There's a very good EY article doing the rounds about modernising risk data. Its argument, made by Jared Chebib and Rohit Garg, is that risk management is shifting from periodic, report-driven work to something continuous, intelligence-led and increasingly “agentified” — and that most institutions can't get there because their risk data is fragmented, poorly governed and impossible to trace. Fix the foundations, the piece says, and you earn the right to deploy AI that monitors risk in real time.
It's right about all of that. But it leaves the most interesting assumption untouched. Throughout the piece, decisioning is treated as a fixed act — the same judgement lenders have always made, just faster, cheaper and better-informed. Modernise the plumbing and the water flows quicker.
I think that undersells what's coming. The plumbing is the easy part to imagine. The harder, more important shift is that the decision itself is about to change shape. In an AI era, what we mean by “a lending decision” expands — in how it's built, what it contains, and how far it reaches. Get that right and it doesn't just make lending faster. It changes what credit is. And the lenders who move first won't just be more efficient — they'll win more good business, run better books, and keep customers the rest of the market can't. This is a competitive move before it's a philosophical one.
What a scorecard has always been.
Strip away the jargon and a credit scorecard is a retrospective. We take a pile of historical borrowers, run regression models to work out which data features predicted who repaid and who didn't, and freeze that relationship into a scoring rule. It's a photograph of the past, used to make bets about the future.
That approach has served the industry well for decades, and there's nothing wrong with the statistics. The limitation is structural. A regression-built scorecard is static — optimised once, then deployed until it drifts far enough that someone rebuilds it. It's backward-looking by construction, because retrospectives can only ever describe borrowers who have already been and gone. And it's narrow: it answers one question — should we lend, yes or no? — and treats everything around that question as separate operational plumbing.
None of that was a failure of imagination. It was the best you could do when modelling was expensive, data was thin, and the only way to generalise from the past was to hold the model still. AI removes those constraints. And once they're gone, three things about the scorecard change.
One: the scorecard becomes fluid.
The first change is that the scorecard stops being a fixed artifact and becomes a living one.
Instead of optimising once against a historical sample, an AI-era scorecard is tuned continuously against how borrowers are actually performing — right now, in the current book, under current conditions. It optimises for two things at once: throughput, meaning how many good customers you can say yes to, and book performance, meaning how that portfolio actually behaves over time. Today those two goals sit in tension, mediated by a cautious analyst who rebuilds the model every so often. In the new world they're balanced continuously, with the scorecard adjusting as the evidence comes in.
That's a meaningful shift in posture. The old scorecard asks, “what predicted repayment among people like this in the past?” The living scorecard asks, “given everything we're seeing across this book today, where is the line that lets in the most good business without degrading performance?” It learns from the present instead of only the past. And the richest signal it learns from is how people are actually paying — which is where payments enter the story.
Two: the scorecard holds rules and questions.
The second change is about what a scorecard is made of.
Today a scorecard is essentially a set of weighted features and hard credit rules — thresholds, cut-offs, policy gates. That doesn't go away. Rules are how you encode regulation, risk appetite and the things you are simply not willing to do. But alongside the rules, the scorecard starts to hold questions.
Not every input that matters can be reduced to a pre-weighted feature. Sometimes the useful thing is an open question — “does this applicant's income look stable and genuine?”, “is this business what it claims to be?” — that AI can investigate across messy, unstructured data to arrive at a robust answer. The scorecard becomes a blend: deterministic rules where certainty and auditability are non-negotiable, and reasoned questions where judgement used to require a human to read the file. The rules keep it safe and explainable; the questions let it see more.
Three: the scorecard covers the whole job, not just the verdict.
The third change is the one the industry has thought least about. A traditional scorecard is obsessed with the decision to lend — the yes or no. But the decision is only a fraction of the actual work of lending. Before and after that verdict sits a long checklist: gather the right documents, verify identity and income, check affordability, satisfy the conditions, prepare everything a clean, compliant loan requires.
In the new world, the scorecard expands to include that end-to-end checklist. It describes not just the judgement but the actions required to lend — and crucially, the outputs of the scorecard drive agents that go and do those actions. Rather than a human chasing a payslip or re-keying a bank statement, agents autonomously acquire and prepare the quality data the decision depends on. This is where EY's argument and mine meet: their AI-ready data foundations are exactly what let those agents work reliably. But the agents aren't just cleaning a data pipeline in the background. They're executing the scorecard — turning a decision into a completed loan.
Payments close the loop.
There's one more piece, and it's the one that makes the whole thing move. Today, decisioning and payments are two separate worlds run by two separate teams. Origination makes the decision, then hands off to a servicing and collections engine whose only job is to execute a fixed schedule. The decision and the repayment barely speak to each other.
In the new world they become the same living flow, because payments play two roles at once.
First, payments are the sensor. A scorecard tuned on the ongoing performance of borrowers needs a real-time signal, and payments are the richest one there is. Every instalment made, missed, settled early or paid in part is a fresh data point that re-tunes the balance between throughput and book performance. With open banking in the mix, the signal gets richer still — you can see income actually land, not just infer it. The payment layer becomes the primary feed into the living scorecard, not an afterthought bolted on at the end.
Second, payments are the actuator. “Payback that breathes” has to physically happen somewhere, and that somewhere is the payment flow. When affordability shifts within its agreed bounds, it's the payment layer that executes the flex — collecting more when income rises to clear the loan faster, easing off when it dips, rescheduling in real time. The agents in the scorecard's end-to-end checklist don't just acquire data to lend; they orchestrate collection, retries and rescheduling too.
Put those together and the loop closes. The same living scorecard that decides also governs repayment — sensing how the book is performing and acting on it continuously, rather than making one judgement up front and hoping the schedule holds. BNPL proved this in miniature: Klarna fused the decision and the payment into a single flow at checkout. The vision here is to do the same across the whole of lending.
Why this is how you win.
It would be easy to read all of this as idealism — a nicer way to lend. It's also, bluntly, how the smartest lenders will win.
A living scorecard that reasons over questions as well as rules approves more genuinely good customers that a static, threshold-based model would have declined — more throughput without loosening standards. Tuning continuously against real payment behaviour means catching deterioration earlier and protecting the book better than a model rebuilt once a year ever could — better performance for the same appetite. Agents running the end-to-end checklist strip cost and delay out of every loan, so you originate faster and cheaper. And credit that flexes with a borrower's life earns something the incumbents struggle to buy: loyalty. A customer whose lender eased them through a rough month, or let them clear early when they were promoted, doesn't go shopping for rates.
More good customers, better books, lower cost to serve, stickier relationships. That's not a trade-off against doing right by the customer — it's the same move, seen from the lender's side of the table. Which is exactly why it will happen: the incentives point the same way for once.
What this does to credit itself.
Add those three shifts together and something bigger happens downstream. If the scorecard is fluid, reasons over open questions, and reaches into the whole lifecycle of a loan, then credit stops being a static product and starts to become a living relationship.
Consider a five-year loan. Today we price and structure it on day one, using a snapshot of the borrower's circumstances, and then we hold that structure fixed for sixty months — as if a person's life freezes the moment they sign. It doesn't. Over five years people get promoted, change jobs, have children, take pay cuts, recover again. Today's model treats all of that change as risk to be priced defensively up front. It builds a straight line and asks the customer to walk it, whatever happens.
A living scorecard can treat that change as information instead of threat. If someone's income rises, the optimal path might be to clear the loan faster. If they hit a rough patch, easing the schedule — within clear, pre-agreed bounds — keeps them on track rather than tipping them into default. Payback that breathes with real-time affordability, rather than a fixed schedule the customer has to survive.
This isn't as speculative as it sounds. Buy-now-pay-later already gave us a taste: terms shaped to the moment rather than to a bureau score pulled once and applied forever, moving towards the customer instead of bending the customer to a rigid product. The leap is to take that instinct — a decision and a payment fused into one responsive flow — and apply it not to a single checkout basket but to a five-year commitment, letting the terms keep adjusting for the whole life of the loan.
I want to be clear that this is a vision, not a finished blueprint. The bounds matter enormously — fluid credit done carelessly is just risk kicked down the road, and the guardrails, the regulation and the governance around this will need real work. But the destination is worth naming, because it's a better one. For decades the industry has optimised a decision that was frozen in time because that was all the technology allowed. It no longer is. The scorecard can become a living thing, and if it does, credit can finally keep pace with the people it's meant to serve.
That's a lending system that flexes to real lives instead of asking real lives to hold still — and that is a better financial services, for all.
← Back to News & Opinion