Most credit teams watch the wrong clock. They track delinquency, roll rates, and vintage curves, and those matter, but every one of them is a rear-view mirror. They tell you about losses that have already begun. By the time an account shows up as 30 days past due, the story is largely written: the cure is harder, the loss is closer, and you are reacting instead of steering.

The teams that protect a book best look through the windshield instead. They watch the signals that fire before the missed payment, and this is the part that separates talkers from operators, they act on them in a disciplined, repeatable way. That discipline is what an early warning indicator (EWI) framework is for. This piece covers the full thing: why lagging metrics aren’t enough, the catalogue of leading signals and how much warning each one gives, how to turn signals into a watchlist that doesn’t drown you in noise, the menu of actions, and how to prove the whole thing actually reduces loss.

There is nothing wrong with delinquency rates, roll rates, and vintage curvesyou need them, and a later issue is devoted to reading them well. The problem is timing. They are confirmation, not warning. A roll rate tells you how fast accounts are deteriorating once they’ve started; a vintage curve tells you a cohort underperformed after it’s seasoned. By then your options have collapsed to damage control.

Leading indicators change the timeline. They surface stress weeks before a payment is missed, while you still have room to act — to reach out, restructure, reduce a limit, or simply watch closely. The difference between the two clocks is the difference between prevention and post-mortem.

Figure 1: Leading signals appear before Day 0, while you can still act; lagging ones confirm after.

Lagging indicators tell you what happened. Leading indicators give you the chance to change it.

Leading signals come from several sources, and the signal is not spread evenly, some give you two months of warning, others two weeks. Here is the working catalogue, roughly in order of how much runway each tends to give.

Cash-flow signals, if you have transaction data, are the earliest and richest: salary landing late or shrinking, balances hitting zero before payday, rising overdrafts, a sudden jump in gambling or BNPL spend. Score migration comes next, watch the slope, not the level; a customer sliding two bands over a few months is telling you something before they ever miss you. Bureau triggers reveal what they’re doing elsewhere — a missed payment at another lender, a flurry of new inquiries, loan stacking. Utilization creep, engagement drop-off, and payment-behaviour shifts round out the set; the last of these, full-pay turning into minimum-pay, autopay switched off, paying later within the grace window, is a quiet confession that arrives closer to the event but is highly predictive.

Figure 2: Illustrative lead times. Earlier signal = more room to act.

A catalogue of signals is useless — worse than useless — if it produces twenty independent alerts per account. That is how you get alert fatigue, where everything is flagged and therefore nothing is acted on. The job is to combine signals into a single, tiered watchlist: one prioritised list, where an account’s tier reflects how many signals are firing and how severe they are.

Figure 3: Signals feed one tiered watchlist; each tier has a mandated action, measured against a control.

Each tier must carry a mandated action and a clear owner. Without that, the watchlist is just a report. The point of tiers is to match the intensity of the response to the strength of the signal, a single soft signal earns monitoring; cash-flow distress plus a bureau miss earns a phone call and a pre-emptive offer.

Watching without acting is just expensive curiosity. The menu of pre-emptive actions, escalating with tier, is roughly this: intensified monitoring; a soft, helpful outreach (a reminder, a check-in, a nudge to update autopay); a limit reduction or step-down to cap downside on a deteriorating account; a re-underwrite when behaviour or bureau data has materially changed; and, for genuine distress, an early hardship or restructuring offer before the account ever rolls. The earlier you intervene, the cheaper and gentler the action can be, which is the whole economic argument for leading over lagging.

Here is where almost every EWI programme falls down: it never proves it works. A signal you act on but never measure is a belief, not a control. The discipline is to treat your interventions like experiments, hold out a control group of flagged-but-untreated accounts and compare default rates. If treated accounts don’t default less than the control, your action isn’t working and you should change it, not scale it.

Don’t try to build the whole catalogue at once. Pick your three highest-signal, earliest indicators — for most digital lenders that’s bureau-trigger inflows, a payment-pattern shift, and a cash-flow signal — and stand up a single weekly watchlist with one mandated action per tier and a control group from day one. Prove the lift on those three, then add signals. A small EWI programme that is acted on and measured beats a comprehensive one that sits in a dashboard.

Stand up one weekly watchlist, with one mandated action per tier, and a control group. The watchlist forces prioritisation, the mandated action forces intervention, and the control group forces honesty about whether any of it is working. That single habit will do more for your loss rate than another month of refining the perfect score.

Next issue: BNPL economics — where the loss really comes from, and the lever most teams miss.

If this was useful, forward it to someone watching a book.

Views are my own and do not represent my employer.