Why is isolation the first question?
Isolation is the first question for any multi-tenant platform because the alternative to it is your data being one bug away from another customer's screen. For an AI platform it matters more, because the tenant data includes the AI employee's memory of your business, its drafts, and the credentials it holds for your systems.
The question to ask a vendor is not whether they isolate tenants, since every vendor says yes, but how, at which layer, and covering what.
What does isolation that holds look like?
Isolation that holds is enforced at the data layer. Every record carries its organization, and every query that reaches the database carries the organization it is acting for, so a request can only ever see rows belonging to that organization. There is no path that skips the scope, including internal tools and background jobs.
On ftes.ai this is how it works: every record is scoped to your organization, and queries that reach the database carry your organization, always. Background work such as measurement sweeps and briefs runs with the same scope as a user request.
What does isolation that only looks like it does look like?
Isolation that only looks like it does is filtering in the interface: the application shows you your data, but the underlying query could return more if asked differently, and a bug or a crafted request exposes it. Another weak form is isolation of the main tables but not of the AI layer: memory, embeddings, drafts, or logs shared across tenants.
Ask specifically about the AI layer. An AI employee's memory of your brand and its drafts of your content are tenant data, and they need the same scope as your customer records.
- Holds: scope on every query at the data layer, including jobs
- Holds: memory, drafts, logs, and credentials scoped the same way
- Weak: filtering in the interface only
- Weak: main tables isolated, AI layer shared
What is property isolation?
Property isolation is a second boundary within an organization. Websites, stores, and ad accounts are separate properties, each with its own memory, targets, and setup, and a Digital FTE working on one never sees another's. For an agency managing several clients' sites, or a group with several brands, this is the boundary that keeps one client's facts out of another's pages.
On ftes.ai property isolation is structural: one FTE's property never sees another's memory, targets, or setup, and connections are granted per property.
How should you verify it?
Verify it by asking for the mechanism in writing and, where the vendor allows, by testing: a request for a record that belongs to another organization should return nothing, as if it did not exist, rather than an error that confirms it does. Ask how background jobs are scoped, how the AI layer is scoped, and who inside the vendor can cross the boundary and under what controls.
The vendor security questions guide on ftes.ai lists the full set. A vendor who can describe the mechanism precisely has usually built it; one who can only say that tenants are isolated has usually not been asked before.
What questions should you ask, and what answers should you expect?
Ask at which layer isolation is enforced, and expect the answer to be the data layer, with every record carrying its organization and every query carrying the organization it acts for. Ask whether background jobs run with the same scope as user requests, and expect yes, because measurement sweeps, briefs, and publishing jobs touch tenant data too. Ask whether the AI layer is isolated the same way, meaning memory, drafts, embeddings, and logs, and expect yes, because a Digital FTE's memory of your business is your data.
Ask who inside the vendor can cross the boundary, under what controls, and whether that access is logged. Ask whether properties within an organization are separated, and expect yes if you are an agency or a group with brands that must not share facts. Ask how you could verify it, and expect an answer that involves a request for another organization's record returning nothing rather than an error. On ftes.ai these are the answers its security page gives. Any vendor that cannot answer the first question with the words data layer has not built what the buyer is asking about.