All work SoyakaAI · Case study
We built the AI credit platform behind SoyakaAI.
Lenders drown in manual review. SoyakaAI's Credit Studio turns an application into an instant assessment, a plain-language explanation and an offer that fits, in seconds instead of days.
Under the hood
- Assess · match · explain workflow
- Knowledge base per lender
- Approvals above threshold
- Model routing per agent
- Bureau and open-banking connectors
- Run log of every decision
01 The challenge
Turn a credit model into decisions a bank can use.
SoyakaAI had a strong credit-risk model. It needed to be something a credit team can read and a bank can try. We turned the model into a platform: personal finance, auto lease, real estate, microfinance and SME lending, each run on the lender's own rules.
02 What we built
Credit Studio, run on the lender's own rules.
Each decision runs as a workflow: assess, match, explain. Every lender's credit rules sit in a knowledge base, so each offer is tested against that institution's policy.
- AI credit decisioningReads the application, bureau and financial data, applies the lender's rules and returns an assessment, a risk memo, a customer explanation and a ready-to-send offer.
- Instant offersIf the request doesn't fit but the customer qualifies, the matchmaker finds the largest amount that does, plus an alternative.
- Open banking intelligenceTurns bank transactions into verified income, affordability and risk signals.
- AI decision advisorRisk, commercial and customer teams ask questions and get answers grounded in that specific decision. It never approves anything itself.
- Admin portal and open APILoans, applicants, models, workflows and the rules engine behind one API, with embeddable widgets that include an AI chat.
- Arabic document pipelineDetects the document type, reads it, checks it against rules and returns structured data to a credit-memo agent.
- Model routingDocument reading, credit memos and the decision advisor each run on their own model, routed by task, cost and the lender's data requirements.
- Audit trailEvery decision is kept in the run log, with its inputs, the rules applied and its reasons.
03 How it works
How a decision runs
One application, from request to offer. Illustrative figures.
- Request: SAR 250,000
- Debt burden 54.5% against a 45% ceiling
- Instant assessment
- Matchmaker: SAR 193,000 within policy
- Adjusted to SAR 160,000
- Instalment SAR 3,227, debt burden 39.4%
- Risk memo, explanation and offer by SMS
04 Guardrails and control
The lender decides.
- AuthorityThe lender approves. The platform assesses and recommends. Approval stays with the institution.
- ThresholdsApprovals above the line. Decisions above the lender's thresholds wait for a person to approve them.
- RulesThe lender's rules, not ours. Every offer is tested inside the institution's own policy before anyone sees it.
- ReasonsEvery outcome has reasons. The factors behind each decision go to the credit team and, in plain language, to the customer.
- DeploymentRuns where the bank needs it. Through the API, a hosted page, on the bank's own servers, or handed over as a working module.
05 Outcome
Proven on real lending decisions.
SAR 2B+of Saudi lending decisions validated
60+financial institutions in its pricing intelligence
10lending model types, from application scoring to fraud
Decision and institution figures as published by SoyakaAI.
06 Stack and channels
What it runs on.
- Interfaces
- Admin portal, open API, widgets
- Channels
- SMS offers, AI chat
- Deployment
- API, hosted, on-premise or module
- Documents
- Arabic pipeline