New AI Is Already Running Ads. Who Controls Its Decisions and Spending?

Until recently, AI in media buying mainly helped with routine work: writing several ad copy variations, generating creative ideas, preparing reports, or quickly analyzing performance data.

Today, its role is gradually changing. AI tools are becoming part of everyday workflows, while agentic systems can do more than provide recommendations: they can perform certain actions independently.

Analyzing a campaign, identifying an issue, suggesting budget reallocation, launching a task, or changing settings can already be automated when the necessary integrations are in place.

But this raises a new question: if AI can make decisions and take actions, who controls its access, budget, and results?

For media buyers, this is no longer only a question of automation efficiency. It is also a question of financial security and team management.

AI Agents in Media Buying: What Can Already Be Delegated

Not every task needs to be completed manually. That is why AI is gradually taking over more repetitive operations.

At the same time, it is important to distinguish between built-in automation on advertising platforms and external AI agents. The former optimize campaigns within predefined settings. The latter, when given the necessary integrations and permissions, can analyze data from multiple sources and perform actions through the APIs of connected services.

Today, media buyers can use AI for several core tasks.

Campaign Analysis

AI can quickly process large volumes of data and identify patterns that are difficult to notice manually.

For example, it can analyze:

• CTR and conversion rate;
• customer acquisition cost;
• spending dynamics;
• the performance of individual creatives;
• audience effectiveness;
• changes in metrics by GEO;
• deviations from usual performance levels.

Instead of manually reviewing dozens of campaigns, a media buyer receives a preliminary analysis and makes a decision based on it.

Creative and Ad Copy Generation

AI can prepare variations of headlines, descriptions, CTAs, video scripts, and other advertising assets.

This is especially useful during testing, when a team needs to generate several hypotheses quickly instead of spending hours creating each one from scratch.

However, the final evaluation remains with a human. AI can suggest dozens of options, but not every one will match the offer, audience, or advertising platform policies.

Finding New Hypotheses

AI can be used to identify potential audiences, analyze previous tests, and form new assumptions.

For example, based on the results of previous campaigns, an agent may suggest testing another GEO, changing the message in a creative, or reallocating attention between audience segments.

Here, AI acts more like an additional analyst that helps teams identify testing directions faster.

Reporting

Reporting is another obvious task for automation.

AI can collect data from several connected sources, structure the results, and prepare regular reports for the team.

For a manager, this means less manual work. For a media buyer, it means more time for analysis and launching new hypotheses.

Bid and Budget Optimization

This is where things become particularly interesting.

There is a difference between AI saying:

“This campaign is performing well. Consider increasing its budget.”

And a system being able to change the budget on its own.

In the first case, AI remains an assistant.

In the second, it becomes a participant in the process that requires clearly defined boundaries of authority.

For example, the Google Ads API allows teams to work with advertising campaigns and budgets programmatically. This makes it possible to build custom automation scenarios in which AI analyzes performance, suggests changes, and a connected system executes them after predefined conditions are met.

However, the technical ability to change a budget does not mean that a system should be allowed to do so without restrictions.

This is an important difference between a regular AI tool and an AI agent.

AI Assistant vs. AI Agent: More Than a Naming Difference

Tools can be divided into two broad categories.

An AI assistant analyzes information and suggests a solution. The final action remains with a human.

An AI agent can perform certain actions independently. For example, it can work with connected services, change settings, or launch pre-approved processes.

The second option requires a much more careful approach to access management.

If AI has access only to analytics, it cannot directly change campaign settings or spend budget through an advertising platform.

If the same AI can change advertising budgets, manage campaigns, or initiate actions through connected payment services, the consequences of an error become far more serious.

Therefore, automation should not start with the question, “What else can we delegate to AI?” It should start with the question, “Which actions does it actually need permission to perform?”

When Automation Starts Creating Problems

The main risk of AI in media buying is not that it will necessarily make a mistake.

The issue is different: AI can perform a logical action based on an incorrect or incomplete goal.

Imagine the following situation.

A campaign shows a strong ROAS over the last few hours. AI sees positive dynamics and suggests increasing the budget.

At first glance, the decision seems reasonable.

But a media buyer knows that:

• the data has not yet accumulated enough;
• some conversions may be delayed;
• traffic quality has not yet been verified;
• actual profit is lower due to additional costs;
• the campaign performs well only during a specific time period.

AI analyzes the data available to it. A media buyer also considers context that is not always reflected in reports.

For this reason, automated optimization does not eliminate the need for human control.

Another Problem: Invisible Spending Growth

Suppose a team is running several projects at the same time.

Each project may use:

• ad accounts;
• AI services;
• creative generators;
• analytics platforms;
• trackers;
• anti-detect browsers;
• proxies;
• additional SaaS tools.

Each individual payment may look small.

But if dozens of subscriptions and services are paid for continuously, the total expense becomes significant.

And if all payments go through one card, a question quickly arises: how much did the team actually spend on a specific project?

This problem exists regardless of whether a human or AI manages the advertising. Automation simply increases the number of processes that need to be controlled.

Who Is Responsible if AI Spends Too Much?

Ultimately, responsibility for automation settings and spending remains with the team that implements AI into its workflows.

AI does not know which financial limit is critical for a specific project unless that limit has been defined.

Therefore, before granting AI access to work tools, teams should establish several levels of control.

What AI Can Do Without Constant Approval

For example:

• analyze statistics;
• prepare reports;
• identify anomalies;
• generate creative hypotheses;
• suggest changes;
• sort campaigns according to predefined criteria.

What Should Require Human Approval

For example:

• increasing a budget above a defined limit;
• launching a new campaign;
• changing key settings;
• connecting a new service;
• changing payment parameters;
• granting new access permissions.

What Should Remain Under Separate Control

First of all, financial operations.

If AI receives overly broad permissions, an error in one setting can turn from an unsuccessful test into a real budget overrun.

That is why it is useful to define in advance:

what AI can view → what it can suggest → what it can change independently → what requires approval.

Restrictions should work at the level of advertising tools and automation systems. Controlling payment methods and actual expenses is another important layer of protection.

How to Control Team Spending on AI and Advertising

The more automation appears in daily work, the more important financial infrastructure becomes.

Imagine a team of several media buyers.

Each of them has ad accounts, AI services, and additional tools. At the same time, the team is testing several offers and GEOs.

If everything is paid for with one card, expenses gradually become mixed.

For example, one statement may include Meta Ads, Google Ads, ChatGPT, Midjourney, an analytics service, and several other SaaS tools.

After a month, it is no longer always clear which project cost how much.

It is more convenient to separate payments before they become a problem.

Separate Cards for Different Tasks

With Pay2.House, a team can issue virtual cards for advertising campaigns and work-related services.

For example:

• Card No. 1: advertising budget for project A;
• Card No. 2: advertising budget for project B;
• Card No. 3: team AI services;
• Card No. 4: a separate work tool or subscriptions.

This approach helps prevent expenses from mixing and makes it easier to understand where the budget is going.

When necessary, cards can be used for individual projects, services, or areas of work.

It is important to understand that separating payments does not replace advertising platform limits. It helps organize financial accounting and monitor operations across different tasks.

Team Expense Control

When several people work on projects, another task appears: not only separating expenses, but also managing access.

In Pay2.House, teams can use My Teams for this purpose.

The tool helps create separate teams, invite members, allocate budgets, and monitor financial operations.

A team lead can organize work with team cards and accounts, view expenses, and manage members.

This is especially relevant as the number of employees, projects, and services grows.

You can learn more about the tool in the article “My Teams in Pay2.House: A Team Lead Tool for Controlling Media Buyers and Budgets.”

Auto Top-Up

Another task that can be automated is topping up work cards.

Pay2.House offers an Auto Top-Up feature that allows users to automatically top up a card when its balance reaches a selected threshold.

The user selects the account for debiting, sets the threshold, and defines the top-up amount.

This helps reduce manual operations and lowers the risk of a required payment failing due to insufficient funds on the card.

However, it is important to remember that automatic top-ups are not a tool for limiting an advertising budget. If spending increases, the card may continue to be topped up according to the configured settings, provided there are sufficient funds in the account.

Therefore, Auto Top-Up should be used together with regular expense monitoring and limits in advertising systems.

A detailed guide is available in the article “Card Auto Top-Up in Pay2.House: How to Set Up the Feature.”

Transactions Instead of Manual Expense Search

Another advantage of separate cards is that expenses become easier to track.

Instead of one general statement, the team gets a clearer picture: which service charged money, from which card, and within which workflow the transaction occurred.

For small teams, this is a matter of convenience.

For a growing media buying department, it is already part of financial control.

How It Works in Practice: AI, Media Buyers, and Payment Infrastructure

Imagine a team running several Google Ads campaigns.

An AI agent is connected to analytics data and helps identify optimization opportunities. At the same time, budget changes require approval from a responsible team member.

The workflow may look like this:

Step 1. AI Analyzes Performance

The system notices that one campaign has a higher ROAS than the others and suggests increasing its budget.

Step 2. The Media Buyer Reviews the Recommendation

The specialist evaluates traffic quality, conversion volume, actual expenses, and whether the recommendation aligns with the overall strategy.

Step 3. The Budget Change Is Approved

After approval, the media buyer or an authorized automation scenario changes the budget in the ad account.

Step 4. Expenses Are Tracked Separately

To pay for advertising, the team uses a separate virtual card from Pay2.House linked to the relevant ad account. This helps separate project payments from AI service subscriptions and other work-related expenses.

Step 5. The Team Lead Monitors the Bigger Picture

The manager reviews expenses across cards and projects. With My Teams, they can also monitor the work of team members.

In this model, each tool performs its own function:

• AI analyzes data and suggests optimizations.
• A media buyer assesses risks and makes decisions.
• An advertising platform manages campaigns and advertising budgets.
• Pay2.House helps organize payments and track expenses.

As a result, automation does not replace financial control. It becomes part of a managed workflow.

Practical Checklist: What to Check Before Connecting an AI Agent

Before granting AI access to work tools, review five key questions.

1. What Data Can It See?

Do not give the agent access to information that is not necessary for its task.

2. What Actions Can It Perform?

Access to performance statistics and permission to change an advertising budget are completely different levels of risk.

3. Is There a Financial Limit?

For automated actions, acceptable limits must be defined in advance and configured in the relevant systems.

4. Is Human Approval Needed?

For critical actions, it is better to require confirmation before execution.

5. Can Access Be Revoked Quickly?

If an agent starts acting incorrectly, the team must be able to stop its work and revoke permissions immediately.

The guiding principle is simple: the higher the potential cost of an error, the more human control should remain in the process.

What Media Buying Will Look Like as AI Agents Develop

AI is gradually taking over part of the manual work of media buyers, and this trend is likely to continue.

Routine analysis, report preparation, hypothesis generation, creative variation, and some campaign-related operations will increasingly be automated.

However, this does not necessarily mean the disappearance of the media buyer’s role.

Rather, the role itself is changing.

Instead of constantly managing every campaign manually, specialists may spend more time setting goals, choosing strategies, evaluating traffic quality, supervising AI agents, and managing budgets.

And the more processes a team automates, the more important it becomes to understand which decisions can be delegated to a system and which must remain with a human.

AI can speed up workflows, but without clear rules, automation can increase not only efficiency but also financial risks.

Therefore, the next stage of media buying development will be connected not only with the question, “What else can AI do?”

A more important question will be:

“Which actions are we ready to trust AI with, and how do we build a system where automation stays within budget?”

Here, financial infrastructure becomes just as important a part of the workflow as an ad account, tracker, or analytics platform.

Pay2.House helps teams separate expenses by cards and projects, organize teamwork through My Teams, use Auto Top-Up, and track transactions.

Because expenses can be automated. But a human should still control them.