The two bulk loaders behind every Fusion implementation, and why the tooling is never the hard part.
FBDI (File-Based Data Import) loads ERP and SCM data into Oracle Fusion using prescribed spreadsheet templates converted to CSV and uploaded for processing. HDL (HCM Data Loader) does the equivalent for HCM, using a structured file format. Both are bulk, asynchronous and validated — and both will reject data that your source system has been tolerating for years.
Every Fusion implementation runs on these two loaders, and every implementation underestimates the cleansing rather than the loading. The templates are well documented and the mechanism works. What takes the time is that Fusion validates relationships your legacy system did not: a supplier without a valid address, an employee with an impossible hire date, a cost centre that was retired.
Oracle publishes an FBDI template per object. The template defines the columns and the required fields precisely.
Map legacy fields to template columns. This is where the real design decisions surface — what to carry, what to retire.
Your team's work, and the long pole. Missing values, invalid references and duplicates all have to be resolved at source.
Upload, run the import, and read the error report. Early runs are expected to fail; that is what they are for.
Record counts and control totals compared against source. A load that completed is not the same as a load that is correct.
A migration team plans to load suppliers, supplier sites, bank accounts and open invoices over one weekend. The files are prepared in parallel by different people and loaded in the order they happen to be finished.
It fails on the first attempt, and the reason is dependency rather than data quality. Supplier sites reference suppliers that must already exist. Bank accounts reference sites. Invoices reference both the supplier and the site, and also a currency, a payment term and a distribution account that must all be set up first. Loading them in the wrong order produces thousands of rejections whose messages describe missing references rather than the actual problem, which is sequencing.
The fix is to establish the dependency order first and rehearse the whole sequence in a test environment end to end — not file by file, which is how dependency errors stay hidden. A full rehearsal also produces the thing most migrations lack: a measured elapsed time, so the cutover plan is built on an observed number rather than an estimate.
| Term | What it means |
|---|---|
| FBDI | File-Based Data Import — spreadsheet templates for loading data into Fusion applications. |
| HDL | HCM Data Loader, the equivalent mechanism for human capital management objects. |
| Control file | The definition telling the loader how to interpret the file's columns and records. |
| Business object | The logical entity being loaded — a worker, an invoice, a supplier — often spanning several files. |
| Load sequence | The order objects must be loaded in, driven by their dependencies on one another. |
It is the single most common cause of a slipped go-live. Begin in the first month and run loads early and often.
Decide deliberately how much transactional history to carry. Carrying everything is expensive and rarely used.
A successful load with wrong totals is worse than a failed one, because nobody investigates it.
FBDI and HDL are reusable for ongoing interfaces. Building throwaway scripts instead creates work later.
| Metric | What it tells you | What distorts it |
|---|---|---|
| Rejection rate by object | Which file is actually the problem. A high rate on one object often causes failures everywhere downstream. | |
| Rehearsal count before cutover | Two or three full rehearsals is normal. Zero is how cutovers overrun. | |
| Elapsed load time per object | The input to the cutover plan, and only knowable by rehearsing. | |
| Reconciliation variance post-load | Counts and control totals compared to source. The only real proof the load worked. |
We build and support the systems this page describes. Start with a free working proof of concept — no cost, no commitment.