How to classify product data
Public, internal, regulated, and irreversible. A plain classification so hosting choices stop being vague.
Edition 1Revised Why editions?
Hosting arguments get noisy when every dataset is treated the same. Classification is how you stop that. You sort the data first. Then you choose where it can live.
This page is for product and infra leads who already asked the sovereignty questions and now need a shared vocabulary on the team.
For the decision frame around location, access, and exit, start with data sovereignty in plain language.
The four classes
Public
Marketing pages, open docs, published OSS, anything already meant for the open web. Leak risk is low because the intent is public. Still keep integrity and takedown paths clear.
Internal
Source drafts, internal notes, non-customer ops data, staging fixtures with no real people in them. Limit who can read it. Do not put it on a random personal drive because "it is only internal."
Regulated
Personal data, payment-related records, education or health adjacent records, anything a law or contract can name. Location, retention, and access logs matter here. Vendor answers must be specific.
Irreversible
The data you cannot replace if it burns: production customer records, unique research, signed contracts, the only copy of a model or dataset, credentials that unlock everything else. This class gets the strictest hosting bar.
If a dataset could sit in two classes, use the stricter one.
How to use the labels
Write the class next to each system in your inventory. One line is enough: "Billing DB: irreversible. Blog CMS: public. Support inbox: regulated."
Then match hosting to the class.
- Public can stay on ordinary SaaS if export still works.
- Internal needs access control and a named owner.
- Regulated needs a clear region story and a retention rule.
- Irreversible needs hosting you can explain in one paragraph, off-box backups, and a restore drill.
A common mistake
Teams spend months arguing about clouds in the abstract. The faster path is classify first. Once irreversible data is named, the next move is usually obvious: protect that system before you redesign everything else.
When you are ready for that move, read more: self-hosting in Nigeria.