Host AI works your entire book — specialized offers for your premium players, and the same intelligence applied to the mid and low tiers nobody has time for.
Now onboarding a small number of properties this quarter.
A host's attention is finite — however good they are, there are only so many calls, comps, and birthday texts in a day. So the book gets worked from the top down: your premium players get the personal touch, and everyone else gets the mass blast, because the blast is the only thing that scales. Working the whole book gives every player a reason to keep coming back — instead of falling off, or finding another floor.
By the time a mid-tier player's absence is obvious, they've found another floor.
Your scarcest resource, allocated on feel, with no read on whether it returned.
The personal, specific, well-timed offer converts — which is exactly why your host can only make so many of them.
One file from your player system — players, trip history, redemptions, any column names — or an API connection we scope during onboarding. The system maps what you send and asks before touching anything sensitive. Refresh whenever you like; regular sends keep the picture current.
The tier-by-worth grid your marketing team already maintains — free play, food, rooms. The system never invents an offer: it selects from your matrix, at your value, or it doesn’t send. Change the matrix and the offers change the same day.
The system watches the whole book — who’s slipping, who’s lapsed, who just wrote in — and queues each one with the reason and a drafted offer, written in your property’s own voice and no-go rules (we lock those in with you at onboarding). Your host clicks approve. Or you set the dial below and decide how much doesn’t wait for the click.
And you set how much runs on its own. The system already watches, prices, and writes across your whole book — how much sends without a person is a dial, set per play — win-backs, trip-decay saves, inbound — and per tier:
Every message waits for a host. Where every property starts.
Offers inside your ceilings send on their own; anything above the line escalates to a host.
The system runs the plays you've signed off on, inside your limits, fully logged and sampled for review.
Set it per play and per tier, not once for everything. Keep every premium message host-approved forever while the volume plays earn a lighter touch — a permanent configuration, not a training-wheels stage.
Every edit your host makes is training. The queue is how the system learns your property’s voice — each play earns its next level from your own approval track record, and nothing moves up without your sign-off.
The players who quietly stopped coming are still sitting in your file — the cheapest revenue you’ll ever recover. The system finds every one of them and puts the right offer in front of them, sized to what they actually used to play.
A regular whose visits are slowing hasn’t left yet — that’s the save that costs a free dinner instead of a win-back campaign. The system spots the slip against their own baseline and gets the nudge out while they’re still yours.
A player writes in at 11pm on a Saturday — and gets answered in your voice before your host’s Monday coffee, not next Friday. Every conversation handled, none of them dropped.
Five stars or one, every review gets answered in your property’s voice — thanks for the praise, a measured answer for the complaint. Your public page always reads like someone’s home.
Every comp that goes out comes back with a number attached — who redeemed, what returned, what it earned. The proof gets its own section below.
Autonomy only works when the rails are absolute. These six are enforced by code — not by policy, not by training, not by a model's memory — and the ones that matter most have no setting attached.
Which is why the worst realistic outcome is a mistimed offer, to one player, at a value you'd already approved — recorded in a log you can read. The rails hold; the downside is a rounding error.
You can already see who redeemed an offer. What redemption can’t tell you is whether the offer made you money — whether that trip would have happened anyway, and whether the play behind it covered the comp. Host AI answers both: every player’s return is measured against their own prior baseline, and a slice of the book is deliberately held out — never contacted — so the lift you see is caused by the program, not coincidence. That’s the number a board can check.
Honest note: attribution on a floor is never perfect. We measure what’s observable and say what’s inferred. A number you can defend beats a number that flatters.
No per-player fees, no usage meters, no long contract. A call gets you a fixed monthly quote — for the whole system, your whole book.