It easily happens that employees do whatever it takes to get their internal tool to show the right thing, or use an automated tool, without ensuring they don't destroy or duplicate existing data. (made-up example: someone is refining the mapping of streets, but destroys information about adjacent/connected footpaths in the process because they don't matter for their delivery-vehicle use case. Or maps street types as they would in country A, but in country B a different norm has been established, so the edits don't fit the surrounding data)
Data quality can be a problem, where the company data source does not match reality (anymore). Or it's difficult for the community to verify your data and be convinced your quality is good, especially if there are other concerns about your edits.
OpenStreetMap operates on soft consensus and can have multiple acceptable ways of doing a thing. Mass-editing that converts one form to other is frowned upon unless wide agreement that one way is deprecated has been sought. Often related to the first point, when company tooling prefers one style or employees only have been taught about one method, but in an area the other mapping style is widely used.
Similarly, communication to resolve conflicts can be lacking: if another editor challenges what you are doing (because they think you made a mistake or are overstepping norms), you need to react and fairly consider their point. OSM has clear expectations that if you are doing coordinated edits you clearly identify and announce your work and respect feedback.
Data sources/licenses can be a concern: what you find acceptable to import into your company database might not be acceptable to include into OpenStreetMap, and how does the community tell that your data is good and won't get the project in trouble later?
E.g. by overwhelming a local community with edits from a swarm of outsourced contractors from a wholly different country. Good luck keeping up with verification of their work.
Or making mass edits / imports without discussing with local community beforehand.
Or as in the first point, but with "AI" as FB did.
