Half of FY27 is gone. Whatever NPA number your NBFC reports next March is being decided right now, in the accounts currently sitting in SMA buckets - because an account that crosses 90 DPD in December was one you could still have saved in September. NPA reduction sounds like a year-end project; it's actually a mid-year one.

Here are five moves with evidence behind them, and two that consume budgets without moving the number.

What works

Move the first intervention before the miss. The single highest-return change available to most lenders. Recovery probability in the pre-due and SMA-0 window sits above 80% and falls below 40% past 60 DPD, so every week of delay in first contact is measurable money. This requires prediction - scoring live accounts for default probability using repayment trajectory, bureau movement and behavioural signals - because you can't intervene early on stress you can't see. Lenders on FrenzoFinserv see their largest uplift, around 38%, in exactly these pre-due buckets.

Stop treating all overdue accounts the same. Uniform treatment is quietly the most expensive policy in collections. When every 30-DPD account gets the same calling effort, resources spread thin across accounts that would have self-cured and accounts that needed triple the attention. Segment by roll-forward probability, not just bucket: a low-risk X-bucket account gets an automated nudge; a high-risk one gets an agent today. Same team, radically different yield.

Watch roll rates, not just bucket totals. PAR tells you how big the problem is; roll rates tell you where it's coming from. Cohort-level roll tracking - by product, vintage, geography, sourcing channel - exposes the specific segments feeding your NPA line, and those findings usually surprise people. One lender's "collections problem" turns out to be a single product's November vintage; another's is one state's field coverage. You can't fix a portfolio average, but you can fix a named cohort.

Make self-cure effortless. A meaningful share of early delinquency is friction, not refusal: the borrower who missed the auto-debit, changed banks, or needs ten days. Every step between "I want to pay" and "paid" costs you a percentage of them. Payment links in every message, a self-service surface for payment and restructuring requests (this is what our APRUVIT borrower app exists for), and restructure options offered by rule rather than by escalation - these convert intent into cash without an agent minute spent.

Close the loop between outcomes and strategy. Every month your collections operation generates outcome data: which treatments worked on which segments. In most operations that data dies in reports. Fed back into the models and routing rules, it compounds - the operation gets a little sharper every cycle, which over a year is the difference between linear and improving performance. This feedback loop is, more than any single feature, what a collections platform actually is.

What doesn't work

Hiring your way out. The reflex response to rising NPAs is more agents, and it produces linear cost for sub-linear return. More callers making the same late, untargeted, uniformly-distributed calls recover marginally more at proportionally higher cost - the pattern shows up in every quarter of results across the industry. Headcount added after intelligence multiplies; headcount added instead of intelligence just dilutes.

Write-off cosmetics. Aggressive write-offs make the ratio look better and change nothing about the flow of accounts feeding it. Six months later the ratio is back, minus the written-off capital. Write-offs have a legitimate place in balance-sheet hygiene; as an NPA strategy, they're an anaesthetic.

The September arithmetic

The moves above share a property: they act on accounts before 90 DPD, which means their effect on FY27's closing NPA number depends on when they start. Started in September, pre-due workflows and risk-based routing are shaping the December and January slippage numbers. Started in January, they're shaping FY28's.

A platform deployment on FrenzoFinserv runs 4–6 weeks on your existing LMS - decision to live portfolio inside a month and a half, with recovery improvement accruing from roughly week five. Lenders on the platform average 35% higher recovery, 25% faster resolution and about 30% lower collections cost. If NPA reduction is on your H2 agenda, we'll model the slippage math on your actual buckets - it's a short conversation and the spreadsheet is yours to keep either way.