AI that reads, explains and waits for a person.
Amanova uses AI to do the reading: bank statements, documents, bureau reports and policy. Every result is a draft until a person accepts it, and every result is kept with the model, the inputs and the reviewer.
What the AI does.
- Reads bank statements: regular income, existing repayments, returned payments and signs that the file was edited.
- Compares documents with what the customer entered, and flags reused documents and shared phones.
- Drafts the credit memo from the file and answers questions about your own credit policy, with references.
- Explains each decision: what helped, what held the application back and what the customer can do.
Eight AI features, one rule: a person decides.
Underwriter copilot
Drafts the credit memo from the file and answers policy questions with references to your own credit policy.
Statement analyser
Finds regular income, existing repayments and returned payments, and checks whether the statement was edited.
Fraud checks
Spots reused documents, shared phones, recycled IDs and income that doesn't match the evidence.
Alternative data
Scores customers with no credit history from wallet, bill and phone data they agree to share.
Reasons, not just scores
Shows what helped and what held each application back, and tells declined customers what they can do.
Early warning
Scores every active loan each night and suggests the next action, channel and a polite message.
Customer assistant
Answers questions about products, instalments, application status and payments on the web, WhatsApp and USSD.
Model governance
Watches approval rates across groups, score drift and officer overrides, and lets you test a policy change on past applications first.
You choose where the model runs.
Built-in only
Rules, scorecards and templates. Nothing leaves your servers.
Inside your network
An open model on your own server, through vLLM or Ollama.
In-country cloud
An OpenAI-compatible service hosted in your country.
Cloud model
Anthropic Claude, OpenAI or Azure OpenAI, with every call logged.
Every AI result is on the record.
- Each result is logged with the feature, the model, the prompt version and a fingerprint of the inputs.
- A person accepts, edits or rejects it. Nothing an AI writes changes a credit decision on its own.
- If the model is unavailable, built-in rules and templates keep every feature working, offline and on your own servers.
- Model governance watches approval rates across customer groups, score drift and officer overrides.
Test a policy before it goes live.
Run a proposed policy change on past applications first and see how approvals, declines and referrals would have moved, by product and customer group.
See it with your own products
We'll set up the demo with your products, your limits and your brand, then walk your credit, compliance and IT teams through it.