Atelier consulting · Credit risk and decisioning

Experienced credit judgement, applied consistently and in real time.

Most lenders have sound credit policy and capable credit teams. Avoidable losses often come from the gap between them and the decisions actually made: policy rewritten by hand as system rules, scorecards built for an earlier book, overrides nobody measures. Zoral’s credit specialists apply the five Cs of credit, with modern models, to your products, deals, portfolios, customer segments and markets. We turn your risk appetite into decision strategies, build and validate the models with your risk team, and run them in real time, with every decision recorded alongside its reasons.

Book an executive briefing Book a credit model review

Credit losses are usually decided long before they are reported.

Risk appetite is agreed in committee, then rewritten as system rules the committee itself cannot read.
Scorecards built for an earlier product, channel or economy are still deciding, and are reviewed less often than the book changes.
Overrides are allowed but not measured, so nobody knows whether they add profit or losses.
Warning signs show in portfolio data well before they show as arrears, but nobody owns the response.
Each policy change is argued on opinion, because there is no clean measure of what the last one did.

A better model on its own rarely fixes this. Judgement and models have to work together, inside a process that records what each decision was based on and what each change achieved.

The five Cs still decide. The evidence behind them has changed.

The questions experienced credit officers have always asked, answered with models and data wherever the answer matters: a product, a single deal, a portfolio, a customer segment, a geography or a regulatory framework.

Character
Repayment record from bureau data and your own book, identity confirmed at source, early signs of fraud or first-payment default.
Capacity
Verified income and commitments, with affordability assessed under your policy and tested against changes in rates and income.
Capital
Savings and net worth for individuals; balance sheet, cash flow and leverage for businesses.
Collateral
Valuation, loan-to-value and enforceability, followed over the life of the facility, not only at the point of lending.
Conditions
The sector, region, rates and economic cycle around each borrower, and the rules of the market you lend in.

Models make these judgements consistent and fast. Judgement decides what the models measure, where they stop, and what happens at the edges.

From risk appetite to a decision on the record.

Policy into decision strategies
Your credit policy and risk appetite, translated into eligibility, cut-offs, referral rules, limits and pricing by product, channel and segment, and written so your credit committee can read them before release.
Scorecards and machine-learning models
Application scores; behaviour scores based on how existing accounts are run; fraud and first-payment default; affordability, limits and pricing. We start from the model library in the Zoral Intelligence Layer and tune it on your data, or work with your data scientists on the models they own.
Validated before it is trusted
Each model is tested on out-of-time samples, measured with standard measures such as AUC and Gini, checked for stability and drift, and set against the scores you rely on now. It then runs in shadow or A/B mode on live applications before you switch.
External review of your scorecards
An external review of the scorecards and models you use today, alongside your validation team, with observations, an assessment of each risk and practical recommendations. We ask whether each model is fit for purpose today, and fit for the plan you are lending against.
Portfolio early warning
Risk indicators by product, segment, sector and geography, each with a threshold, an owner and an agreed response, so a signal leads to action before it becomes arrears.
Decisions in real time, on the record
Strategies and models run in Zoral Decision Engine, part of the Zoral Finance Operating System (fOS), in real time at very high volumes. Every decision is recorded with its data, the policy and model versions that applied, and its reasons.
Every policy change measured
The Business Outcomes Layer shows what each change to policy, cut-offs, limits or pricing did to approvals, conversion, losses and margin, by product and channel, against your plan. The next policy debate starts from evidence, not opinion.

Fraud kept apart from credit risk.

When fraud is booked as default, losses are misread and credit models learn the wrong lessons. We build fraud and identity controls into the same decision flow, so credit risk is judged on genuine borrowers:

First-party and third-party fraud identified separately from credit losses, so each is managed for what it is.
Identity, application data and documents checked at source and for alteration, with honest mistakes told apart from deception.
Early warning of organised fraud, with responses agreed in advance, from closer checks to pausing a channel.
Checks routed by risk, so only the cases that need a person reach one, and good customers are not kept waiting.

Where your regulator and the customer’s consent allow, what identity checks reveal also helps predict default.

How a decision reaches production with your committee’s approval.

Our credit models are built with Zoral’s data-science-driven MVP methodology. For a credit decision it runs like this, with your risk, data science and compliance teams involved at each step:

01
Start from your book
Workshops on your products, customer lifecycle, credit policy, KPIs, and the models and data you already use. We agree the decision to improve first and how success will be judged.
02
Close the predictive data gaps
We test the quality of the data behind each decision, then look for consented, lawful evidence it lacks, or collects and leaves unused. In credit, better evidence is often worth more than a new algorithm.
03
Prototype against your current decisions
Strategies and models are prototyped in a test environment, then run in shadow or A/B mode beside your current decisions, showing what would change in approvals, losses and margin.
04
Review and approve
Documentation of purpose, data, development, validation, limitations and monitoring is reviewed with your risk, data science and compliance teams before release and before any retuning. Your model risk process and credit committee decide.
05
Run, monitor and retune
Agreed thresholds on model performance, stability and the early performance of new lending trigger a review. Recalibration follows the same route.

See how we build and run decision models

The team that implements Zoral’s decisioning, and works on yours.

Atelier implements Zoral fOS and its solutions, so the specialists who design a credit strategy also configure it in Zoral Decision Engine, connect it to your bureaus, documents and core banking system, and measure it in the Business Outcomes Layer. This work sits behind Zoral’s solutions for underwriting, loan origination, portfolio optimisation, alternative finance and credit cards. If you run another decision engine, or your scorecards live on your own data science platform, we work there, reviewing, rebuilding or retuning what you have. Some credit engagements use no Zoral product at all.

Underwriting Automation · Loan Origination · Portfolio Optimisation · Alternative Finance · Credit Cards · Customer lifecycle and collections

Where credit engagements usually begin.

Credit model review
One or more of your scorecards, reviewed with your validation team. It ends with recommendations ranked by their value to the business.
Policy-to-decision assessment
We trace one product from its risk appetite statement to the decisions actually made, overrides included, and show where policy and practice have drifted apart.
One decision, improved in production
One decision, such as limits for existing customers, taken from data review through shadow testing to live use, measured against plan.
Portfolio early-warning framework
Indicators, thresholds, owners and responses for one portfolio, agreed with your risk team and run from your own data.

Under your chief risk officer, alongside your validation team.

We work under the direction of your chief risk officer and credit committee, alongside your model validation function and the advisory firms and systems integrators you already use. Our review adds to your validation; it does not replace it. Every decision on a model or strategy stays with your bank.

Founded in 2004, Zoral has delivered over 100 projects, with 350+ years of combined experience. We have worked with clients regulated by the FCA in the UK, the CFPB in the US and the central banks of Malaysia and Thailand, and built fraud, identity and document-forgery models for lenders on several continents.

Frequently asked questions

Judgement and control

Do you replace our credit team’s judgement?

No. Your credit policy sets which applications are decided automatically and which go to a person, and your lending authorities stay as they are. Our work applies your team’s judgement consistently and frees your credit officers for the cases that need them. A customer who asks for a person to review an automated decision can be routed to one.

Will faster decisions mean riskier ones?

They should not. Speed comes from removing manual steps, not from loosening policy: clear cases are decided within the policy you approve, and the rest are referred with their reasons. Each change is shadow-tested before release, and new lending is tracked against plan, so any weakening shows early.

Do AI agents make credit decisions?

No. Credit decisions follow the policy, scorecards and models your bank approves, and accountability stays with your bank. AI agents can prepare work for your credit teams, such as reading documents or drafting a case summary, under guardrails and human-in-the-loop supervision. They act only through the permissions, workflows and approvals you set, and what they do is recorded.

Is your review independent of our validation function?

It is an external review alongside your validation team, not a substitute for it. Where your rules require independent validation, your validation function remains responsible, and our observations and evidence are available to it.

Models and data

Can you work on our existing scorecards and decision engine?

Yes. We review, retune or rebuild scorecards where they run today, and configure strategies in the decision engine you already use. Where Zoral Decision Engine is in place, your existing models can run in it, or stay on your own platform and send their scores to it.

How do you handle thin-file customers?

With consented, lawful evidence of capacity and conduct that a bureau file lacks: verified income and commitments, bank statements or open-banking data where the customer agrees, and your own relationship data for existing customers. Affordability stays part of every decision. Where evidence is limited, a strategy can start with a lower limit and review it as the account is run, subject to affordability.

How do you avoid unfair bias?

By designing for fairness from the start, with your compliance team. Model inputs are agreed and documented, characteristics the law protects are excluded where the law requires, and variables that could stand in for them are reviewed. Each decision carries its main reasons, so it can be explained to the customer and examined by your teams. Where your bank tests outcomes across customer groups, we provide the data and documentation the test needs.

Does your model library carry other lenders’ data?

No. The library holds model designs that have worked in earlier deployments, not another client’s data. Each model is tuned on your own data, under your governance.

Which credit decision worries your committee most?

It may be a scorecard that has drifted, a product where losses came in above plan, or a segment you would like to lend to but cannot yet price. A senior Zoral credit specialist will look at the evidence with you and agree a first step.