Almost every Indian company that has been in business for fifteen years has one: a billing, inventory or production system written in Visual Basic 6, FoxPro, Delphi or early PHP, maintained by one person who has been around since it was installed. It still works. Nobody wants to touch it. And every new requirement, from GST changes to a customer portal to a mobile app for the sales team, has to be bolted onto it with exports, Excel sheets and manual re-entry.
This guide is about replacing that system without the one thing owners fear most: a weekend cutover that goes wrong and stops billing on Monday morning.
Why does the big-bang rewrite keep failing?
The instinct is to commission a brand-new system, build it for a year, and switch over on a fixed date. Martin Fowler, who popularised the alternative approach, notes that full replacements of this kind usually take far longer than planned, users cannot wait that long for new features, and the existing behaviour turns out to be much harder to specify than anyone expected. Much of that behaviour is undocumented rules that only exist in the old code.
The costs of a failed cutover are not theoretical. In April 2018 the UK bank TSB moved its customer data to a new IT platform in one migration. The platform failed immediately, disrupting branch, phone, online and mobile banking for a significant share of its 5.2 million customers, and services did not return to normal until December 2018. In December 2022 the UK regulators fined TSB a combined £48.65 million, finding that it had not adequately organised and controlled the migration programme or managed the risk from its outsourced IT supplier. TSB had also paid £32.7 million in redress to customers.
Your business is not a bank, but the lesson scales down: one irreversible switch-over date concentrates every unknown into a single weekend.
What is the strangler fig approach, in plain terms?
The strangler fig pattern replaces the old system one piece at a time. A routing layer (often called a facade or proxy) sits in front of both systems. At first it sends everything to the old system. As each function is rebuilt, say customer master, then quotations, then dispatch, the facade starts sending that function's traffic to the new system. Users keep working; the old system shrinks until it can be switched off. Microsoft's architecture guidance describes exactly this sequence, with the facade finally removed once nothing depends on the legacy system.
The trade-off is honest: for some months you run two systems and some temporary "glue" code that will later be thrown away. That costs money. It is still far cheaper than a failed cutover.
What should you move first?
Do not start with the hardest, most central module. Start where the payoff is visible and the risk is contained. A practical ranking for most Indian SMEs:
- Reporting and dashboards. Read-only. Copy data out of the old database every night (or continuously) into a new store and build the MIS reports management actually needs. Nothing in the old system changes, and the owner sees value in weeks.
- Customer-facing or field-facing additions. A dealer portal, a sales app, a service-request form. These are new capabilities that write into the old system through a controlled interface, so the old screens keep working.
- Self-contained back-office modules. Purchase requests, HR leave, gate passes. Few links to the core ledger.
- Master data. Customers, items, vendors. Once the new system owns masters, everything after becomes easier.
- Transactional core last. Invoicing, stock movement and accounting entries move only after the team trusts the new platform.
A quick scoring table for each module
| Question | Move early if | Move late if |
|---|---|---|
| Does it write to the ledger or stock? | No | Yes |
| How many other modules depend on it? | Few | Many |
| Is the business logic documented? | Yes, or simple | Lives only in old code |
| Is it a daily pain for users? | Yes, quick win | Rarely touched |
| Can you roll back in an hour? | Yes | No |
How do you run old and new in parallel safely?
Parallel running is where most of the real engineering happens. Four rules keep it safe:
- One system of record per data item at any moment. Customer addresses are owned either by the old system or the new one, never both. Write it down in a simple ownership table and review it weekly.
- Sync with change data capture, not manual exports. Read changes from the old database (triggers, transaction logs or a scheduled delta job) and apply them to the new store. Manual CSV exports drift within days.
- Put a translation layer between them. Old systems carry odd conventions: dates stored as text, status codes like "Y/N/X", item codes padded with spaces. Keep that mess in an adapter so it does not leak into the new design. Microsoft calls this an anti-corruption layer.
- Reconcile every day. Run an automated comparison of totals: invoice count and value, stock by warehouse, outstanding by customer. When the numbers match for two or three consecutive weeks, that module is ready to cut over.
Keep a rollback path for each module until it has run cleanly through at least one month-end close. Delete legacy tables only after that, as a deliberate final step.
What if the source code is lost or nobody understands it?
This is common with FoxPro and VB6 systems built by a vendor who has since closed. You still have options. Most such systems keep their data in a database you can read (DBF files, Access, SQL Server or MySQL). You can intercept at the data layer: read changes from the database, and for writes, have the new system insert records in the format the old screens expect. Before you do that, spend a week documenting what the old system actually does, by watching users and by tracing which tables change when they press each button. That week is the cheapest insurance in the whole project.
Microsoft's guidance is clear that the pattern is hard to apply when you can neither intercept requests nor change the legacy system or its database. If that is truly your situation, a carefully planned phased replacement by department or branch may be the better route.
What are the signs you have waited too long?
- The system runs only on an old Windows version that no longer receives security updates, and you keep one machine unpatched "because the software needs it".
- Only one person can fix it, and they are planning to retire or leave.
- You cannot meet a compliance change, such as e-invoicing or an audit-trail requirement, without a separate workaround.
- Month-end close needs someone to export data and fix it in Excel before it reconciles.
- A customer or bank has asked for API integration and the honest answer is "not possible".
- Backups are a copy of a folder, and nobody has tried restoring one.
Two or more of these together mean the risk of doing nothing is now higher than the risk of change.
A realistic timeline for a mid-sized Indian business
| Phase | What happens | Typical output |
|---|---|---|
| Discovery | Map modules, data, users, integrations; record undocumented rules | Module map and migration order |
| Foundation | New database, data sync, facade, reconciliation reports | Live dashboards from old data |
| Module waves | Rebuild and cut over one module at a time | One module live every few weeks |
| Core cutover | Masters, then transactions, after a clean month-end | New system becomes system of record |
| Decommission | Archive old data read-only, switch off old servers | One system, documented |
Testing is not a phase at the end; automated regression checks should run on every module before it goes live. That is why an automated QA suite belongs in the foundation phase, not at the end.
Common mistakes to avoid
- Copying every old screen exactly. Ask what the screen is for; often three old screens become one.
- Migrating fifteen years of junk data. Archive closed years read-only and migrate only what is live.
- Leaving the users out. The accounts clerk knows rules no document records. Involve them from week one.
- No owner on the business side. Someone senior must decide which system is the truth when numbers differ.
How Cehpoint can help
We approach legacy replacement module by module, starting with a working demo on your own data before you commit, and you own the code and infrastructure at the end. If you want to understand your options first, a paid one-to-one consultation with a senior engineer is a sensible first step. For the full engagement, see our legacy application modernization service, or read how our demo-first delivery works.