Every member question, answered by their own institution
Members ring the office to ask what is in their FOSA account, where a loan application stands, or why a limit changed. wekanode answers those calls out loud, from your own ledger, and shows the entries it read. Your staff put the same questions to the whole book from a console. The credit models run underneath it.
Members can reach their money. Reaching an answer is the hard part.
The account works. The statement arrives. What is missing is anyone to explain it — at the hour the member is free, in the language they think in, with the ledger open.
What members run into today
- A branch queue or a call centre that is open only while the member is at work
- Statements written for an auditor, handed to a member
- Decisions delivered as a yes or a no, with no reason attached
- A bureau file that most adults in these markets have never had
What wekanode does instead
- A member asks in their own words, out loud, and gets an answer
- Every figure traced back to the ledger entry it came from
- The reasoning behind a decision, in the same words for the member and the auditor
- Scoring models underneath, so a thin file is not an automatic no
“How much is in my FOSA account, and why did my loan limit drop?” One member, one call. The assistant retrieves the entries that answer it, says so in plain language, and shows its working — to the member now and to the auditor later.
- FOSA ledger24 mo. of entries
- Loan bookschedules & arrears
- Member recordKYC & standing
- Product rulescurrent terms
engine
ANSWER
- FOSA ledger · 12 Aug depositcited
- Loan schedule LN-4471cited
- Limit policy v4 · savings ratiocited
Seven modules. One member record.
The conversation layer is the product; the models sit under it. Every module reads and writes the same member record, so what the assistant told a member on Tuesday is the same thing the audit log says on Friday.
From your ledger to a member's question, in one pipeline.
You keep the member relationship and the final policy call. We handle the part between your records and an answer the member — and the regulator — can both follow.
- 1
Connect your core
FOSA ledger, loan book, member records and product rules, through one API or a batch drop. Consent is captured and logged at the source.
- Core banking connectors
- Batch CSV
- Consent ledger
- 2
Ground the model
The assistant answers from your record and nothing else. Anything it cannot retrieve and cite, it does not say — it escalates to a person instead.
- Retrieval over the ledger
- Citations required
- Human handoff
- 3
Answer and act
A member asks by phone; a teller asks from the console. Balances, statements, loan status, next repayment — and the scoring models underneath when a decision is needed.
- Voice and text
- Staff console
- Per-market models
- 4
Explain and improve
Every answer carries its sources and every decision its ranked reasoning. Transcripts, retrieval traces and model versions are retained for regulatory review.
- Full audit trail
- Drift alerts
- Versioned models
Same engine. Different book.
A SACCO underwriting against member savings and an agri-lender underwriting against a harvest need different models, not different vendors.
For microfinance institutions
Your loan officers spend their day answering the same questions in person. wekanode takes the routine ones by phone — balances, schedules, where an application stands — and gives officers a console for the rest, so the branch network stops being the bottleneck on growth.
Talk to us about microfinance- Routine member queries answered by voice
- Loan officer console over the whole book
- Group and individual lending models underneath
It cites its sources, or it says it cannot.
A model that guesses at a member's balance is worse than no model at all. Everything wekanode says to a member is retrieved from your record, retained with the trace that produced it, and retrievable when someone asks.
Grounded, never improvised
The assistant answers only from records it retrieved and can cite. Asked something outside them, it says so and routes the member to a person — it does not fill the gap.
Every conversation is on the record
Transcripts, the retrieval trace behind each answer, and the model version that produced it are all retained and queryable — the same evidence a member complaint would need.
Explainable by default
Every score returns ranked reason codes. Adverse action notices are generated from the same payload the model used — no post-hoc rationalisation.
Data residency you choose
Deploy in-region or in your own VPC. Member data never has to leave the jurisdiction that regulates it.
Fairness testing built in
Disparate-impact metrics run on every model version across gender, region and age bands, with results in the audit log — not a slide deck.
Encrypted end to end
TLS 1.3 in transit, AES-256 at rest, field-level encryption for PII, and scoped API keys with full request-level audit trails.
One endpoint. An answer you can trace.
REST and gRPC, idempotent by default, with a sandbox that returns deterministic fixtures so your tests don't depend on our uptime. Every response carries the records it was built from.
- Voice and text on the same session, with the transcript returned
- Every answer ships the retrieval trace that produced it
- TypeScript, Python, Go and Java SDKs
- Sandbox keys with seeded member fixtures
- Versioned endpoints with a published deprecation policy
import { wekanode } from "@wekanode/sdk"; const wekanode = new wekanode(process.env.wekanode_API_KEY); // Answer a member, grounded in the institution's own recordconst reply = await wekanode.answer({ memberId: "mbr_8f21c04b", channel: "voice", utterance: "Salio langu la FOSA ni ngapi?", ground: ["fosa_ledger", "loan_book"], escalateIfUncited: true,});Ready to lend smarter?
Bring us a sample of your declined applications. We'll run them through the engine in shadow mode and show you which ones you should have funded — before you commit to anything.
- 30-minute technical walkthrough
- Shadow-mode pilot on your own portfolio
- No integration required to evaluate