Make credit decisions on checked documents, not retyped ones.

OCR turns a page into text. A lending decision needs to know what the text means, whether it agrees with the other documents and whether the document has been altered. Zoral does that work and passes the result into the decision. You can keep your existing scanning and upload channels. Zoral takes documents from them.

Zoral’s models are trained to understand financial documents, not to follow a fixed template for each layout. A new payroll format or a skewed photo is usually still recognised as the same document type. Where confidence drops, the field goes to a person instead of passing silently. You can see confidence and review rates by document type, and you set the thresholds.

Accuracy depends on your document mix, image quality and market, so we don’t quote one number. Instead, we measure it on your own documents, field by field, against results your team has already checked. You see where it is reliable, where people still need to review, and decide on that evidence.

It flags signs that a document may have been altered, or that it disagrees with the rest of the file, and gives the reason for each flag. Your risk team sets the checks for each product and country. Your fraud team decides whether a case is fraud. Every flag and outcome is recorded.

Zoral Document Fabric runs in your own AWS or Azure account, or your private cloud, in your region and under your control. Retention can be configured to your policy. Our controls are independently audited under SOC 2, with details on the Zoral Trust Portal.

Zoral combines its own models, trained over years on financial documents, with language models where they add accuracy. You choose those models, from enterprise model services to open-weight models hosted in your own environment. Your knowledge bases, conversation records and client data stay under your governance, within the environments and agreements you control. Any use of your documents to improve Zoral’s own document models is agreed with you in writing. You remain the controller of your customers’ data. Every extracted value links back to its place on the page, with its confidence score and any human correction. AI agents, configured to your policies and workflows, can assist your reviewers inside the screens they already use. Each has its own guardrails and approval steps. They act only through the permissions, workflows and approvals you set, and every prompt, tool call and answer is traced for review. They do not make credit decisions.

No. It works best inside Zoral fOS, where verified data goes straight into decisions, work queues and customer journeys. It can also feed your existing loan origination system, decision engine or document store through documented interfaces, including widely used mortgage loan origination systems. You choose where the data goes.

Yes. Your agents connect to Zoral fOS through a standard interface (MCP) and can read extracted data, run checks or start workflows only within the permissions you grant. Guardrails, budgets and human approval steps are set per agent, and every action is traced for review. Your analysts can also set up document checks with an AI assistant, as our specialists do with AI for Zoral. Each change is still versioned and approved by a second person before release.

The library covers 650+ document types and 14,000+ data fields: identity, income and employment, bank statements, tax returns, property, insurance, company accounts and their notes, and trade invoices. Models are trained and used in North America, Europe and ASEAN, with documents in English, Malay and Thai today. For each country, document type and language you need, Zoral trains the models and configures the document workflow, through a tested, approved release.

Zoral’s specialists maintain the models and add new document types with you, from your sample documents. Each change is tested against known answers and released as a new version only after approval. Your analysts set which documents each product needs and which checks to run.

Every document, extracted value, check, flag and correction is recorded with who did what and when. That gives you evidence for credit files, KYC and anti-money-laundering reviews and audits. Zoral is designed to support GDPR and UK GDPR obligations, including data minimisation and regional processing. The obligations stay with your bank. Zoral keeps the record you need to meet them.

With a proof of concept on your own sample documents, chosen by you, measured against results your team has checked. Samples are anonymised, or a data processing agreement is in place before any customer documents are shared. We set out a realistic plan for production only after you have seen the results.

Zoral Document Fabric runs in your own AWS or Azure account, or your private cloud, in your region and under your control. You choose who runs it, and the rest of Zoral fOS: Zoral can run and support the platform for you, your own technology team can run it, or we share the work and hand over as your team grows. Your people review the documents and make the decisions.

Atelier, Zoral’s consulting and implementation team, works with your teams and, where you choose, with your advisers and systems integrators. The same team can bring the data Zoral Document Fabric extracts into your data lake and into the models that use it.

Data, AI & Automation