Here is the failure this sequence exists to prevent. A team exports 4,000 files from the old room, uploads them into the new one, invites everyone from the old participant list and opens the room. Two days later a bidder finds a folder of employee compensation data that was restricted to file names in the old room and inherited the new room's group default. Fixing the permission takes a minute. What does not get fixed is that a bidder has now seen it, and your access record has a period of over-disclosure sitting in the middle of it.
The sequence
Order matters more than tooling. Structure before files, files before permissions, permissions before invitations.
- 1. InventoryExport the file list, folder tree, participant list, group memberships and audit trail from the old room. Before you cancel anything.
- 2. Rebuild the treeRecreate the folder structure empty, keeping the old numbering. Naming is load-bearing.
- 3. UploadFiles land in folders that already exist, so nothing has to be moved afterwards.
- 4. Rebuild permissionsThe heavy step. Groups first, then folder overrides, then per-document exceptions. Budget more time here than for everything above combined.
- 5. Verify from a seatLog in as a member of each group and walk the tree from their view, not from an admin view.
- 6. Invite, then openInvitations go last, so nobody sees the room in a half-configured state.
What actually transfers
One row in this table moves on its own. The rest are rebuilt by a person, which is the whole reason a migration takes a week rather than an afternoon.
| Criterion | Comes with the files? | What breaks if you skip it |
|---|---|---|
| The documents | Yes | Nothing. This is the part every team gets right, and it is why migrations get underestimated. |
| Folder tree and numbering | No | Every line in the diligence request list and every reference in the Q&A thread points at a path that no longer exists. |
| Group definitions | No | You rebuild them from memory, and the group that carried one narrow exception quietly loses it. |
| Folder and document overrides | No | The exception is the entire point of the override. Losing one is silent, and it fails open rather than closed. |
| Who saw what, and when | No | Your access record restarts. For everything before the switch you now depend on an export from a vendor you have stopped paying. |
| Open Q&A threads | No | Answered questions become unanswered ones, and the counterparty asks them a second time. |
Build the folder tree before you upload anything
Recreating the structure first turns the upload into a single pass. Uploading first and organising afterwards turns it into thousands of individual moves, each one a chance to drop a file into a folder with a different permission.
Keep the old numbering, including the parts you dislike. If the previous room used a numbered index, request lists and open questions refer to those numbers, and renaming 3.2 to something more descriptive during a migration breaks every one of those references at the exact moment people are relying on them. Rename after close, not during.
Permission mapping is a translation, not a copy
Every vendor names its tiers differently and the tiers do not line up one to one, so write the mapping down before you configure anything: a row per group, a column per folder branch, and the tier each combination should land on. Two rules keep it safe. Map to the lower tier whenever the translation is ambiguous, because under-granting produces a support request while over-granting produces a disclosure you cannot withdraw. And map groups, not individuals, because per-person exceptions are how a room becomes impossible to reason about six weeks later.
Sifrsys resolves five levels, most specific first: a per-document override beats a folder override, which beats the group default. The levels are none, fence, view, download and full, and download and print are separate switches from the view level, so “can read it” and “can keep a copy” are configured independently. If you are translating from a product that bundles those together, that separation is the first place your mapping will need a decision. The tier most often mistranslated is the name-only one, which is the subject of what fence view does and does not prevent.
Export the old audit trail before you cancel
This is the only step with a hard deadline. Access to the old room's log usually ends with the subscription, and that log covers the period when diligence actually began. Pull it as CSV or JSON, with whatever integrity hashes the vendor provides, and store it with the deal file. If you cannot export it, you are about to lose it, and a gap in the record sits exactly where the disclosure argument would start. The properties that make an audit trail defensible are worth checking on the new room too, before you commit.
Verify from a participant's seat, not an admin's
An admin view shows you the configuration you intended. A participant view shows you the configuration you have. Log in as one member of each group, walk the tree, and confirm that the folders which should be invisible are invisible and the ones that should be name-only show a name and nothing else. Do this before invitations go out, because an invitation is the point at which a mistake stops being reversible.
The offer that exists because of step 4
For the Founding 10, the first ten paying companies, we do the migration ourselves at no charge: documents, folder structure and permission setup out of your current data room. Those companies also keep their signup price for the life of the subscription. The migration offer exists specifically because the permission rebuild is the step that does not scale down, and it is not work a buyer should have to do twice to find out whether a platform suits them. Terms are on the pricing page, alongside the flat plans and the 14-day trial that needs no credit card.
Before you migrate anything, it is worth running the vendor through the security checklist. Moving twice costs more than asking six questions once.