Every marketing platform now has an AI feature, and most of the conversation is about what it can do. The more useful question, especially for software that spends money and sends messages on a brand's behalf, is what it should be allowed to do on its own.
We build FORTVISION and run client accounts in it every day, so we have had to answer that question concretely. This is where we drew the line, and the reasoning behind it.
The principle: reversible work is fine, irreversible work needs a person
A draft that turns out to be wrong costs a minute to fix. An email sent to two hundred thousand people cannot be unsent. A budget raised by mistake overnight is money already spent. A deleted list is gone. Almost every decision about AI in the product follows from that asymmetry.
AI can do as much as it usefully can on work that is cheap to undo. Anything that is expensive or impossible to undo goes through a person who sees exactly what will happen before it happens.
What AI does in FORTVISION
Drafting
The AI assistant writes subject lines, email and SMS copy, ad text variants and landing page sections. It works from the brand's existing material and the brief, and its output is a draft in the same editor a person would use. Drafts never send. Nothing reaches a customer because a model wrote it.
Analysis
Asking “which campaigns lost money last week?” or “why did revenue per recipient drop in October?” in plain language, and getting an answer built from the account's own data, is where AI saves the most time. The dashboard's suggested actions come from the same place: a short list of things worth looking at today, each with the numbers behind it.
Building segments
“Customers who bought twice in the last year but nothing in the last ninety days, excluding anyone who unsubscribed from SMS” becomes a segment definition you can read, check and edit before anything uses it.
Conversations, inside rules
AI agents answer calls and chats, qualify leads and book appointments. They work inside rules the business sets: what they may discuss, what they may promise, and when to hand over to a person. Every conversation is logged to the contact, so a human can see exactly what was said.
What AI does not do on its own
- Press send. Sending a campaign, a sequence or a bulk message is a separate, deliberate action from creating it.
- Move money. Changing campaign budgets or bids, pausing or launching paid campaigns.
- Delete. Contacts, lists, campaigns or anything else that cannot be restored.
- Make bulk changes. Updating hundreds of records at once.
- Make commitments. Prices, discounts or terms that the business has not configured.
How the line is enforced
A rule that lives only in a prompt is a suggestion. The line has to be in the software itself, so it holds whether the request comes from our own assistant or from an outside AI tool connected to the platform.
- Read freely. Looking at data has no side effects, so AI can explore an account without asking.
- Write drafts. Creating or changing a single record, or saving a draft, is allowed and logged.
- Preview, then confirm. High-impact actions, such as sending, deleting, bulk changes and budgets, do not run when first requested. They return a preview of exactly what would happen: who receives what, which records change, how much spend moves. The action runs only after a person confirms that exact preview. Change anything and it has to be previewed again.
- Scoped access. Connections act inside one workspace with the permissions they were granted, so an assistant set up for one brand cannot touch another.
- No double actions. Writes carry a unique key, so a retry after a timeout cannot send or charge twice.
Why not let it do more?
The models are good, and they keep getting better. The reason for the line is not that AI makes more mistakes than people. It is that the cost of a mistake falls on the brand and its customers, and accountability has to sit with someone who can be asked why.
There is a practical reason too. Most of the time saved by AI is in drafting and analysis, where the work is slow for people and cheap to check. The final click to send or approve takes seconds. Automating it saves almost nothing and removes the last point where someone looks at the whole thing with fresh eyes.
Where the line might move
Some actions sit closer to the edge than others. Pausing an ad that is clearly overspending, or stopping a sequence for a customer who has just complained, are cheap to reverse and expensive to delay. Those are the cases where we expect to allow more automation over time, inside limits each business sets: a maximum change, a narrow trigger, a notification every time it happens. The principle stays the same: the less reversible the action, the more a person needs to see it first.
In short: in FORTVISION, AI drafts, analyses, builds segments and holds conversations inside your rules. Sending, spending, deleting and bulk changes are previewed and confirmed by a person, every time.