Problem
The working capital of Brazilian SMEs costs 30-40% a.a., with relevant parts captured by intermediaries. Brazil has R$ 2,5 Tri in stock of credit PJ (BACEN, 2025) and more than 20 millions of active SMEs. The problem is structural: (1) credit models analyze operations in isolation, ignoring systemic dynamics and the links between companies in the same production chain; (2) the traditional credit score is a black box, meaning the borrower does not understand its rate or how to improve its risk profile; (3) Open Finance data already exists, but they are not integrated into systemic and explainable credit models. The result is expensive, opaque, and suboptimally allocated credit across the chain.
ChainCredit Intelligence solves all of this with knowledge, innovation and innovation.
Solution
Three integrated pillars: (1) Explainable AI with Gradient Boosting (XGBoost) with SHAP values, where each discount rate is explained variable by variable, with native regulatory transparency (LGPD/BACEN) . (2) Financial graphs in production chains modeled as targeted networks, with risk propagation, centrality, and contagion (Eisenberg-Noe) between suppliers and buyers. (3) Operational Research for a network flow problem (min-cost flow) that optimally allocates liquidity, minimizing the total capital cost of the chain as an integrated system, not as isolated operations. A pipeline of 6 layers (data → Credit/ML → XAI → graphs → optimization → output) transform raw Open Finance data into optimal liquidity allocation with end-to-end auditable explicability.
Business model
B2B SaaS + licensing, with recurring revenue linked to the volume financed: (1) SaaS by volume with a fee of ~0,12% a.m. on the volume financed, for fintechs, FIDCs and digital banks; (2) XAI+Grafos engine license with annual license + customization, for medium-sized banks and cooperatives; (3) Analytics & Data with subscription to systemic risk reports for managers, insurers, and regulators. Projected average ticket: ~R$ 50 millions of volume/year per customer, generating around R$ 720 thousand of ARR. Estimated gross margin of 70-75%.
Market
At this stage we are focused on the Brazilian market. TEAM: R$ 800 bi as the total credit market PJ, SCF and advance receivables in Brazil. SAM: R$ 180 bi with SMEs in production chains with more than 5 suppliers and access to Open Finance. SOUND: R$ 2,5 bi, focus on the former 2 years in fintechs, FIDCs, and SCF platforms as B2B clients. Reference: PJ credit stock in Brazil of R$ 2,5 trio (BACEN, 2025) and more than 20 millions of SMEs active in the country.
Competitors
Three categories, none with systemic intelligence: (1) Traditional banks, without XAI, without network/graph vision, without financial contagion. (2) Credit Fintechs (e.g.: IQ Tech, today a Brazilian unicorn valued at more than US$ 1 bi, with a round led by General Atlantic), use ML for PD, but explicability is partial and there is no chain modeling. (3) Global Supply Chain Finance platforms, such as C2FO (more than US$ 580 My recruits, investors such as SoftBank, Temasek and Mubadala), strong in anticipating receivables, but without systemic optimization via min-cost flow or structural explicability (SHAP). ChainCredit competes at the intersection of these three fronts, with a differential that none of them covers in isolation.
Competitive differentiation
Compared to banks, credit fintechs, and SCF platforms, ChainCredit is the only one that simultaneously combines: ML for PD (XGBoost) + structural explicability (SHAP, not a subsequent interpretive layer) + network/graph view (Eisenberg-Noe) + financial contagion (PD adjusted to the network) + systemic optimization (min-cost flow) + native compliance by design (LGPD/BACEN, aligned with the European AI Act's explanability requirement). Banks and fintechs charge the most 1-2 of these dimensions; no mapped competitor covers all six.
Entry barrier
Two barriers that are reinforced over time: (1) Cumulative graph - the production chain map grows with each operation, creating a proprietary data asset that is difficult to replicate without the same transactional history. (2) Learning curve where PD models continuously improve with real data, expanding competitive advantage with each cycle. In addition: industrial secrecy and INPI registration planned for the software, the technological engine planned for a patent, and the scientific project with the algorithmic formulation already deposited with the National Library as proof of precedence.
Traction
Pre-recipe, with institutional and commercial validation: qualification in the FAPESP PIPE-JT Phase process 1 (2026/15550-6), with completed technical due diligence; pilot contribution from R$ 240 thousand under structuring with the owner of a FIDC, to validate the MVP based on real data from a FIDC and a commercial bank; and advanced conversations with a large securitization/FIDC/Factoring group for complementarity and possible sponsorship. Commercial MVP in development (FastAPI, 5 pipeline modules), with min-cost flow optimization module already validated in 3 scenarios.