Kita Risk Score
Vision AI reads any file.Alternative credit signal, extracted.Scored into your model.
Kita’s vision models look at whatever the borrower sends, a PDF, a screenshot, a phone photo of a receipt, and read the page the way a credit analyst would. What comes back is alternative credit signal as numeric scores, ready for the model you already run.
Verde Logistics S.A. de C.V.
Thin-file borrowers rarely arrive with clean, predictable data.
Thin files stay thin
Third-party data does not cover every borrower. Telco and device data require a mobile footprint, while open-finance data requires a financial account. Globally, 86% of adults own a mobile phone and only 80% have a financial account, leaving qualified borrowers outside the reach of existing models.
Unexpected uploads are untapped signal
Real applications contain collateral photos, screenshots, unexpected images, partial records, and files that do not resemble a standard financial document.
Documents are a universal data source
Every borrower sends something. Kita scores whatever arrives, in any format, turning content, visual structure, and application context into features your model can consume.
Uncover the risk signals buried in every file.
Kita reads financial content, layout, and capture quality in whatever format it arrives, then turns those observations into numeric scores tested against real repayment outcomes.
Cash flow, income stability, employment formality, affordability, debt burden, and liquidity scored from the evidence inside the document.
Image quality, layout, consistency, and capture patterns become measurable predictors instead of review noise.
Whether a file matches what the borrower was asked to submit and whether the submission is reasonable in its application context.
Raw signals spanning financial capacity, document fidelity, and upload intent, pulled from any file a borrower sends, structured or not.
Signal categories combined into numeric repayment scores that return over API as model-ready features.
See what Kita's 8,000-microloan test revealed.
The backtest paired borrower-uploaded financial documents from a global microlender with bureau scores and observed repayment outcomes, allowing Kita to measure the incremental predictive value of document signals against existing credit data.
Read our backtest8,000 self-reported borrower uploads spanning bank statements, payslips, bills, IDs, photos, and screenshots. No template required.
524 raw signals extracted blind from financial content, document structure, visual quality, and layout.
Every signal regressed against the training outcome set. More than 25 were meaningfully predictive of repayment.
Every score judged against the lender's real repayment record.
Kita predicts repayment more accurately than credit scores.*
The score creates a monotonic risk split: as the document score improves, observed default rates decline.
A 15% relative improvement in predictive power over the bureau score alone.
A +0.036 AUC lift and 57% relative improvement over the bureau score alone.
Information Value by signal category
Financial signals lead, but non-financial uploads also carry measurable signal.
Capture-fidelity score
linear R² 0.97
* Based on Kita's historical backtest of 8,000 microloans from a global microlender, measured against bureau scores and observed repayment outcomes. See the case study to learn more.
Lower risk.
A larger addressable book.
Expand your offering
In thin-file markets, lenders must keep credit limits low to manage risk. By analyzing any borrower upload, Kita gives microlenders the evidence to offer larger limits and new products safely.
- Offer additional loan products
- Upsell customers with model-driven limit increases
- Write larger-ticket loans at the same risk
Sharper risk ranking
Kita processes any file upload in real time, adding incremental predictive power that meaningfully splits repayment risk.
- Set starting limits from verified capacity, not a default
- Fewer defaults at the same approval rate
- More approvals at the same risk
Frequently asked questions.
01What file types can you score?
Any of them. PDFs, images, photos taken on a phone, screenshots, spreadsheets, scans, and files that match no template at all. If a borrower can upload it, Kita returns a score for it.
02Does the score replace our third-party data or existing risk model?
No. It is an input, not a replacement. Kita returns numeric scores you feed into the scorecard or model you already run, adding incremental predictive power alongside third-party and first-party data.
03Can the engine support real-time decisioning?
Yes. Kita scores files as they enter your application flow and returns them over API in under 15 seconds, so your existing decision engine can consume them immediately.
04How do we know it will work on our own portfolio?
We begin with a historical backtest against your own files and repayment outcomes. You see the incremental lift, score stability, and explainability before the scores go into production.