Keeping the AI bill in check: usage limits, accounts and who approves what

Controlling AI usage cost in a hotel is a monthly routine, not a negotiation: list every AI account and who owns it, set a limit per team and per tool, review the bill against actual use, require one named approver for new tools and upgrades, and ban shared logins so every charge and every action has a person behind it.

The card statement arrives with four AI subscriptions nobody remembers approving, and one of the logins is used by six people on three shifts. The bill is not the real problem. The missing owner is. After reading this you will be able to run a monthly control loop, build an account inventory, set limits per team and per workflow, record approvals, and exit a tool without losing the work.

What does keeping AI usage cost under control actually mean?#

It means running a short repeating loop: inventory, limits, review, approvals and exit. Each step leaves one artifact behind, so the next month starts from evidence instead of memory.

The five artifacts are a sheet listing every account, a written cap per team, a monthly review note, a recorded approval, and an export taken before any cancellation. None of them is software. All of them fit in one spreadsheet and one page of policy.

The two billing shapes matter more than the vendors. A per-seat plan charges for each person who can log in. A usage plan charges for what the tool does, counted in messages, credits, minutes or documents. The first is controlled by headcount: remove the person, remove the seat. The second is controlled only by caps and a monthly review, because the meter moves without anyone signing anything. Read more on how AI tools are priced: seats, usage and what drives the bill.

Give the loop one owner. In most hotels the general manager or the IT lead keeps the sheet, and the accountant checks the invoice against it before paying. Two people, one sheet. Treat control as direction, not refusal: a hotel that bans every tool pushes the same work onto personal accounts, where nothing is visible and nothing can be cancelled.

Count the costs the invoice does not show. Staff time spent re-checking output is a cost. Guest details pasted into the wrong tool is a risk. Moving to another tool later is a project. Run the loop monthly, because an annual review is too slow for tools that bill by use and renew automatically.

How do you build an AI account inventory for a hotel?#

One sheet, one row per account, with the fields that let someone act: who holds it, who pays, how it is billed, who uses it, what data it receives and when it renews.

Record one row per account, not one row per vendor. The same vendor can appear twice, for example once as a team plan and once as a single seat bought on a card. Write the name of the person who can log in, never the department. A row that says "front office" cannot be acted on when that person leaves; a row with a name can be closed in five minutes.

Include the invisible accounts. Free tiers that convert to paid, trials that renew on their own, AI features bundled inside another subscription, and app-store subscriptions on staff phones all belong on the sheet. Record the payment route, because it decides who sees the charge and when: hotel card, personal card reimbursed, or annual invoice.

Add a data field: what kind of information the tool receives, such as public text, internal documents or guest data. This field decides who must approve any change. Add an exit field too: how to export the work, the notice period, and whether the plan renews automatically. Date every change and keep the sheet where the general manager and the accountant both see it.

If you want a deeper version of this sheet, see an AI register for hotels: every AI use, its owner and its risk tier on one sheet.

What usage limits should a hotel set, and per what?#

Set two limits: a seat limit per team, which is how many named people may hold a login, and a usage cap per workflow, which is how much work the tool may do in a month. Then write down what happens when a cap is reached.

The seat limit is the cheapest control and the easiest to audit. It is a number on the sheet, and it changes only when someone asks. The usage cap is set per workflow, not per person: for example, a cap on documents summarized per month for the sales team, or a cap on assistant messages for reservations.

Use the tool's own meter in the policy. If the product counts messages, credits or minutes, write that word, so staff recognize the warning when it appears. Choose hard or soft for each cap: a hard cap stops the tool, a soft cap warns the owner. Use soft caps everywhere except where a stop would be harmless.

Write the exception first. Front desk and reservations must never lose a tool mid-shift, so those workflows get a higher cap and a named person who can raise it within minutes. Pick the starting number from last month's actual use, not from ambition. A cap set at twice current use teaches nobody anything. Decide what a breach triggers: a message to the sheet owner, a review within two working days, and a decision written down.

Limits work better when staff understand what the tools may and may not do with hotel data, which is covered in training hotel staff to use AI: a short in-house program by role.

Why are shared logins a problem for the bill and for the data?#

A shared login makes the bill unattributable and the history anonymous, and it outlives the person who should have lost access months ago.

When usage doubles, nobody can say which shift or which team caused it, so the only available response is to cut the tool for everyone. Offboarding breaks quietly: you cannot remove one person without changing the password for all of them, so in practice nobody changes it.

The data surface widens too. Chat history holds whatever anyone pasted, including guest details, and every holder of the password can read it. It usually breaks the vendor's own terms, which puts the account at risk exactly when the hotel depends on it. And the practical question is the audit one: if a guest asks what happened to their data, a shared account cannot say who did what. A hotel policy on what staff may share with AI chat tools is the companion rule.

Replace shared logins with named seats, a shared team workspace where the vendor offers one, and a written handover when a person leaves. Weigh the cost of a seat against the cost of one unattributable incident. The seat is usually the cheaper line.

Who approves what, and how is the decision recorded?#

Split approval by consequence, not by amount alone: a routine renewal, a new tool, a price change and a change in the data a tool receives are four different decisions with four different owners.

One named approver per decision. A committee delays the answer and hides who said yes. Ask for a one-line request: what the tool does, why now, the monthly cost, and what data it will receive.

Keep the data decision with the person who owns the data, and take accounting or legal advice before any new category of personal data goes to a tool. Record the approval on the same sheet as the inventory, with the name and the date, so next month's review can see it. Give the accountant the authority to stop any payment that has no row on the sheet. That single rule catches most leaks.

Set a small monthly amount a team lead may spend without asking, and hold everything above it for the general manager. Review the approval table once a year, or whenever a tool changes what it does with the data.

How do you review the AI bill every month?#

Thirty minutes on a fixed day, comparing three things: the invoice lines, the account sheet and the seats actually used in the last thirty days.

Fix the day and put it before the renewal dates, not after the charge lands. Chase every invoice line with no row, and every row with no use. Both are errors, and one of them costs money.

Find the tool whose usage moved most and ask the team what changed. A jump is either a new workflow worth keeping or a mistake worth stopping. Ask each team lead one question: what did this tool do last month that you would miss? Silence is an answer.

Decide one of four things for each tool: keep, downgrade, cancel, investigate. Write the decision and the date on the sheet. Keep the note to three lines per tool, so next month's comparison takes minutes rather than an afternoon. Send the note to the accountant before the next payment run, not after it.

How do you exit an AI tool, and what belongs in the one-page policy?#

Plan the exit when you sign, and put the rules on one page that staff actually read. An exit without an export is a deletion.

The exit checklist#

Agree the exit terms at signing: export format, notice period, auto-renewal, and who confirms the exported file opens. Export before you cancel, and open the file yourself. A tool that closes the account on the last day of the notice period can take the history with it.

List everyone who holds a login, including people who have left, and close each one. Remove the card or the invoice route on the same day, and mark the row closed rather than deleting it, so the history stays readable. Keep the outputs that matter in the hotel's own storage, under the hotel's own naming, not inside the tool you are leaving.

The policy line#

Write the policy in plain sentences: named accounts only, approved tools only, no guest data in unapproved tools, a cap per team, one approver, a monthly review, and report a mistake the same day. Publish the page where staff already look, and repeat it at induction rather than sending it once.

What to do next#

Open one sheet this week, add every AI account you can find, and write the one-page policy line above it. The inventory, the limits, the monthly review, the approvals and the exit all hang from that sheet, and a hotel without it is negotiating with its own card statement.