Use cases

When does AgentGate protect you?

Six real situations where an AI can spend money — and how AgentGate keeps you in control without blocking the automation that saves you time.

01
Marketing

An agent that manages your online ads

The situation

You use an agent to run your Meta or Google ad campaigns. It adjusts budgets, creates new ads, and pauses underperformers — 24/7. Very convenient, but a bug or misconfiguration can spend €5,000 overnight without anyone noticing.

The risk

Without a safeguard, a model error or a badly-written instruction can trigger massive ad spend on an account you're not monitoring around the clock.

What changes

The agent runs 90% of campaigns on its own. You only step in for unusual cases — around 5 to 10 times per week, from your phone. No budget overrun in 3 months.

Rules configured
  • BlockedSpend blocked if it exceeds €500 per day
  • Needs approvalAny ad spend over €150 at once → you approve first
  • Needs approvalUnknown new ad network → manual review
  • AutoRegular weekday campaigns between 8am and 10pm → auto-approved
02
Software & Cloud

An agent that tops up your AI subscriptions and credits

The situation

Your agent monitors your cloud usage and automatically tops up OpenAI, Anthropic, or AWS credits when they run low. Handy for avoiding outages — but an infinite loop or a bug can trigger repeated top-ups within minutes.

The risk

Cloud platforms have no per-top-up limit. Your agent can trigger 20 top-ups of €500 in an hour if something goes wrong in the code.

What changes

An infinite loop was detected and blocked at €400 of spend — instead of the thousands it could have cost. The auto-approval rate stays at 87%.

Rules configured
  • Needs approvalTop-up over €200 → your approval first
  • BlockedMore than €1,000 in top-ups per day, even if the agent keeps asking
  • BlockedReserved capacity purchases (very expensive) → always blocked
  • AutoSmall routine top-ups → pass automatically
03
Customer support

An agent that issues refunds automatically

The situation

You have an AI agent handling refund requests. It reads customer messages, checks orders, and decides whether to refund. Efficient for small disputes — but some customers know how to phrase messages to get more than they're entitled to.

The risk

A savvy customer can write a convincing message to get a refund far higher than their order value. Without a hard limit, the agent might comply.

What changes

78% of refunds handled in under 2 minutes. Average resolution time dropped from 8h to 12min. Three suspicious patterns spotted through the history before they became incidents.

Rules configured
  • AutoRefund under €50 → processed immediately
  • Needs approvalBetween €50 and €300 → a supervisor reviews
  • BlockedOver €300 → never automatic, always handled manually
  • BlockedCap of €1,000 total refunds per day per agent
04
Finance & Investment

An agent that manages transfers or investments

The situation

Your agent watches market signals and executes transfers or rebalancing automatically. Useful for fast reactions — but a configuration error can send a large sum to the wrong place. And with some assets, that's irreversible.

The risk

Unlike regular payments, some transfers can't be cancelled. A bug or a wrong address can cost a fortune with no way to get the funds back.

What changes

No unauthorized movements since deployment. The automatic history serves as compliance evidence. Partners can independently verify every transaction.

Rules configured
  • BlockedAny transfer to a non-pre-approved account → blocked
  • Needs approvalTransfer over €5,000 → two people must approve
  • Needs approvalAny movement outside business hours → mandatory review
  • BlockedHard cap of €25,000 in outflows per day, no exceptions
05
Purchasing & Supply chain

An agent that restocks your inventory automatically

The situation

Your agent monitors stock levels and places orders with your suppliers when a product runs low. You save time — but a forecasting error can trigger an order 20x larger than your usual need.

The risk

Forecasting errors compound. An unusual signal + an overly-reactive agent = weeks of stock ordered in a panic, with storage costs exploding.

What changes

Restocking lead times fell 22% for routine orders. Two large over-orders were blocked before they shipped. Internal audit signed off on the agent's autonomy.

Rules configured
  • Needs approvalOrder over €2,000 → purchasing manager approves
  • BlockedMore than €8,000 in orders per day total
  • Needs approvalAny new unlisted supplier → verification first
  • Blocked'Capital expenditure' category → always handled manually
06
Marketplace & Payouts

An agent that pays dozens of sellers every week

The situation

You run a platform and regularly pay sellers, freelancers, or affiliates. Your agent handles weekly transfers. The risk: fake accounts, fraud rings, or errors in the amounts to be paid.

The risk

Fraudulent sellers can create fake accounts or inflate their figures to receive more. Without verification, your agent can send thousands of euros to people who don't deserve it.

What changes

A group of 14 fake seller accounts was spotted within two weeks through the history and alerts. €82,000 saved before the verification team had time to update their own detection rules.

Rules configured
  • Needs approvalSeller without full identity verification → manual review every time
  • Needs approvalExceptional payout over €500 (99th percentile) → check required
  • BlockedCap of €20,000 in total daily payouts for this agent
  • BlockedOnly this agent can make seller payments — no other agents
Don't see your case?

Every rule is customisable. If your AI can do it, you can govern it.