“We have OA” is sometimes self-comfort. Open it: contracts are not enabled, seals say no permission, leave fields are another industry’s “annual-leave quota.” People vote with their feet and go back to WeChat. You think it is better than none; they think it is more half-hearted than none—because you showed a system that cannot be used.
Twenty tiles on the home page; half open into demo data.
Approval nodes follow the product preset; changing one means a ticket to the vendor.
The phone shell opens; attachments and countersign still need a PC.
Permissions are ticked by name; one transfer scrambles them.
A false signal hurts more than a blank, because you chose to go live
A blank at least lets people think digitalization has not started. An unclickable package tells the floor the process bar is low. Feishu, DingTalk, and custom differ on whether exceptions can change; see Feishu, DingTalk, and a custom back office: what actually differs. If the system made work slower, diagnose fields and nodes first; see The system made work slower: how to diagnose it. When the annual fee still bills and work has returned to WeChat, the loss is harder to see than never buying—because the books say “we already have OA.”
The fix is not a fuller module tree. Return to the work: phase one opens only the three–five flows that actually run, fields match your tickets, permissions follow roles. Treat the system as a work entry, not a display case; see Filing cabinet vs. work entry: two ways to treat the system. If changing one node means a vendor ticket, daily exceptions return to the group, and the package is as bad as none.
Do you need a full ERP? See Do you need a full ERP?. Extend or rebuild: see Should you extend a legacy OA or rebuild it?. To replace a package with a maintainable back office, go to the Management systems service page with live screenshots. Four approvals people can actually walk beat twelve gray tiles.