How to Convert Legacy Data Before an ERP or CRM Migration — Odoo & ERP article by Rui Codex

Convert legacy data by cleansing it at the source first, mapping each old field to a target field with a documented rule, and proving the mapping through repeated test loads before cutover. Each step exposes problems the previous one could not, so skipping a step moves the cost of errors into live operations.

Key takeaways

  • Cleanse data at the source before writing any mapping rules.
  • Map every legacy field to a documented target field, including fields that will be dropped.
  • Run several test loads from fresh copies and reconcile record counts and sample values after each one.
  • Treat personal data as a regulated asset throughout the migration, not only after go-live.

What does data conversion involve?

Data conversion is the process of extracting records from legacy systems, transforming them into the structure the new ERP or CRM expects, and loading them into the target. It is usually the workstream that takes longest to finish, because it depends on data that nobody has reviewed for years.

Conversion has three distinct activities. Cleansing corrects or removes bad records. Mapping defines how each source field becomes a target field. Test loading runs the whole chain on copies of the data so errors surface before go-live. Teams often treat these as one task, which is why problems appear late.

For background on how these projects are scoped, see our page on ERP and CRM integration services.

Why should cleansing happen before mapping?

Cleansing should happen before mapping because mapping rules assume the data is consistent. If the source holds three spellings of the same customer or dates in two formats, the mapping rules turn into exception handling and the target system inherits the mess.

Typical cleansing tasks

  • Remove exact duplicates and identify near-duplicates, such as the same company under different legal suffixes.
  • Standardise formats for dates, currencies, phone numbers, postal codes and units.
  • Archive records that are obsolete, such as customers with no activity for years, once the business agrees on the rule.
  • Fill mandatory fields from authoritative sources, or record that they are empty and why.

Cleansing needs a business owner. An engineer can find duplicates, but only the sales or finance lead can decide which record is correct. Without that decision, the migration team ends up guessing.

Tip: Keep a log of every cleansing decision with the rule applied and the person who approved it. Auditors and the business both ask for it months later.

How do you map legacy fields to the new system?

You map legacy fields by writing one documented rule for each target field that states the source, the transformation and the default value when the source is empty. A mapping sheet with these three columns is the minimum a project needs.

Where mapping fails

Mapping fails most often in three places. Fields that look equivalent can carry different meanings, such as a status code that means 'active' in one system and 'has a contract' in another. Many-to-one cases combine two legacy fields into one target field and lose information. One-to-many cases split a free-text field into structured fields that were never reliably filled in.

Custom fields and notes fields need explicit decisions. Each one is either mapped, moved to an attachment or history record, or dropped with approval. Silently dropping data is the most common source of later complaints.

Integration patterns matter too. If the new system connects to other tools through APIs, the mapping should reflect the identifiers those connections use. Our overview of system integration and API engineering articles covers this from the connection side.

How should test loads be run?

Test loads should run on a recent copy of production data, repeat until the results are stable, and be reconciled against the source with counts and sampled values. A single successful test load proves very little on its own.

  1. Load a small sample of records that covers every object type and known edge case.
  2. Load a full copy and time it, since the cutover window depends on that measured duration.
  3. Reconcile record counts per object, and total balances wherever quantities or money are involved.
  4. Check a random sample and every edge-case record by hand together with the business user.
  5. Fix the mapping or cleansing rule rather than patching the output, then repeat the load.

Each repetition should start from a fresh copy of the source, because the live system keeps changing. Otherwise a fix made in one run cannot be trusted in the next.

Tip: Record the duration of each full test load. If it grows between runs, investigate before go-live rather than assuming production hardware will absorb the difference.

Where do data conversions usually fail?

Conversions usually fail at the boundaries: between systems, between teams, and between the test environment and production. The data itself is rarely the only cause.

  • Source data changes after the last test load, so the cutover load differs from what was tested.
  • Reference data such as product codes, price lists or the chart of accounts is not aligned before loading.
  • Identifiers are regenerated by the new system, which breaks links to documents and to other integrations.
  • Nobody owns the reconciliation sign-off, so go-live proceeds on assumptions.

Choosing a delivery partner does not remove these risks, but it does determine whether someone checks them. The questions in our guide on how to choose a software development partner in Belgium include testing and handover topics that apply to conversion work.

How do GDPR obligations apply to migrated personal data?

Under the GDPR, moving personal data into a new system is processing, so principles such as data minimisation, accuracy and storage limitation apply to the migration itself and not only to the live system. The regulation text is published on EUR-Lex. This article is general information and not legal or tax advice.

In practice, test environments should not hold more personal data than necessary, copies should be deleted when testing ends, and retention rules should be applied during cleansing rather than after go-live. Responsibilities for each step should be recorded in your own processing documentation.

Hypothetical examples of a conversion

Here are two hypothetical scenarios:

In the first, a distributor finds that its customer master contains the same company under several names. Cleansing merges them only after the sales team confirms which account holds the open invoices, so the mapping rule for the new CRM stays simple. Without that confirmation, the migration would have created duplicate accounts and split the order history.

In the second, a manufacturer maps a free-text production note field into structured fields. Test loads show that many notes cannot be parsed reliably, so the team keeps the original text as an attached history record and maps only the fields that are consistently filled.

What should be checked before cutover?

Before cutover, confirm that the final test load matches the source within the agreed tolerance, that every open decision in the mapping log is closed, and that the business has signed off the reconciliation report. If any of these is missing, the go-live date should move.

  • Final reconciliation report signed by the business owner of each data object.
  • Mapping sheet frozen with a version number and a change log.
  • Rollback plan tested, including how the legacy system is kept read-only.
  • Integration identifiers verified against every connected system.
  • Access rights reviewed so migrated data is visible only to the people who need it.

Frequently Asked Questions

How long does data conversion take in an ERP migration?

There is no fixed duration, because it depends on data volume, the number of source systems and how much cleansing the business approves. Teams should plan for several test cycles rather than a single load, and should time each full load so the cutover window rests on measured results. Conversion is commonly the longest workstream, so it should start early.

Should we migrate all of our legacy data?

No, migrate only the data the new system and the business need to operate. Obsolete, duplicated or expired records should be archived or deleted under documented rules. Keeping the old system in read-only form or as an archive can answer historical questions without loading everything into the new platform.

What is the difference between cleansing and mapping?

Cleansing corrects the content of source records, while mapping defines how each source field is transferred to a field in the new system. Cleansing answers whether a record is correct, and mapping answers where it belongs. The two depend on each other, so they should be designed together even though they run in sequence.

How many test loads are enough?

The number is not fixed, but testing should continue until a full test load reconciles cleanly and the last two runs show no new defects. Each run should use a fresh copy of the source, because live data keeps changing. Stopping after one clean run often means the result depended on a convenient dataset.

Can conversion be done without a specialist?

Yes, a capable internal team can perform conversion, especially for simple datasets. The main risk lies in reconciliation and sign-off, which need someone who understands both the source and the target. Rui Codex offers technical implementation support for these steps through its integration services, which you can discuss via the contact page.

Conclusion

Legacy data conversion succeeds when cleansing, mapping and test loads are treated as separate, owned steps rather than a single import. Start with the data itself, document every decision, and repeat test loads from fresh copies until the reconciliation holds. Rui Codex is led by a founder with 200+ projects delivered, and published work is described at our case studies page. To discuss the conversion scope for your own migration, request a project consultation.

Sources

Tags: legacy data conversion ERP migration CRM migration data cleansing data mapping test loads system integration GDPR

Need Help Implementing This?

Our team can help you put these insights into practice. From AI automation to custom software development, we build solutions that deliver real results.

Send a project request