Credit risk & decisioning · Dubai · Riyadh

Lending decisions made on expected value.

NYK Advisory works with banks, digital banks and finance companies across the GCC on the three places a retail book makes or loses money — who you acquire, who you approve, and what you do with the customers you already have.

Consumer credit — cards, personal finance, BNPL Conventional & Shariah-compliant UAE · KSA
Three decisions, one thresholdSchematic — not measured data
Three decisions, valued rather than ruled Schematic, not measured data. Three decision points in sequence: acquire, approve and manage. Each opens into a fan of options placed by expected value, with the highest-value option marked. A single dashed policy-rule threshold runs across all three; at the manage stage every option falls below it, including the best one. Two bands beneath are labelled decision economics and decisioning infrastructure. ↑ higher expected value 01 · Acquire 02 · Approve 03 · Manage policy rule — one threshold best option, below the rule Decision economics · NPV · test-and-learn · pricing Decisioning infrastructure · execution · data · models in production

Every decision is a set of options with different expected values. A single policy rule applies one threshold to all of them; a valuation takes the best available at each point — including where the threshold would have said no.

The problem

Most retail books here are still priced in aggregate and decisioned on rules.

That was survivable while growth covered the losses and the addressable population was thick-file. It is not survivable now. Every new entrant — digital banks, finance companies, BNPL — is chasing the same thin-file customer, and the bureau cannot see that customer clearly.

The techniques that resolve this are not new. NPV models for decisioning, risk-based pricing, disciplined test-and-learn, cash-flow underwriting and continuous portfolio action have been standard practice in mature consumer credit markets for two decades. They are still largely absent here.

That gap is the whole of our work. Not a transformation programme — the specific, measurable decisions where a lending book is leaving money on the table or taking losses it cannot see.

Approach

Six things we believe, and will be held to.

These are not values. They are commitments a client can check us against while the work is running — and, increasingly, the same places the regulator is looking.

01

Value the decision, don’t rule it

Every approval, limit, price and treatment has an expected value that can be estimated before it is made. A policy rule is a decision taken once, in the past, for a population that has since changed. Wherever the data allows it, we replace rules with valuations.

CBUAE Model Management Standards (Notice 5052/2022) apply to all models supporting decision-making, pricing models included. A credit policy rule set is a decisioning model: it belongs in the inventory, it is tiered for risk, and it is validated — whether or not anyone has ever called it a model.

02

Nothing counts without a control group

An intervention not measured against what would have happened anyway is a story, not a result. We design the control before the programme runs, and we report what the control says — including when it says our own work did not pay.

CBUAE MMS requires ongoing performance monitoring against defined metrics and limits, escalation, and reporting to a Model Oversight Committee at least quarterly. It also names “cherry-picking, reverse-engineering and positive bias” as grounds for supervisory action. A clean control group is the cheapest defence against all three.

03

Built for this market, not translated into it

Credit models carry the assumptions of the market they were built in. Here the dominant factors are different: employer concentration, income that stops discontinuously along with residency, a bureau that is thinnest exactly where the growth is, and structures that must be Shariah-compliant by design rather than by adaptation. Those are not adjustments to an imported model. They are the model.

CBUAE AI & ML Guidance Note (11 February 2026) requires training data to be “sufficiently accurate, relevant and representative of the customer populations to which the models will be applied.” A vendor model trained on another market’s population does not clear that bar by default, and the burden of showing it does sits with the institution.

04

Parameterised and production-grade

What we build is meant to run, not to be presented. Thresholds, weights and segment definitions are exposed as parameters your own team can move, documented well enough to know what moves when. If changing it requires us, we built it wrong.

CBUAE AI & ML Guidance is explicit that institutions “should not employ AI models that they have no control over,” and must retain the immediate ability, with human intervention, to stop using one. MMS adds documentation sufficient for an independent party to assess it, plus archiving for rollback. Control is now a requirement, not a preference.

05

Explainable to the customer, not just to the committee

A decision you cannot explain in a sentence to the person it was made about is a decision you do not fully understand yourself. Explainability is not a reporting layer added at the end — it constrains what you build, and it is worth the constraint. It is also the single least-prepared-for obligation in this market right now.

CBUAE AI & ML Guidance (Feb 2026) requires that consumers can request human review or an explanation of an AI-generated decision, with alternative arrangements for customers who do not want an AI decision at all; disclosures in plain language in Arabic and English; annual bias testing; and human oversight calibrated to risk. Most decisioning stacks here were not built to answer that.

06

Loss avoided is not value created

A credit function measured only on losses will produce a book that is smaller and safer than it should be, and will lose the customers worth having to whoever prices them properly. The objective is the value of the portfolio, and often that means approving more.

Practice

Three places money moves, and the two disciplines beneath them.

Each area stands alone as a mandate. Most books need more than one, and the sequence usually reveals itself in the first four weeks.

01Acquisition economics 02Underwriting & credit risk 03Portfolio management & customer value 04Decision economics & risk-based pricingNPV · test-and-learn · pricing 05Decisioning infrastructure & applied AIExecution · data · models in production
01

Acquisition economics

Pay for booked accounts, not for leads.

Channel design and economics. Pay-for-performance structures for DSAs, brokers and digital partners, built so the payout ladder tracks credit quality rather than application volume. Pre-qualification and lead scoring at the top of the funnel, so acquisition spend follows applicants you will actually approve. Full-funnel attribution from spend through to a performing account.

What it moves
  • Cost per booked, performing account
  • Approval rate by paid channel
  • Channel contribution after loss
  • Spend on unapprovable volume
02

Underwriting & credit risk

Approve more of the customers you are declining for the wrong reasons.

Credit policy design and rebuild. Application and behavioural scorecard development and validation. Cash-flow underwriting on open-finance data for thin-file and new-to-credit segments. Independent assessment of bureau performance on your own portfolio, measured forward-looking on a fresh application population rather than on headline statistics. Portfolio diagnostics and vintage analysis.

What it moves
  • Approval rate at constant loss
  • Forward-looking discrimination
  • Loss rate by vintage and segment
  • Size of addressable population
03

Portfolio management & customer value

The book you already own is the cheapest growth available to you.

Line increase and decrease programmes run continuously against risk and utilisation rather than as an annual bulk exercise. Proactive repricing. Activation and utilisation campaigns for dormant and low-engagement accounts. Retention and offer design against customer lifetime value rather than product margin. Attrition propensity modelling. Collections strategy driven by propensity to pay and to cure, rather than by days-past-due bands and a uniform treatment ladder.

What it moves
  • Incremental balances per programme
  • Activation and utilisation rate
  • Attrition in the profitable decile
  • Roll rates and cost to collect
04

Decision economics & risk-based pricing

Value the decision before you make it.

NPV models for account-level decisioning — the discipline of valuing a decision rather than approving to a cut-off — applied at origination, at every limit and price change, and at the point of a retention offer. Risk-based pricing frameworks, including the transition path away from a single aggregate price. Champion-challenger and test-and-learn design: the measurement architecture that makes a result readable, and the governance that lets a test run long enough to mean anything.

What it moves
  • NPV per account decision
  • Margin at constant volume
  • Decisions valued, not ruled
  • Learning per test cycle
05

Decisioning infrastructure & applied AI

A policy you can change in a week, not a quarter.

Decision engine selection, specification and implementation oversight. Open banking and open finance integration, and the attribute layer that turns raw transaction data into decision-ready features. The execution capability to action a portfolio programme at scale and measure it against a clean control. Model deployment, monitoring and refresh. AI applied where it earns its place in a credit business — income and document verification, credit operations, collections treatment, agent assist — rather than as a programme of its own.

What it moves
  • Time to change a credit policy
  • Straight-through processing rate
  • Model refresh cycle time
  • Cost to serve per application
Engagements

Scoped, time-boxed, answerable to a number.

We work as practitioners inside your team rather than as a parallel programme. Every engagement is defined by the decision it changes and the metric that should move as a result.

Four weeks

Portfolio diagnostic

Where the book is leaking, and by how much. Acquisition, approval, portfolio action and collections examined against expected value, with a ranked set of interventions and the size of the prize on each.

Ten to fourteen weeks

Portfolio action programme

A line management, activation, repricing or retention programme designed, tested against a control group, measured and handed over as a repeatable cycle rather than a one-off campaign.

Eight to twelve weeks

Model or policy build

A scorecard, an NPV decisioning framework, a pricing structure or a rebuilt credit policy — developed, validated, documented and handed over with the transfer of method, not just the artefact.

Six to nine months

Zero to first loan

Standing up the credit function of a new lender. First-generation credit policy, build versus buy on the decisioning stack, vendor selection, data and bureau strategy, and the shape of the team you need in year one.

Retained advisory runs alongside any of the above, at CRO, chief data officer or founder level — monthly cadence, access between cycles, and a second opinion on the decisions that are hard to reverse.

Point of view

What we think is wrong, and what is coming.

Four arguments that inform how we work, each written up as a paper. Three are published; the fourth follows before the end of the year.

I

A bureau score is built to rank everyone. No lender lends to everyone.

A bureau score’s measured performance is a property of the test it was measured in, not only of the score. Measured the way a lender actually uses it — forward, on the population that applies, over the horizon the exposure runs — the same score is answering a harder question, and answers it less well. Paper I sets out why, and the four levels a credit function can run to measure the score on its own book.

Read the paper →
PublishedSeptember 2026
IV

The back book is where the unclaimed money is.

Almost nobody in this region runs continuous portfolio action — line management against live risk, propensity-driven collections, retention priced on lifetime value. The usual objection is that debt burden rules constrain limit increases, and they do constrain their size. They do not constrain the discipline of targeting them, and most books are leaving permitted room unused rather than running into the cap.

Forthcoming

Papers publish through Q4 2026 — get in touch to be notified as each is released. All papers →

The practice

Built by an operator, not a reviewer.

We do one thing — the economics of consumer lending decisions — and we work with a small number of clients at a time, so the person you meet is the person who does the work.

The method comes from Capital One, where scorecards, NPV decisioning, portfolio action and test-and-learn were not projects but the way the business was run day to day. The market judgement comes from twenty years in the region since: running a cards portfolio at a UAE tier-one bank, and building consumer lending businesses here from the ground up, including cash-flow underwriting for segments the bureau does not serve.

That combination matters more than it sounds. Analytical technique without a feel for how these markets behave produces elegant models nobody deploys. Market experience without technique produces confident decisions with no arithmetic underneath them.

Both conventional and Shariah-compliant structures are in scope. In Islamic products the credit question does not change, but the mechanics of pricing, profit and treatment do — and they need to be designed in rather than retrofitted.

  • Capital OneConsumer credit: scorecards, NPV decisioning, test-and-learn, portfolio optimisation
  • Emirates NBD · Emirates IslamicRetail banking and cards, latterly Acting Head of Cards
  • Booz & Company / Strategy&Management consulting, financial services
  • Consumer lending ventures, UAEFounder: cash-flow underwriting for thin-file and new-to-credit segments
  • Greenfield digital bank, KSAAdvisory: target operating model, decisioning and data
  • Consumer businesses beyond lendingSKU, store and price optimisation · working capital · loyalty programme design and lifetime value
  • IIT DelhiBachelor of Technology
Beyond lending

The same decisions exist wherever a customer relationship carries a balance.

Every decision on this page is the same decision in different clothing: what is this customer worth over time, and what should we do about it under uncertainty.

Lending is where those decisions are most developed, most measurable and most regulated, which is why it is where the method was built. It is not where the method stops. Any business that delivers value before it is paid for is carrying consumer credit risk whether or not it has a credit function — and most of them underwrite on rules of thumb, price everyone the same, and collect on a fixed ladder regardless of who the customer is.

The distance between that and a managed book is large, and it is closed more cheaply than most of these businesses expect. The customer-value half of our work — acquisition economics, activation, retention priced on lifetime value, churn propensity, test-and-learn — carries further still, into any recurring consumer business.

This is not theoretical for us. Alongside the lending work we have run a full commercial optimisation for a nutrition and supplements business — SKU rationalisation, store rationalisation, price optimisation and cash flow — and built loyalty programme economics elsewhere. Those are the disciplines on this page under different names: which SKUs earn their shelf space is portfolio optimisation, a store is valued the way an account is valued, price optimisation is risk-based pricing with the risk dimension removed, and a loyalty scheme is lifetime value with an earn-and-burn wrapper. The unit of decision changes. The arithmetic does not.

What does not carry is the regulatory layer above. Those obligations bind licensed financial institutions, and we are careful not to imply otherwise.

  • Postpaid telco and device instalment plans
  • Utilities — deposits, payment arrangements, disconnection strategy
  • Healthcare and education billed in instalments
  • Rental and property — deposit sizing, arrears treatment
  • Merchant BNPL and embedded finance partners
  • Employers and platforms advancing pay or float
  • Any subscription business with churn and a lifetime value
Get in touch

If a number in your book is moving and nobody can tell you why.

That is usually the right moment for a conversation. A first call costs nothing and is generally enough to tell whether there is real work here or not.

hello@nykadvisory.io