Lending Infrastructure

What Has to Move Together When Your LMS Migrates

A loan book migration breaks when systems move on different clocks. What has to move together, in what order, so classification, the ledger and bureau reporting stay correct through cutover.

What Has to Move Together When Your LMS Migrates: cover art
Contents 10 sections

Whether your book lives in a spreadsheet or an old loan management system, you move it in the same order. Clean the records. Fix one freeze date. Load what each loan owes on that date, and run both side by side until the numbers agree for at least one full billing cycle, which is usually a month. An old system adds years of history that have to move with the balances, and that’s most of this post. Starting from Excel? The section below covers the shorter path.

You already know your loan management system has to go. The symptoms are on the table: manual workarounds for a workflow the vendor was supposed to handle, a change request that takes six weeks to price and ship, a report your team rebuilds by hand every month because the system cannot produce it. What nobody in the vendor’s demo room asks is the only question that decides whether the switch is safe: what happens the night you cut over.

Quick answer

An LMS migration is a sequencing problem, not a data export problem. Six things have to move together against one freeze date, the loan master and terms, the repayment schedule as billed, the ledger and GL history, the full classification history, the charges and waivers ledger, and the bureau reporting history. Move any of these on its own timeline and the new system starts reporting a different reality than the one your borrowers and your regulator have been living in.

Most migration plans read like an IT project: export, transform, load, go live. That plan works for a CRM. It does not work for a live loan book, because a loan is not one record. It is a chain of events, disbursal, repayments, a fee waived here, a rate reset there, a DPD counter that has been running since the day a payment was missed, and every one of those events has to agree with every other one on what day it happened. Split the chain across two systems that disagree about the freeze date, and the new system does not fail loudly. It fails quietly, in a classification date, a reopened waiver, or a bureau file that no longer matches the account’s real history.

Pick one freeze date, and hold every number to it

Before any file moves, agree on a single moment: the exact date and time every number in the old system becomes the number the new system has to match. Not “end of month”, a timestamp. Every extract, every reconciliation, every parallel-run comparison gets checked against that one point, because a loan book does not stop moving while you migrate it. Repayments post, charges accrue, and if two team members pull “current state” twelve hours apart, they are pulling two different books.

The freeze date is also the point every downstream check refers back to. When something looks wrong three weeks into the new system, the first question is not “what changed”, it is “did this account match the freeze-date snapshot on day one.” Without a named freeze date, that question has no fixed answer to check against.

What moves as one unit, not four exports

A migration plan that treats the loan master, the schedule, the ledger and the classification history as four separate exports is where most of the damage starts. They are not four data sets. They are one loan, described from four angles, and each angle has to agree with the others about the freeze date.

What has to moveWhat breaks if it moves on its ownWhat it must agree with, at the freeze date
Loan master and termsThe new system computes a different EMI or maturity date from day one.Every live loan’s rate, tenor and product code, as they stand today, not as originally sanctioned.
Repayment schedulePast and future instalments look inconsistent with what the borrower was actually billed.The schedule as billed after every reset and part-payment, not the original disbursal-day schedule.
Ledger and GL postingsOpening balances do not tie out, so the first month-end close on the new system does not reconcile.Every posting since disbursal, not just the current outstanding balance.
Classification history (DPD, SMA, NPA)An account’s NPA date resets to “today”, understating provisioning and misstating the account’s real risk history.The full days-past-due trail, not just the bucket the account sits in right now.
Charges, waivers and part-paymentsA previously waived charge reappears, or a part-payment silently changes the tenor instead of the EMI.Every charge event tied to the loan it belongs to, not the net balance after the event.
Bureau history and CKYC linkageThe bureau file shows the account as newly opened, confusing the borrower’s own credit history.The account’s existing bureau reference number, not a freshly generated one.

Notice what the middle column has in common: none of these are IT failures. They are audit-by-design failures. A system that treats history as optional metadata, rather than the thing that proves a classification date or a waiver decision was correct, produces a migration that looks clean in a demo and falls apart the first time an auditor asks why an account’s NPA date changed.

Event-based LMS migration: replay the loan history

You’ve got 2 ways to load a live book into a new LMS. You can load each loan’s opening position as on the freeze date, and keep the old history as a record. Or you can load the history as dated events, and let the new system rebuild balances, DPD (days past due) and SMA or NPA status from them. The second is an event-based migration. It’s more work. What you get for it is a test of how the new system processes your past activity, loan by loan, before you trust it with new activity.

Here’s an illustration, not a customer. A ₹1,00,000 loan over 12 months at 18% a year pays out on 6 March, with an EMI of ₹9,168 due on the 6th. Take 5 events from its life: the disbursal, a failed EMI collection on 6 June, a ₹500 bounce charge on 7 June, a ₹9,168 receipt on 10 June and the charge waived on 13 June. They aren’t 5 GL entries or the whole history. A replay also needs the schedule, the earlier receipts and the rules. Assume monthly reducing-balance interest, every earlier EMI paid, the 10 June receipt applied in full to June’s EMI, and no other charges. At a 14 June cut-off, the replay should land on the old system’s balance, 0 DPD and nil charges by itself. If it doesn’t, you’ve found something to check. Better now than in your first month-end close.

You can check a replay in 5 steps:

  1. Freeze the figures you’ll compare against: each loan’s balances, DPD, SMA or NPA status under RBI’s rules, and open charges at one agreed cut-off (14 June, in the example). Keep the full export, not a summary tab in Excel. You’ll need it in step 4.
  2. Export every event with its effective date and its posting date. You also need the rule versions in force on each date: rates, charges, allocation order, rounding and day count (Actual/365 or 30/360, say). Miss one and you’ll chase ghosts.
  3. Map each old event type to a new one, and list the ones that don’t map, such as old manual adjustments. Don’t bury them. Look into each one, and load it only as an adjustment the lender has approved, with its source reference, dates and reason. Never add one just to force a match. That way you can show an auditor, 2 years from now, why that loan’s figure moved on the day you went live.
  4. Replay the events in the new system. Then compare its figures with your step 1 export at the same cut-off.
  5. Look into every difference. Is it the old system, your mapping or a rule? Write down which. We’d log the loan ID too. Keep the open ones on a list the parallel run in the next section has to close.

There’s a catch. A full replay needs complete, clean history, and it can’t test a period that’s missing. Where history before, say, April 2022 is missing, check an approved opening position as on that date and replay the events after it. Then say plainly, in your sign-off note, which periods sit outside the test. A match with the old LMS also isn’t proof either was right. If the old system ran interest from the sanction date, not the disbursal date, and the new one is set up the same way, you get the same wrong answer twice. Your GL carries it too. You’ll check the rules too. How to test for that is in ledger drift in loan servicing.

The parallel run: prove agreement before you trust it

Data mapping tells you the fields lined up. A replay tests the history. Neither tells you the new system will handle next month the way the old one does. For that, you run the old and new systems side by side, on the same live loans, for at least one full billing cycle, and reconcile every output before you let the old system go.

A test on a handful of sample accounts will not catch what a parallel run catches. A small sample tends to miss the complex accounts. The failures live in the accounts that are not simple: the one with three part-payments and a waived late fee, the one that moved from SMA-1 back to standard last quarter, the co-lent account split two ways. A parallel run puts the whole book through both systems and asks a narrow question about every single account: does the new system’s number match the old system’s number, for the ledger balance, the classification bucket and the next instalment due. Anywhere it does not match, you find out before go-live, not after.

This is also where synthetic test data earns its keep. Running edge cases, an account near an NPA boundary, a part-prepayment mid-cycle, a co-lending split, through a realistic but non-production book before the real parallel run starts catches structural mismatches without risking a live account. It is the same discipline behind building a living digital twin of a lender’s book for QA: break the system on a copy first, so the parallel run on the real book is confirming, not discovering.

Cutover weekend: the order that keeps classification correct

When the parallel run agrees, cutover itself is short, and the order matters more than the speed. Stop new transaction entry in the old system first, so nothing posts after the freeze date on one side and not the other. Take the final ledger and classification snapshot second, timestamped to the freeze date, not “whenever the export finished.” Load that snapshot into the new system third, and run the new system’s own day-end classification job against it before anyone touches a live account, so you catch a bucket that shifted on load before a borrower or a bureau ever sees it. Only after that day-end run agrees with the old system’s last known state do you open the new system for live transactions.

Skip the order and run the new system’s day-end job before confirming the loaded snapshot, and you are debugging classification and production traffic at the same time, on the same weekend.

Day one in the new system: what has to reconcile first

The morning after cutover, before a single new payment posts, three numbers have to tie out against the freeze-date snapshot: total ledger balance across the book, count of accounts by classification bucket, and total charges outstanding. These are cheap to check and expensive to discover wrong three weeks later, once a month’s worth of new activity has been layered on top of a bad starting point.

Bureau reporting is the other first-day check, and it is unforgiving of a bad migration in a specific way: a lender reports to credit bureaus using four reference dates a month, and the first file the new system produces has to carry the account’s existing bureau reference, not a new one. Get this wrong and the borrower’s own credit history looks discontinuous, which is the kind of error that generates support calls and bureau complaints, not just an internal reconciliation flag.

Moving from Excel to a loan management system

If your book runs in Excel or Google Sheets, check how many years of history it holds, and how clean it is, before you scope the move. Say you have 400 loans across two tabs and two people who both update them. The same four steps apply, in this order:

  1. Clean the sheet. You want one row per loan, with terms that match the signed agreement: amount, rate, tenure, first due date and charges. Date every receipt, including NACH bounces and cash collected in the field. If two people keep copies, decide which one is the book before you start, because you can’t reconcile against two.
  2. Fix the freeze date. Pick one timestamp, say 6 pm on 31 March. Every receipt up to it counts. That’s the snapshot you load. Until cutover, record every later receipt in both the sheet and the new system, so the two can be reconciled.
  3. Load opening balances. For each loan, you load principal outstanding, interest and charges due, and days past due as of the freeze date. A loan 20 days overdue on that date has to arrive as SMA-0 (1 to 30 days), one at 35 days as SMA-1 (31 to 60 days), one at 75 days as SMA-2, and one past 90 days as NPA under the RBI’s prudential norms, not as a fresh account. If you’ve rebuilt a schedule by hand after a part-prepayment, load the one you actually billed, not the original.
  4. Run in parallel. Keep updating the sheet for at least one billing cycle while the new system bills the same loans. You’ll check every EMI, every receipt split between principal, interest and charges, and every overdue count. Don’t stop updating the spreadsheet until the two agree.

Fewer loans mean fewer schedules to check: 400 loans, 400 schedules. Missing history and complex terms still add time. Moving a small book off Excel covers what that looks like on Lokta, including who checks the migrated schedules against your spreadsheet before you switch.

What goes wrong when a step is skipped

The failure modes are not exotic. They are the same handful, every time a step above gets skipped or rushed.

DPD resets to zero

Classification history did not migrate as a trail, only as a current bucket, so the new system has no record of how long an account has actually been overdue. An SMA-2 account looks brand new.

A waived charge reappears

The waivers ledger moved on a different export than the charges it applies to, so the new system re-bills something a collections team already agreed to drop.

A part-payment changes the wrong thing

The schedule migrated as a snapshot of today’s instalment, not a history of every reset, so a part-payment the borrower expects to shorten the tenor instead recalculates the EMI, or the reverse.

A co-lender’s share drifts

The split was reconstructed from the current balance rather than carried across as its own record, so each co-lender’s share of the same loan slowly disagrees between the two books.

Every one of these traces back to the same root cause: something was treated as a snapshot when it needed to travel as history. Loan management systems that hold classification, charges and postings as an event-sourced ledger rather than a mutable current-state table narrow this problem, because the history is the record, not a derived report. The lender still checks the imported data and the rules.

The migration is a sequencing problem, not a data problem

Every field in the loan book will map to something in the new system. That was never the hard part. The hard part is making six different views of the same loan agree with each other on the same freeze date, proving that agreement in a full parallel run before you trust it, and cutting over in an order that checks classification before it goes live rather than after. Get the sequence right and the migration is uneventful. Get it wrong and you find out from an auditor, a bureau complaint, or a borrower who calls asking why a fee they were told was waived just showed up again.

None of this is free. A proper parallel run costs a full billing cycle you might be tempted to compress, and a small team feels every week of it. There are three ways to spend that cost. Run the migration yourself, on your existing tools, and treat the sequence above as a manual checklist your team owns end to end. Bring in a migration specialist to run the parallel-run process for you, which buys speed at the cost of building no internal muscle for the next system change. Or move onto a platform where the loan master, schedule, ledger and classification history are one event-sourced record instead of four systems that have to be kept in sync, which does not remove the freeze-date discipline but shrinks what has to migrate down to one history instead of four exports.

Frequently asked questions

How do I move my loan book from Excel or an old loan management system to a new one?

You move both the same way, in four steps, and an old system just brings more history. From Excel, clean the sheet first: one row per loan, terms that match the signed agreement, and every receipt dated. Then pick a freeze date, such as 6 pm on 31 March, load each loan's opening balance and arrears as of that date, and check the new schedules against the spreadsheet for at least one billing cycle before you stop updating it. On a 400-loan book, that's 400 schedules you'll check line by line. From an old loan management system, the steps don't change but you'll carry more history: the schedule as billed, the ledger, the DPD and SMA or NPA history, charges and waivers, and bureau history all move against one freeze date. Then you run both systems in parallel until they agree.

What data has to migrate together when an NBFC switches loan management systems?

Six things, tied to one freeze date: the loan master and terms, the repayment schedule as actually billed, the ledger and GL postings to date, the full DPD and SMA/NPA classification history, the charges and waivers ledger, and the bureau reporting history and CKYC linkage. Moving these on separate timelines is what breaks classification and reporting after cutover, because each one depends on the others agreeing about the same freeze date.

What is an event-based LMS migration?

It's a migration where you load each loan's history into the new LMS as dated events: the disbursal, every EMI receipt, bounces, charges, waivers and rate resets. With the schedule and rules in force, the new system rebuilds balances, DPD and SMA or NPA status, and you reconcile them against the old records loan by loan. An opening-position load can keep the history as a record too, but it doesn't test how the new system processes it. Where older history is missing, start from an approved opening position, replay the events after it and write down what wasn't tested.

What is a parallel run in an LMS migration, and why does it matter?

A parallel run is at least one full billing cycle, usually a full month-end, where the old and new systems process the same live loans side by side and every number is reconciled before the old system is switched off. It matters because a mismatch that looks small in a test account (a wrong DPD count, a missing waiver) shows up in production as a wrong SMA date or a wrong bureau file, and by then the old system is already gone.

How long does a loan management system migration take?

There is no single answer, because it depends on book size, product variety and how much classification and charge history has to carry over, and any migration vendor who quotes a fixed week count before seeing the book is guessing. What is fixed is the shape of the timeline: data mapping and validation, then a parallel run of at least one full billing cycle, then cutover, then a reconciled first month-end close on the new system.

What usually goes wrong in an LMS migration?

The same handful of failures recur: DPD counters reset to zero because classification history did not migrate, a waived charge reappears because the waivers ledger moved separately from the charge it applies to, a part-payment silently changes the tenor because the schedule migrated as a snapshot instead of a history, and a co-lender's share of a loan drifts because the split was reconstructed rather than carried across.

Read next:


Chandramouli is the co-founder and CEO of Lokta, the agentic loan servicing platform. He has spent two decades building AI for decisions that change people’s lives, and has served as an independent director on an NBFC board. He writes here about the mechanics that decide whether a system change is safe.

Bring us your live book

See what agents can do after approval.

Talk to us
Founder-led adoption

Adopt the agentic loan servicing platform.

Lokta is built for enterprise deployment, VPC or single-tenant cloud, with an audit trail in every state change. We work with a select group of institutions through a founder-led model: deep adoption, deliberate scope, a delivery window the team commits to in writing.