Trust Flow

• 

  Ai initial customer prompt

• 

  Ai chat response

Inspiration

In Nigeria’s ₦200 billion informal market, over 40 million small business owners drive 58% of the GDP—yet only 12% have access to formal bank credit. Microfinance banks (MFBs) want to lend to informal traders in Lagos, but rigid repayment terms and high default rates create massive risk. Traditional collection relies on manual outreach, leading to dispute friction and alienating borrowers. We built TrustFlow to bridge this gap: an empathetic, AI-driven assistant that makes debt collection fair, localized,

What it does

• TrustFlow is an intelligent, automated loan reminder and repayment management system built natively into WhatsApp.
• Localized WhatsApp Reminders: Sends automated nudges in plain English and Pidgin tailored to Lagos market dynamics.
• Repayment-History Driven: Eliminates friction by evaluating creditworthiness and hardship claims purely through past repayment patterns—no daily sales logging required from traders.
• Zero-Pressure & Human-in-the-Loop: Capable of handling routine reminders automatically (up to 3 nudges), but immediately routes complex hardship or pattern-dispute cases to a human loan officer.
• NDPA-Compliant: Adheres strictly to Section 37 of the Nigeria Data Protection Act (NDPA 2023) by keeping humans accountable for all adverse financial decisions.

How we built it

• Agent Architecture: Structured a multi-agent workflow featuring a Reminder Agent (handles zero-decision automated outreach), a Concierge Agent (captures trader context), an Analyst Agent (evaluates repayment history patterns), and a Liaison Agent (communicates officer decisions).
• Localized Messaging Engine: Designed templated WhatsApp flows for T-2 days, due date, T+1 grace periods, and post-decision holding messages in both formal English and Lagos Pidgin.
• Officer Dashboard Integration: Built automated case-card generation that presents loan officers with clear, audited summaries of a trader’s repayment history and stated hardship before they sign off.

Challenges we ran into

• Calibrating Conversational Tone: Finding the right balance in Pidgin was tricky. If a message sounds too formal, traders ignore it like spam; if it sounds too aggressive or miscalibrated, it feels predatory.
• Structuring Guardrails for Ethics: Ensuring the AI never makes autonomous legal or financial decisions on its own required strict prompt-engineering boundaries, capped retry limits, and explicit human escalation rules to stay NDPA-compliant.

Accomplishments that we're proud of

• Removing the Friction Barrier: Realizing that requiring traders to self-report daily sales created unnecessary friction, we successfully pivoted the core intelligence model to rely on repayment history alone for the MVP.
• 80% Officer Outreach Reduction: Architected a flow where the vast majority of repayments settle seamlessly via automated reminders, saving loan officers hours of manual follow-up while maintaining trust.

What we learned

• Simplicity Drives Adoption: Informal market traders don't want another app or complex transaction log. Meeting them where they already are (WhatsApp) with zero-friction requirements is essential for adoption.
• Ethical AI by Design: Human-in-the-loop architecture isn't just a safety feature—it's a critical legal and trust requirement when dealing with financial livelihoods in emerging markets.

What's next for Trust Flow

• v2 Optional Sales Pattern Analysis: Adding optional self-reporting capabilities for traders who want to upgrade their hardship claims from "plausible" to "evidenced" to unlock higher credit limits.
• Direct WhatsApp Payment Links: Integrating instant local payment rails (like Paystack or Monnify) directly into the WhatsApp chat so traders can settle loans in two taps. Core Banking System (CBS) Plugins: Building turn-key API connectors for popular microfinance banking software across West Africa.

Built With

• rag
• sandbox
• supabase
• twilio
• whatsapp