- Is it really worth relying on AI agents for your company's paid campaigns?
- What is really happening at the big international agencies?
- Who decides what: how is the work split between AI and humans in campaigns?
- What changes in practice for an SME or Italian company working with a media agency?
- What governance tools and approaches exist for AI agents in campaigns?
- What legal rules do you need to know before using AI agents in campaigns?
- Why everyone is talking about it now: is the market really ready or is it just hype?
- How to get started: the steps to evaluate and govern AI agents in your campaigns
- What are the most common mistakes of those who rely on AI agents without the right rules?
- How can you tell if the governance of AI agents in your campaign is working?
- So, is it worth relying on AI agents in campaigns for your company?
- How we see it at SHM
- Frequently Asked Questions
- The leading international agencies are already using AI agents to plan, buy and optimize paid campaigns: the change is happening right now.
- The main risk isn't the AI "going rogue," but the lack of approval thresholds, human checks and clear rules on who decides what.
- 78% of European organizations have already appointed a formal AI governance lead: if your agency hasn't, that's a question to ask right away.
- Asking your agency how it monitors and governs AI agents in campaigns is the first concrete step to protect your budget and stay in control.
Two-thirds of media buyers are already testing AI agents for campaigns, but only 30% have scaled automation across the entire cycle. The gap between those who start and those who truly govern is still huge — and the budget at risk is yours.
Is it really worth relying on AI agents for your company's paid campaigns?
It depends on how well your agency has structured its controls: agentic AI brings speed and scale, but without clear governance the risk of runaway spending is real.
An AI agent — a software system that makes decisions on its own, without a human approving every single action — is becoming the standard tool for buying online ad space. It's not science fiction: according to Digiday, in a single week of October 2026 three major international groups announced agentic systems running on their clients' campaigns.
For an Italian marketing manager or business owner, the right question isn't "is AI ready?" but "does my agency have the rules to use it safely?"
| Your situation | Should you rely on AI agents? | why |
|---|---|---|
| Agency with documented approval processes and a dedicated AI lead | Yes | You can benefit from speed and optimization with limited risk |
| Agency that uses AI but has no spending thresholds or automatic alerts | Evaluate first | You need a written agreement on who approves high-impact decisions |
| Agency that can't answer questions about AI governance | No, for now | The risk of unauthorized spending or undetected errors is too high |
| Campaign budget under €5,000/month | Wait | The governance overhead often outweighs the benefit at these volumes |
What is really happening at the big international agencies?
In just a few weeks, three of the world's leading media groups have launched operational agentic systems: these aren't future announcements, they're tools already used on live campaigns.
Dentsu announced that its agency Carat is working with NBC Universal, Freewheel and the AI company Newton to launch an agent dedicated to planning, buying and optimizing ad space. A luxury retail client is already using it.
The independent agency Tinuiti has opened its proprietary measurement system directly to AI agents, bringing the management of audiences, creative, media and measurement under a single server. Stagwell has relaunched its media agency with an agentic system called Machine OS, designed to react in real time to market changes.
These aren't lab experiments. They're production systems, running on real budgets of real clients. If you work with a media agency — or are thinking about it — this changes the questions you need to ask.
| Group / Agency | Agentic system | What it does on its own |
|---|---|---|
| Dentsu / Carat | Agent with Newton + Freewheel | Upfront planning, buying and optimization |
| Tinuiti | Bliss Point MCP Server | Unified management of audiences, creative, media and measurement |
| Stagwell / Stagwell Media | Machine OS | Real-time reaction to market changes |
| WPP Media | In-house system with structured governance | Inventory analysis, AI recommendations with human approval |
Who decides what: how is the work split between AI and humans in campaigns?
The model the most advanced agencies are adopting draws a clear line between decisions that can be automated and those that require human approval, especially where there's a financial impact.
The keyword that comes up at every agency Digiday interviewed is "human in the loop" — a human who stays in the decision loop, especially for high-impact choices. It's not an abstract ethical stance: it's a practical necessity, because AI agents make mistakes and, without controls, those mistakes turn into unauthorized spending.
- 1 · AI agentAnalysis of available inventoryThe agent scans the ad space available across multiple platforms and evaluates it against the campaign's goals.
- 2 · AI agentPurchase recommendationProposes a buying plan optimized for audience, format and timing. It doesn't execute yet.
- 3 · Human ownerApproval or changesA senior media buyer reviews the proposal, checks that it fits the strategy and approves within predefined thresholds.
- 4 · AI agentReal-time execution and optimizationOnce approved, the agent executes and optimizes on its own, but within the approved parameters.
- 5 · Human ownerMonitoring and alertsThe team gets automatic alerts if the agent exceeds spending thresholds or performance falls out of range.
What does this change in practice for an Italian SME or company working with a media agency?
The main risk isn't AI "going rogue." It's not knowing which decisions it's making on its own, and having no written agreement on who is responsible for mistakes.
Picture an Italian manufacturing company investing €20,000 a month in Google and Meta campaigns. Today, an account manager approves every significant change to that spend. With an AI agent, the same budget can be optimized in real time. But if the agent decides to move €8,000 to the wrong channel and nobody notices, the damage is already done.
The question to ask your agency isn't "Do you use AI?" but "Which decisions can AI make on its own with my budget, and which need my approval?"
| Type of Italian company | Main risk | Question to ask the agency |
|---|---|---|
| E-commerce with seasonal campaigns | Spend accelerating on the wrong products during peaks | Which daily spend thresholds trigger a human alert? |
| Professional firm (B2B) | Low-quality leads optimized by AI for volume, not value | How does the agent tell a qualified lead from an unqualified one? |
| Manufacturing company with exports | Wrong geographic targeting on non-priority markets | Who approves targeting changes above X%? |
| Chain of physical stores | Budget shifted to digital channels that don't fit the local area | Can the agent change the geographic distribution of the budget? |
What governance tools and approaches exist for AI agents in campaigns?
The best-structured agencies use three levels of control: automatic approval thresholds, separation between recommendation and execution, and periodic audits of the agent's behavior.
There's no universal standard yet, but the model emerging from international agencies' statements, and one you can ask your own agency for, works on three levels. We also cover this in our article on how to stay in control when Google and Meta automate campaigns, which looks at the problem from the platform side.
| Level of control | What it involves | Who manages it |
|---|---|---|
| Approval thresholds | The agent can't spend more than €X/day or change targeting by more than Y% without approval | Account manager + client |
| Separation of recommendation and execution | AI proposes, a human approves before execution on high-impact decisions | Senior media buyer |
| Periodic behavior audit | Weekly or monthly review of the agent's decisions and results | In-house team + client |
| Automatic alerts | Real-time notifications if the agent exceeds predefined parameters | Automated system → human owner |
| Separation of probabilistic AI and deterministic execution | AI recommendations are labeled as such, distinct from certain executions | Architecture of the agentic system |
What legal rules do you need to know before using AI agents in campaigns?
Since August 2, 2026, the EU AI Act imposes transparency obligations on AI-generated content: those who fail to comply risk fines of up to €15 million.
Since August 2, 2026, the obligations of Article 50 of the EU Regulation on artificial intelligence (AI Act) are fully in force. In practice: if an AI agent generates text, images or video used in campaigns, that content must be recognizable as artificial. Violations are punished with fines of up to €15 million or 3% of worldwide annual turnover. The details are clearly explained by Studio Cravero in its guide to Article 50 of the AI Act.
For an Italian company that entrusts its campaigns to an agency, this means one concrete thing: you need to know whether the creative content produced by the AI agent is labeled correctly. If it isn't, responsibility may fall on the advertiser too, not just the agency.
Why everyone is talking about it now: is the market really ready or is it just hype?
The numbers show adoption is real but uneven: large organizations are far ahead of SMEs, and governance is still the industry's weak point.
The IAB Europe report "Impact of AI on Digital Advertising 2026" (September 2026, 50 European respondents) shows that 86% of organizations already use AI for marketing and 72% have increased investment compared to 2025. But the current performance of AI agents in advertising operations earns an average rating of 2.76 out of 5: we're still in the maturing phase, not full trust.
The gap between large and small organizations is clear: among companies with more than 500 employees, 86% have agentic systems at autonomy level 2 or higher. Among those with fewer than 500 employees, the share drops to 48%, according to IAB Europe. For Italian SMEs, this means the risk of relying on tools that aren't yet mature is higher, not lower.
Our article on agentic buying and automated media purchasing explores the market context and the adoption dynamics underway.
How to get started: the steps to evaluate and govern AI agents in your campaigns
Before activating any agentic system, you need an audit of your agency's governance and a written agreement on who decides what.
The starting point isn't choosing an AI tool. It's understanding how your agency already handles automated decisions. Many agencies have used automation for years (automated bidding on Google Ads, automatic optimization on Meta) without ever formalizing it in a governance document. New-generation AI agents amplify this problem, because they act on more variables at once and faster.
To learn how AI agents can scale campaigns without growing the team, also read AI agents for campaigns: scaling without hiring.
- 1 · You (the client)Ask your agency for an AI governance documentIt must specify which decisions the agent makes on its own, which require approval and who the responsible human owner is.
- 2 · You + agencySet spending thresholds and alertsAgree in writing on the daily and weekly thresholds beyond which the agent can't act without explicit approval.
- 3 · AgencyCheck AI Act complianceAll AI-generated creative content must be labeled as such. Ask for written confirmation that the agency complies with Art. 50 of EU Regulation 2024/1689.
- 4 · You + agencySet up a monthly report on the agent's behaviorA performance report isn't enough: you want to know how many decisions the agent made, on which variables and with what results.
- 5 · You (the client)Review the agreement every quarterThe AI agent market moves fast. Spending thresholds and governance rules should be updated at least every three months.
What are the most common mistakes of those who rely on AI agents without the right rules?
The most common mistake is handing high-stakes financial decisions to AI without defining who is accountable if something goes wrong.
The original source puts it plainly: "not every decision should be delegated to an agent." The problem is that many companies, and some agencies, haven't yet drawn the line between what AI can do on its own and what requires human oversight. Here are the most common mistakes we see in the market.
A related topic is unreviewed AI content: risks, rules and how to protect yourself is a useful read for understanding where legal responsibility sits.
| Error | Concrete consequence | How to avoid it |
|---|---|---|
| No automatic spending threshold | The agent can burn through the monthly budget in a few days on the wrong audience | Set daily and weekly caps in writing |
| Confusing an AI recommendation with a certain decision | Actions are taken based on probabilities, not hard data | Ask your agency to separate probabilistic outputs from deterministic executions |
| No named human point of contact | When something goes wrong, nobody knows who is responsible | Put the name of the responsible senior buyer in the contract |
| Auditing only results, not the agent's behavior | You can see the campaign performed badly, but you can't tell why | Request reports on the agent's decision-making process, not just KPIs |
| Unlabeled AI content | Risk of AI Act penalties of up to 15 million euros | Check compliance with Art. 50 of EU Regulation 2024/1689 |
How can you tell if the governance of AI agents in your campaign is working?
Governance works when you can answer three questions: what decisions did the agent make this week, who approved them, and what results did they produce?
Monitoring the performance of a campaign run by AI agents calls for different indicators than a traditional campaign. Looking at cost per click or return on ad spend isn't enough: you also need to know how the agent got those results. Without that transparency, you can't tell good performance from a lucky streak.
For those running campaigns across multiple platforms, cross-platform measurement has become a priority for 72% of media buyers in 2026 (up from 64% in 2025), according to the IAB 2026 Outlook Study: a sign that the market is realizing agentic AI requires new measurement tools.
| Indicator | What it measures | Recommended frequency |
|---|---|---|
| Number of autonomous agent decisions | How many actions it carried out without human approval | Weekly |
| Human intervention rate | % of decisions blocked or modified by the team before execution | Weekly |
| Deviation from approved budget | Difference between planned spend and the agent's actual spend | Daily |
| Lead quality / conversions | Is the agent optimizing for volume or for value? | Monthly |
| AI Act compliance | Is AI-generated content labeled correctly? | Every campaign |
| Human contact's satisfaction | Does the senior buyer find the agent's recommendations reliable? | Monthly |
So, is it worth relying on AI agents in campaigns for your company?
Worth it if your agency has an AI governance document, a named human point of contact, defined spending thresholds, and complies with the AI Act. In that case, the speed and scale of AI agents are a real advantage.
Somewhere in between: your agency already uses automation (automated bidding, Meta optimization) but hasn't yet formalized rules for next-generation agents. It's worth asking for a dedicated meeting before expanding automation.
Not yet if your agency can't answer questions about who approves the agent's decisions, or has never heard of the AI Act. In that case, the risk outweighs the benefit: better to wait or change partners.
How we see it at SHM
For us, AI agent governance isn't a technical issue: it's about the relationship. An AI agent managing your ad budget is an agent in the legal sense, and like any agent, it needs clear instructions, defined limits and someone accountable for its actions. We believe simplifying digital also means making what AI does with your budget understandable, not hiding it behind a performance dashboard.
We work with Italian companies that want to bring AI into paid campaigns without losing control. Our approach always starts with an audit of your existing governance, then we define thresholds, approval processes and monitoring indicators together. If you're thinking about how to structure this path, the starting point is our strategic AI integration service. If you want to understand how AI is changing the organic visibility of campaigns, SEO for AI is the natural complement. And if you run Google Ads campaigns and want to know where automation helps and where it risks getting out of hand, our Google Ads campaigns team can give you a concrete assessment.
You'll find all our articles on paid media, automation and AI in our Marketing and performance ADV hub.
Consider a manufacturing company that asked us to review its relationship with its media agency after noticing unexplained gaps between the approved budget and actual spend. It wasn't fraud: no one had ever defined which decisions the automation could make on its own. We helped put together a simple governance document, just two pages, that solved the problem and built more trust in the relationship with the agency. They didn't need a more sophisticated AI system: they needed clarity on who decides what.
Frequently Asked Questions
An AI agent in media buying is a software system that makes decisions on its own: it analyzes available ad inventory, proposes or executes purchases, and optimizes campaigns without a human approving every single action. The most advanced agencies already use them on live campaigns, with control systems that limit the agent's autonomy on high-stakes financial decisions.
The simplest route is to ask directly: ask your agency whether it uses advanced automation systems or AI agents on your campaigns, and which decisions those systems can make on their own. If the agency can't answer or doesn't have a governance document, that's a red flag.
The main risk is unauthorized spending: an agent without control thresholds can shift budget to the wrong channels or audiences before anyone notices. On top of that comes legal risk: if AI-generated creative content isn't labeled as such, the company may face penalties under the European AI Act.
Yes. Starting August 2, 2026, Article 50 of EU Regulation 2024/1689 requires AI-generated content, including text, images and video, to be identifiable as artificial. This also applies to ad creatives produced by AI agents. Violations are punishable by fines of up to 15 million euros or 3% of worldwide turnover.
According to the IAB Europe report from September 2026, 58% of European professionals expect media buying through AI agents to reach regular operational use within a year. However, the current performance of AI agents in advertising operations earns an average rating of 2.76 out of 5: we're still in a maturing phase.
There are four essential questions: what decisions does the agent make on its own with my budget? Who is the responsible human contact? Which spending thresholds trigger an alert or require my approval? Is AI-generated content labeled in compliance with the AI Act? If the agency doesn't have clear answers to all four, you need to formalize the rules before moving forward.
The real question isn't whether to trust AI, but whether you've put in writing who is accountable when AI gets it wrong.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
