What data would this system access? Where would that data be processed and stored?

The longer version

This is the question every business owner should ask before signing anything, and most don't — because most consultants can't answer it precisely. "The AI needs access to your systems" is not an answer. Which systems? Read or write? Does data leave your environment? Does it cross borders? Stored for how long, by whom?

A serious answer looks like a list, not a paragraph. For a typical small-business automation, it might be: read access to the job schedule in JobTread, write access to nothing except a draft folder someone reviews, processing inside the vendor's US data centers, no retention beyond 30 days. That's a boundary you can evaluate. "Don't worry, it's secure" is not.

The reason this matters beyond caution: it's also how you get value faster. When the boundary is explicit, there's no six-week security review, no nervous owner second-guessing every step. Everyone knows what the system touches, so everyone relaxes and the thing actually gets used.

How BIS handles this

I build every engagement around a data-access boundary document, and it's the core of my boundary offer: default-deny (the system starts with zero access and earns each permission), three tiers — Lite, Standard, Regulated — sized to how sensitive your data is, and every boundary labeled by how it's enforced: technically, procedurally, or documented. You get the document before we build, not after.