One decision engine, from first lead to final payment

An essential part of every lending product

Selling online is in many ways like selling face to face. You learn as much as you can about the customer and ask the right questions. You adapt as you learn more, and you make consistent, accurate decisions.

Many decision engines were designed before banks sold most of their products online. They did not need to take in large amounts of varied data and respond instantly. Zoral Decision Engine, part of the Zoral Finance Operating System (fOS), was designed for this from the start.

It takes in application, bureau, financial, document and third-party data, plus device data for fraud checks. It adds to that data as the journey goes on. Your credit team defines the rules and decision workflows without writing code. The engine applies them in real time, and adapts the journey as the customer’s profile develops. It also automates work that used to be manual, which lending products need if they are to grow without costs growing at the same pace.

The decisions are simple to state. Making them well is not.

When you lend, you make many decisions, from first lead to final payment:

• Do we want to pay for this lead?
• Is it likely to become a profitable customer?
• Which product should we offer? If they do not qualify for product A, should we offer product B?
• What is the risk, and how should we price it?
• Should we lend, and how much?
• Are they who they say they are? Is this fraud?
• Will they pay on time? Will they renew, or leave?
• What is their likely lifetime value?
• What is the next best action for this customer?

Zoral Decision Engine answers these questions in real time, and drives the customer journey across every touchpoint. But lending is more than a series of scores or yes-or-no answers. The decisions have to be balanced together. Your team sets the targets, and the engine balances the factors to meet them across the portfolio.

Things change

Marketing changes who arrives at your website. Competitors respond. Fraud methods, regulation and data quality change, sometimes suddenly. Rules and models that worked last quarter may not work today.

So the engine has to make change safe and fast:

• your credit team changes rules, thresholds and decision workflows themselves;
• new workflows can be added for special cases or new products;
• the bank’s own models, and models from the Zoral model library, run inside decision workflows;
• models are monitored in production; when results get worse, your team is alerted, retrains the model in the test environment, and the new model goes live only after approval;
• every change is tested in separate development, test and production environments.

For technical teams:

• extend data sources and register new data providers;
• define how data passes between workflow steps;
• create and register workflow versions;
• build dashboards and reports, or connect your BI tools;
• set scaling, resilience and deployment settings;
• manage users and detailed permissions.

Built for business users, at very high volumes

Your credit team writes rules in a form people can read, or draws them as a diagram, without needing IT. The engine works out the most efficient order to run rules and data calls. Your team does not have to tune the logic by hand.

Zoral Decision Engine handles each request as it arrives, or in small batches. It works in real time, at very high volumes, and shares the work across servers automatically. In production, it makes up to 5,000 credit decisions a second, around the clock, for a large online retailer’s pay-later checkout.

The engine runs as a set of services, each with a specific job. The services run independently, so a fault in one has little effect on the rest. Rules, workflows and settings are held in memory for speed. Data providers and model services can be scaled separately, and models can run in Python, Spark, R, PMML or C++.

One canvas for rules, models and tests

Zoral Decision Engine has one authoring canvas. There, business users and data scientists define and monitor business rules, models, workflows, settings, permissions, data sources, tests, audits and reports. Many people can work in it at once.

The engine works alongside your existing systems: core banking, payments, internet banking, CRM, loan management, collections and more.

Test before you change

The workflow tools are simple for non-technical users, and the engine handles the complexity. Shadow testing is one example. A new version of a workflow runs alongside the live version, without affecting real decisions, so your team can compare the results first.

Checks you can rely on

The engine runs the data checks that straight-through processing needs: identity and KYC, company, card, bank details, employment, personal information, location, documents, financial statements, blacklists and device data. Your team can define alternative checks, so a decision does not stop when a data provider is down or a result is unclear. The design also catches serial, linked and velocity fraud.

Every decision is recorded with its data, rules, scores and workflow version, and explained in plain English. Supervisors and auditors can trace any decision back to its source.

Feature summary

Where it fits

Zoral fOS has one centre. Zoral Data Fabric remembers every application, decision and change. Zoral Decision Engine makes every decision. The Business Outcomes Layer measures every result against plan. Around that centre, the other components do the work. Every Zoral solution reaches out from it into a product or a channel, and draws on the same data, rules and measurement. So each new product costs less to launch, and starts smarter, than the last.

Zoral Decision Engine sits at the centre of Zoral fOS. It works most closely with Zoral Data Fabric and the Business Outcomes Layer.