You find out after it's already happened.
A month-end report is a photograph; the business is a film. By the time the ledger shows a miss, the intervention window has closed.
Credit Risk Orchestration · Nairobi
The borrower you just wrote off was your best customer.
A read-only intelligence layer that shows you a business heading into trouble weeks before your own reporting does — early enough to restructure instead of write off. It sits on top of the core system you already run. No replacement, no downtime, and every credit decision stays with your committee.
- Borrower records analysed
- 101,000
- Corrected position after duplicate detection
- 83,631
- Loan book monitored
- KES 1.4B
- Fewer non-performing loans
- ~25%
- Improvement in portfolio at risk
- ~30%
- To disburse
- 21 days → 48 hours
Observed across deployments and pilots. Direction, not guarantees.
If you run a loan book
Banks, microfinance banks, SACCOs and digital lenders.
See distress earlier, restructure instead of writing off, and recover more of what you have already lent.
ExploreIf you manage a fund
DFIs, impact lenders and programme managers.
Enforce mandates at origination, monitor downstream portfolios, and evidence where capital landed.
ExploreIf you insure, buy debt, or make introductions — we also work with insurers, distressed debt buyers and referral partners.
Start here
A backtest on your own loan portfolio, at no cost.
Send us a CSV export of your loan ledger — 24 months if you have it. We run it through UNBRDN and show you what our screen would have flagged, and when, against what actually happened. No integration, no IT involvement, no commitment. If it tells you nothing you didn't already know, that's a useful answer too.
The machinery is built to write people off.
Not through negligence. Through timing. Every part of a conventional credit process reports on the past, and the past is the one thing the person accountable for the loan portfolio cannot change.
The same borrower gets different answers.
Similar files, different officers, different outcomes. Portfolio risk you can't predict because the decisions weren't consistent.
You can't show your working.
When an auditor asks how a loan was approved, “the officer used judgement” is an increasingly difficult answer.
Recovery starts too late to work.
By the time the recovery team mobilises, the borrower is in crisis and most of the recoverable value has gone.
It is not a moral failure by lenders. It is an information failure, and that is solvable.
What UNBRDN does
Recovery and rehabilitation first. Screening supports them.
UNBRDN reads from the systems you already run. All credit approvals remain with your own committee.
Post-lending
Recover more of what you have already lent.
Continuous monitoring of every account against your own policy thresholds, a structured turnaround path instead of a write-off, and a collections command centre for the accounts that do go bad. Promises kept, right-party contact and unworked case queues, tracked per collector.
Pre-lending
Price the borrower on real cash flow, not a snapshot.
Upload an applicant book and every borrower is scored against your own debt service ratio threshold, your mandate and your risk limits. Bank and mobile-money statements read, normalised and turned into a decision-grade view.
Analyse
Turn documents into a defensible credit view in minutes.
Borrower-level analysis producing an end-to-end financial profile, and standalone statement analysis focused on inflows, outflows and cash-flow signal.
Behind these three sit origination, mandates, market discovery, sustainable finance, transfer pricing and a debt marketplace. Ask and we will show you.
That third step is the point. A borrower you restructured who then repaid is someone you have seen under pressure — on any sensible measure a better risk than a stranger with a clean application.
A marketplace layer for portfolios that need to change hands is in development.
The loop most lenders never close.
- Screen
- Monitor
- Rehabilitate
- Re-lendreturns to Screen as a known risk
Most lenders never close this loop. The cured borrower is the best-qualified lead you already own.
Access
Two ways into the platform.
For lenders and funds
UNBRDN runs as a private, isolated workspace for each institution. Access is by invitation so your environment is provisioned and secured before anyone signs in.
Passkey authentication. No passwords, ever.
For small businesses
If you are a business carrying debt, run a free assessment yourself. Share a few details and get an analysis of your position and what to tackle first.
For commercial lenders
Your NPL ratio is decided in the first thirty days, not the last ninety.
An account a week late is a timing problem. The same account at ninety days has already paid somebody else first, the goodwill is spent, and it is heading for classification, provision and security realisation. The money has not changed. The recoverability has.
- Every account graded daily into arrears buckets, with movement between buckets tracked
- Roll rates and cure rates by officer, branch, sector, product and vintage
- Flags on behavioural signals before the first missed payment — late-but-within-grace payments, creeping part-payments, falling collection-account turnover
- A watchlist where every line carries an owner, an action and a date
- Recommended action with the reasoning shown; your team can accept, amend or overrule, and whichever they do is recorded
- Promise-kept rate as a metric a credit manager can actually manage
For funds and development finance
Prove where the capital landed.
- Mandate enforcement at origination — sector caps, demographic targets, geographic and impact criteria applied when a loan is written, not discovered in a quarterly audit
- Downstream visibility — live portfolio performance across intermediaries, replacing static quarterly reporting
- Verification at scale — evidence gathered and recorded per decision, with an audit trail a funder or examiner can follow
- Climate and sustainable finance mandates configured alongside every other parameter
- IFRS 9 Stage 2 tracking automated, with expected-credit-loss scoring in audit-ready form
Also serving
Insurers
Distribution and risk data for embedded cover; the insurer remains risk carrier and product owner.
Distressed debt buyers
Propensity scoring so you know which accounts will pay before you bid.
Referral partners
Introduce a lender or fund and share in what your network unlocks.
Get in touchIt sits on top. Nothing gets replaced.
Your existing systems
Core bankingCollectionsPortfolio toolsExternal data sources
UNBRDN intelligence layer
Read-onlyRecoverRehabilitateMonitorScreen
Outputs
Early-warning alertsCredit memosCommittee packsFull audit trail
Credit AI that runs inside your environment.
Borrower data remains within the client-controlled deployment and approved data plane. VALR Extract and VALR Analyst can process documents locally, with external LLM egress disabled by default.
- Read-only by design.
- We read from your systems and never write to them.
- Every decision reconstructable.
- Cryptographic audit trail on every check, filing and committee decision.
- Bounded autonomy.
- The platform performs the analysis. Approving, restructuring or writing off a loan remains a decision of your credit committee.
- ISO 27001-aligned
- GDPR and Kenya Data Protection Act 2019
- IFRS 9 aligned
- Passkey authentication
- Enterprise SSO and SCIM
- Multi-tenant isolation
Data sovereignty
Built for institutions that cannot hand their borrower data to anyone.
VALR Extract
Reads bank and mobile-money statements. Resolves transaction direction against the running balance rather than inferring it from the description, because on a mobile-money statement the same paybill can be income for one borrower and an expense for another.
VALR Analyst
Reasons through the credit decision and produces a defensible, reviewable output.
We built two proprietary models — one that reads bank and mobile-money statements, one that reasons through the credit decision. Both are designed to run within the bank-controlled environment, so borrower data does not have to be sent to third-party LLMs to get a decision. In our own benchmarking they approach the quality of leading frontier models at a materially lower cost per decision.
- Read-only by design — UNBRDN reads from your systems and writes nothing back
- Borrower data is not sent to third-party LLMs; AI processing can run within the bank-controlled environment
- Every recommendation traces back to the signals behind it — no black-box score
- Cost per decision falls as your loan portfolio grows
Benchmark figures are from internal testing and describe direction, not a guaranteed result in any given deployment.
Who built it.
Alex Ndubai
Founder & CEO
Two decades in credit recovery and SME lending.
Erick Oluoch
Co-Founder & CFO
20+ years in institutional credit risk. Founder Chairman, Association of Debt Recovery Agents Kenya.
Ernest John Ndung'u
Co-Founder & CTO
Architect of the UNBRDN platform. Senior Software Engineering Manager at Microsoft.
We came at this from the recovery end. Twenty years on the wrong side of these decisions shows up in what we chose to build.
What we have observed so far.
Observed across deployments and pilots — direction, not a guarantee.
- ~90%
- faster underwriting — mixed borrower data to a decision-grade view in under five minutes
- 83,000
- borrowers in a full portfolio risk analysis
- 27%
- of a live lending pipeline declined by our pre-lending screen
Amber flags fire against your own thresholds, typically weeks before the ledger misses a payment. We do not publish a fixed prediction window, because the right window is the one your own policy defines.
Backed by
Nairobi International Financial CentreCertified — July 2026 cohort- ISO 27001-aligned
- GDPR and Kenya DPA 2019 compliant
- IFRS 9 aligned
Start with your own loan portfolio.
A CSV export, a backtest, and an honest read of what our screen would have flagged and when. No integration, no commitment.
See what your own book looks likeOr just talk to a founder — support@valrcapital.co