration risk.
- Regulatory pressure to demonstrate systematic risk controls before resuming retail lending.
The Berix Solution
BCM engaged Berix in Q1 2023 to develop an alternative data credit scoring framework. The engagement had two phases: first, a data audit to identify which Berix signals correlated with historical default behavior in BCM's own pre-crisis loan book; second, the construction of a new scoring model using those validated signals.
The final model incorporated 14 Berix data signals across four domains: neighborhood economic stability indices, business registration and continuity data, utility payment proxies, and employment sector stability scores. Critically, the model was designed to be explainable — each score component could be articulated to loan officers and, where required, to regulators.
Results
Key Learnings
The BCM engagement demonstrated that alternative data can substitute effectively for traditional credit bureau data in crisis contexts — but only when the data provider has genuine local depth. Generic international alternative data products performed poorly in validation testing; Berix's Lebanon-specific signals were the differentiating factor.
The project also highlighted the importance of model governance. BCM's risk committee required full documentation of each data signal's source, update frequency, and potential bias vectors before approving the model for production use. Berix's data provenance documentation was central to clearing this governance hurdle.
“We went from making credit decisions based on gut feel to having a systematic, data-driven framework. The default reduction speaks for itself.”
— Chief Risk Officer, BCM (anonymized)