Switching HR systems without losing data

Most HR migrations go wrong in the same two places: something that looked like data turned out to be a process, and something that looked exportable turned out to be a screen. Neither is dramatic. You discover in week three that leave balances did not come across, that the salary history behind each person is gone, or that the offboarding you thought was one workflow was actually three spread over modules you have already cancelled.

This guide is the sequence that avoids that. It applies whichever tool you are leaving, and the tool-specific guides underneath it cover the parts that differ.

Decide what you are actually moving

Before you export anything, list what you hold and where it lives. Most teams find four or five piles rather than one:

• The directory. Who works here, in what capacity, from where, on what terms. This is your master index and the thing everything else is checked against.
• Documents. Contracts, offer letters, tax and identity forms, policy acknowledgements, anything uploaded to a profile.
• History. The effective-dated chain behind fields that change: salary, job title, department, status. This is the pile people forget, because it is invisible in a normal export.
• Balances and policies. Leave balances, accrual rules, holiday calendars, working weeks.
• Whatever is in an add-on. Payroll runs, timesheets, benefit elections, device records, spend. Read your invoice, not your memory: if you are paying for it, it holds data.

If you cannot say which pile a fact belongs to, that is the fact that will go missing.

Export in order, while the account is live

Do this before you give notice. In every tool, access to reports and documents is a function of your subscription, and a former customer is a lower support priority everywhere.

1. Directory first. A full spreadsheet export with status, role, country, start date and manager. Everything that follows gets reconciled against this list.
2. Documents second, and test the mechanism on one profile before you start. If documents come out per person rather than as an archive, your timeline is a function of headcount and you need to know that on day one.
3. History third. Run whatever history or change report exists, separately from the directory export. If none exists, decide now whether you need the history badly enough to capture it another way, because after cancellation the answer is no.
4. Balances and policies fourth. Export current balances and write the accrual rules down in plain words. You will be re-entering these by hand.
5. Add-ons last, one at a time. Each one exports separately. Tick them off the invoice.

Then reconcile: does the document count match your headcount, do the exported balances match what people believe they have, does the payment or payroll history agree with your accounting records. Do this while you can still ask someone.

Expect three things to be manual

Be realistic about import, because this is where schedules slip.

Leave balances almost never import. A balance is meaningless without the accrual policy that produced it, and no two systems model accrual identically. Plan to configure the policies in the new system and load opening balances by hand, then check a sample of people individually before you trust the set.

History rarely imports. Most systems will take a current-state directory and nothing behind it. If your salary history matters for audit or for a future dispute, keep the exported history reports as your archive rather than expecting the new tool to absorb them.

Documents are re-filed, not migrated. You attach them, person by person, against records that now exist. There is no clever way around this and it is the single largest time cost in most moves.

It is worth being specific rather than reassuring here, including about our own product. Capstan imports employees well, with a preview-and-invite flow, and that is where its import framework currently ends. There is no file import for org structure, leave balances, rosters, projects, assets or contractors, and holiday calendars are bundled sets for a handful of countries rather than an upload. If your migration depends on bulk-loading any of those, ask before you plan around it, and ask every other vendor the same question in the same words. “We support import” usually means employees.

Loading the data is only half of the move. The other half is configuring the system it lands in, in an order that does not force rework, and that is a sequence of its own: the implementation checklist for the new system covers what to set up before you load anybody, what to leave until the data is in, and what to test before you invite the team.

Run an overlap, then cancel

Keep both systems live for one full cycle. A cycle is a complete month of leave requests and approvals, one pay period handed to whoever runs your payroll, and any recurring compliance task that falls in the window. Compare the outputs. One extra month of a subscription is cheap next to finding out in week three that something you relied on exists nowhere.

Only after that overlap has run clean, and after every document and history report is safely downloaded to a dated folder you control, do you cancel.

Test the new exit before you commit to the new entrance

You are switching, at least partly, because getting data out of the last tool was harder than it should have been. Do not repeat the mistake in the other direction. Ask four questions before you sign, and hold the answers against what you have just been through:

1. What container does the export arrive in? A plain, documented archive that any machine can open, or something proprietary.
2. Does it include the original document files, or only their metadata? An export of document names is not an export of documents.
3. Is there a manifest you can check the archive against? A list of each entity and its row count turns “we exported everything” into something you can verify in a minute.
4. Does the export still work if a payment lapses? This is the one that matters, because a billing dispute is exactly the moment you will want your data, and it is exactly the moment a badly designed system will refuse.

For what it is worth, that is the test Capstan is built to pass: the export is a plain archive holding documented JSON as the authoritative form, a flattened CSV of the same rows, and the original file of every document, with a manifest listing each entity and its count. The export endpoints deliberately sit outside the billing check, so a workspace that is suspended for non-payment can still take a complete copy and leave. Ask any vendor to state their equivalent in the same detail.

Where to go next

For the tool you are actually leaving, the specific guides cover what differs: leaving Deel for contractor-heavy teams with an employer of record in the mix, leaving Rippling for untangling HR from IT and finance, leaving Gusto for timing a move around payroll and year-end filings, and leaving BambooHR for pulling effective-dated history out of a mature HRIS.

If you have not yet chosen the replacement, start with the HR software for startups guide, which covers what a small team actually needs and the packaging patterns worth avoiding.