Credolab

Credolab — Device-level behavioural intelligence captured at the moment of a loan application: what is on the applicant's handset at install time (app inventory, app categories, apparent income and spending apps, device make/model/age, device integrity and tamper signals, emulators and test devices, multi-application clustering), plus acquisition-channel attribution and the KYC/onboarding events around it. Credolab layers this on top of bureau data at the decision point and then observes the disbursal and servicing outcome. The key artefact is a benchmark against an African digital bank's own…

Credolab offers (Alternative, Sentiment) — Device feature vectors per assessment plus a matched outcome label for each disbursed loan. Reasoned from the revenue range and per-assessment pricing, the corpus is plausibly tens of millions of assessments, of which the highest-value subset — assessments followed by a known repayment outcome — runs in the low millions. The rarest component is the documented same-population, same-window paired benchmark, which volume cannot substitute for..

Training device-behaviour credit models for thin-file and underbanked populations; building reject-inference and lift-curve datasets from approved/denied cross-classification between old and new models; fraud and first-party-default model development using device integrity and emulator signals; acquisition-channel quality and marketing-spend optimisation models for lending apps; model champion/challenger benchmarking methodology with a documented real-world lift; transfer-learning for new market entry where no bureau history exists

The data is with 10 years of history.

Coverage spans US, APAC, Other; Financial Exchanges & Data, Application Software; alternative, sentiment; equities.