I run plant monitoring for six customers this way, so here’s the setup that’s held up.
One organisation per customer company. You keep one account and are a member of each, with Admin, and switch between them from the organisation picker. The customer’s people are invited as User. A User can see devices, profiles, dashboards and alarms, acknowledge alarms, send basic commands and build personal boards. Viewing flows needs Admin (the permission is automation.read), so your Script block stays out of their sight. Never give a customer Admin “just for a minute”.
Dashboards: build them once in a template org, download the board as a file and import it into each customer org. Flows don’t travel in the board file, so you bring the flow over separately: Export it from the row menu on the flow list and Import it in each customer org (secrets have to be typed in again, and imported flows arrive switched off). With 4 orgs I keep the Script block code in git as well, it’s a 10-minute job per change. If that gets painful, think about moving the dosing calculation into the firmware and sending dose_g as telemetry; then the logic lives in one signed image.
Firmware: what Laura said, sign with Ed25519 in your pipeline, register the public key on the Firmware page, and leave Allow unsigned images off. Unsigned or re-signed images are then refused by the upload check.
The role table is on Users and roles. Branding with your logo is per organisation too, if you want it to look like your product.