ARRIVE AI is six separate features, not one. Each has its own place in the product, its own switch, and its own limits. Two of them never send anything to a model at all — they are arithmetic over your own records, and they are labelled below.
Everything here reads only your organization's own data. Nothing here writes to it.
The six features
| Feature | Where you'll find it | What it produces |
|---|---|---|
| Daily Briefing | Dashboard | A short written read of your own numbers for today |
| Recommendations | Dashboard | Cards for thresholds your own data has crossed — no model involved |
| Insights | Reports | A line of context beside a figure that moved unusually — no model involved |
| Ask bar | Dashboard, and the "Ask ARRIVE AI" pill in the top bar | An answer to a question about your organization |
| Workspace assistants | The AI Assistant tab on an event, location, program or performer | An answer about that one record |
| Report summaries | Reports, as Executive Summary | A paragraph describing the report below it |
Those are the exact names used in Settings → AI, so a switch there always matches a thing you have seen on screen.
Daily Briefing
On your dashboard, under the heading Daily Briefing — "What your organization's own numbers say today."
Nothing is generated when you open the page. One briefing is written per organization per day, shared with everyone, and somebody with AI access has to ask for it. Opening the dashboard only reads what is already there. If today's has not been written yet, a member with AI access sees a Generate today's briefing button; a view-only member is told a colleague can write it.
Because it is written once and then cached, it can be hours old. That is why the time it was generated leads the card rather than sitting in the small print, and why the live counts stay on screen above it.
Recommendations
On your dashboard, under Recommendations — "Thresholds your organization's own data has crossed."
No model writes any part of this. Every card is produced by a fixed rule over your own records, and every card will show you the figure, the line it crossed, and how many records that was based on. Expand Why am I seeing this? on any card.
The thresholds are your organization's own, not ARRIVE's opinion of how you should operate — an event you scheduled with nothing booked against it, a follow-up date you set on a contact, a performer you marked as one to work with, your own distribution of booking value across your own locations.
If nothing has crossed a threshold, the panel says so. It does not invent something to say.
You can Set aside a card you have dealt with; it returns if it becomes more urgent. A view-only member sees no Set aside control at all.
Insights
On the Reports page, between the figures and the tables.
No model writes any part of this either. An insight says one thing: a measure moved further than your own history says it usually moves. It does not say whether that is good, does not say what caused it, does not predict what happens next, and never suggests a next step. How this is measured on each card shows both periods, both figures, the sample size and the range it was compared against.
Most of the time there is nothing to flag, and the page says exactly that rather than manufacturing a card.
Ask bar
Ask ARRIVE AI, on your dashboard and from the top bar.
The bar answers questions on a closed set of topics — your revenue, bookings, upcoming events, locations, performers, programs and organization health. That list is fixed in the product. It is not a suggestion of what to try; it is the complete set.
A question outside that set is refused, not guessed at. You get a plain sentence saying it can't be answered from your data, together with the topics that can be. Opportunities is a topic the bar understands but deliberately does not answer, because the underlying figures are not part of what it is given — so it says so instead of improvising.
Text that reads as an instruction rather than a question ("ignore your instructions and…") is rejected before it reaches the model at all.
Every answer carries a Based on… line naming what it was derived from. The one exception: if your organization has no events and no bookings yet, the bar answers with a plain count of your locations and team members instead. That answer has no Based on… line, and it usually does not use one of your 25 daily requests.
Workspace assistants
An AI Assistant tab on four record types: an event, a location, a program, and a performer. Each asks about that one record and nothing else.
There is more on these in Asking ARRIVE AI about a record.
Report summaries
On the Reports page, under Executive Summary — "What this report says, in a paragraph."
Like the briefing: one per organization per day, shared, and never generated just because someone opened the page. Someone with AI access asks for it.
The disclaimer on this one is deliberately stronger than elsewhere. The summary narrates across many figures, several of which count bookings over different statuses and different windows, so a correct number can still end up in the wrong sentence. Read a sentence against the table it describes.
The four things ARRIVE AI never does
These four guarantees hold for every feature above:
- It never writes to your data. No AI feature creates, edits or deletes an event, booking, program, contact or anything else. Its output is text for you to read.
- It never moves money. It cannot charge, refund, transfer, or change an amount.
- It never sends anything to anyone. No email, no notification, no message leaves your organization because of it.
- It never acts on its own. Output is a draft. Nothing is carried out because ARRIVE AI produced it.
These are not settings, and they are not policy statements sitting in a document somewhere. In the code, each of them is a value that can only ever be "no" — an AI feature that wrote your data or moved your money could not be built without the change being rejected outright.
What this does not do yet
- There is no AI on Contacts or Messages. A contact record is largely your own notes and someone's contact details, which ARRIVE AI is never given.
- The Ask bar's topic list is fixed. New topics arrive by being built and reviewed, not by rephrasing.
- Nothing generates on a schedule. Every briefing, summary and answer is asked for by a person.
Related: Turning ARRIVE AI on and off · ARRIVE AI: what it does, how to switch it off, and what it keeps