For a law firm, insurer, or bank, where documents live matters as much as what the AI can do with them. There are three broad deployment models for document intelligence, and they trade convenience against control in different ways.
Shared SaaS
The simplest model: many customers share the same application and infrastructure, separated by software permissions. It is fast to start and cheap to run. The trade-off is that your documents sit on shared infrastructure, and you rely on the provider's logical separation rather than a physical boundary. For some teams that is fine; for others it is a non-starter.
Isolated cloud account
A stronger boundary: the application runs in a cloud account dedicated to you, not shared with other customers. Your data, your keys, your logs - separated at the account level, not just by a permission flag. This is the model InVerba uses for enterprise clients: a dedicated AWS account per client, with data held in the EU (AWS eu-central-1, Frankfurt) and encryption keys you control.
On-premise / your own account
The strongest control: the system is deployed inside infrastructure you own, and documents never leave it. This suits organisations with the strictest data-sovereignty requirements. It asks more of your own operations team in return for maximum control.
How to choose
The right model follows from your obligations, not from a preference for "cloud" or "on-prem" in the abstract. A few questions usually settle it:
- Does a regulator or client contract require a specific data location or isolation level?
- Do you need to hold your own encryption keys?
- What audit evidence must you be able to produce, and how often?
- Who is expected to operate and patch the system?
Whichever model you choose, the controls that matter are the same: encryption in transit and at rest, data residency you can point to, complete audit logs, and a signed data processing agreement. The deployment model decides how strong the boundary around those controls is.