Lentera Microfinance Loan Officer Assistant
A Jakarta microfinance NGO gave 34 field loan officers an AI co-pilot for KYC, credit memos, and follow-ups.
Impact
What changed.
Officer leverage
Average borrowers per officer rose from 350 to 540 within the first year, without quality regression. Portfolio grew 54% on flat headcount.
Risk capture
Early repayment-stress signals now reach officers an average of 19 days before formal default. Portfolio-at-risk over 30 days fell from 4.7% to 2.3%.
Donor reporting
Donor portfolio reports now generate continuously instead of through three-week cycles. The NGO won two new institutional donors partly on the strength of reporting credibility.
The challenge
Before
Lentera is a Jakarta-based microfinance NGO serving 12,000 women-owned micro-enterprises across Java and Sumatra with average loan sizes of IDR 8M. Their 34 field loan officers were spending more time on paperwork than with borrowers — KYC documentation, credit memo drafting, repayment follow-up, group-meeting notes. Each officer was managing 350 active borrowers on average, and the paperwork load capped their ability to grow the portfolio without losing service quality.
- 34 field officers managing 350 active borrowers each on paper-heavy workflows
- KYC documentation captured on forms, photographed, then re-keyed at branch
- Credit memos drafted on laptops in evening hours after field work
- Group meeting notes captured by hand, transcribed weekly
- Repayment follow-up done by SMS templates and phone calls
- No way to surface borrowers at risk of repayment slip before it happened
- Promotion to larger loan tiers slow because credit history compilation was manual
- Donor reporting consuming three weeks per cycle for portfolio analyst team
The solution
What we built
We deployed a mobile-first loan officer assistant that runs on the officer's phone in the field, with offline capability for remote villages. KYC captures borrower identity through ID photo and a structured questionnaire; the agent extracts and validates fields, flags anomalies, and produces a clean KYC pack. Credit memos are drafted by the agent from the structured intake plus business assessment notes; the officer reviews and signs off. Group meeting attendance, savings deposits, and discussion notes are voice-recorded and transcribed by the agent in Bahasa Indonesia with structured tagging. A risk-watch model flags borrowers showing early signals of repayment stress (missed group meetings, smaller-than-usual deposits, sentiment shifts in officer notes) so the officer can intervene supportively before a default. Donor-facing portfolio analytics roll up automatically from the same structured dataset.
Core workflow connections
How the system flows.
- Borrower VisitID CaptureKYC ValidationAnomaly Flagging
- Business AssessmentStructured CaptureCredit Memo Draft
- Officer Sign-offCredit Committee SubmissionApprovalDisbursement
- Group MeetingVoice RecordingIndonesian TranscriptionTagged Notes
- Repayment WatchRisk SignalOfficer Intervention Prompt
- Promotion EligibilityCredit History CompilationTier Upgrade
- Donor ReportingPortfolio Roll-upOutcome Metrics Surfaced
- Offline mobile capability for remote village field work
- Risk-watch flagging early signals before repayment default
- Voice-first input respecting field-officer workflow realities
Process
How we built it.
Borrower Visit → ID Capture → KYC Validation → Anomaly Flagging
Business Assessment → Structured Capture → Credit Memo Draft
Officer Sign-off → Credit Committee Submission → Approval → Disbursement
Group Meeting → Voice Recording → Indonesian Transcription → Tagged Notes
Repayment Watch → Risk Signal → Officer Intervention Prompt
Promotion Eligibility → Credit History Compilation → Tier Upgrade
Donor Reporting → Portfolio Roll-up → Outcome Metrics Surfaced
Offline mobile capability for remote village field work
Risk-watch flagging early signals before repayment default
Voice-first input respecting field-officer workflow realities
Start a project
Field officers buried in paperwork instead of with borrowers?
We build field-officer agents that respect mobile-first realities — offline-capable, voice-friendly, and tuned to the language and culture of the work.
No retainer lock-in · Month-to-month · Full transparency
More work
Up next.
A 72-person Melbourne architecture firm replied to 3x more public-sector RFPs after deploying a tuned response agent.
Vertice Architects AI RFP Response Engine
A Dubai B2B medical supplier handles 720 daily quote requests with an AI agent wired to live inventory and tiered pricing.