A credit model that works perfectly in Bulgaria can't simply be copied to Mexico or Kenya. That is the main takeaway Zvezdelin Stoyanov, Head of Data Science at MFG, shared with Digitalk. MFG is a fintech company that started in Bulgaria and Romania and now also operates in Mexico, the US, Peru and Kenya.
Why data doesn't "speak one language"
Credit risk and scoring models depend heavily on local data, and that data differs in three ways:
- Availability. One country has a credit bureau with a rich history; another has almost nothing.
- Type. In some markets bank data is the main signal; in others it's mobile payments.
- Usage rules. The law decides which data can be used at all.
So for a fintech company, the first challenge in a new market is access to data. Entering the market comes second.
The approach: data first, model second
The MFG team first checks what data it can legally obtain and what that data says about customers, and only then discusses a specific model. According to Stoyanov, a common mistake is to transfer a model before understanding the logic behind it.
The parallel with international websites
We see the same mistake in web projects built for several markets. The site gets translated, but not adapted. Here is what usually gets missed:
- Different search behaviour. People in different countries search for the same service using different words. Translating keywords directly rarely works.
- Different legal requirements. Business disclosures, cookies, privacy and consumer protection rules vary.
- Different trust signals. Payment methods, couriers, currency and address formats.
- Technical structure. Separate URLs for each language, correct hreflang tags and canonical URLs.
If you are planning a multilingual website or online store, see our approach to multilingual websites. Related: why structured data has become a strategic asset.
Source: Digitalk: Data doesn't speak one language (in Bulgarian). This article is an independent analysis by Penkov Studio.
