AI is now embedded in media planning and campaign optimization, but when an agency talks about its use of AI, it can mean anything from automated reporting to algorithmic decision-making that changes budgets and targeting while campaigns are live.
For marketers, the goal isn’t to ban automation – far from it. AI, when used well, can significantly improve outcomes – including savings and value. The priority is to make sure it is being used responsibly, transparently, and in a way that improves outcomes (not just efficiency), with clear guardrails, human oversight and an understanding of what’s happening inside the ‘black box’.
Below are five questions that advertisers can bring to their media agency, and what to look for in the response.
Ideal response: Your agency maps their workflow and shows where AI recommends, where it executes, and where humans must approve.
Why it matters: AI can speed up optimization, but if it’s effectively ‘driving’ without a human in the loop, you can end up with changes that are hard to explain after the fact, or that optimize to the wrong goal. AI should complement rather than replace experienced media professionals - over-reliance on automation without human oversight introduces both performance and governance risks.
What good looks like:
✔️A clear explanation of who is making which decisions - AI vs human - and where responsibility sits
✔️ Named owners for decisions (not 'the platform'
✔️A defined escalation path when performance shifts or the model behaves unexpectedly
Red flags:
❗ 'It's all automated' or 'the algorithm handles it', with no governance
❗Difficulty explaining why a major change has been implemented
Ideal response: A clear, accessible list of data inputs, what's first-party vs third-party, and how privacy and compliance are managed by market.
Why it matters: AI models are only as good as the data feeding them, and the compliance risk sits with both the advertiser and the agency. That makes privacy, data governance and market-by-market compliance a core part of any AI-driven approach, not an afterthought.
What good looks like:
✔️A documented view of inputs: platform signals, conversion signals, attention/viewability/verification signals, brand safety signals etc.
✔️Clear answers on consent, retention and where data is processed and stored - this is particularly important for advertisers who operate across markets or regions.
✔️A clear explanation of how the system behaves when data signals weaken, for example, if fewer users consent to tracking, and how it maintains control and reliabilty in those scenarios.
Red flags:
❗One-size-fits-all privacy settings across markets, or vague assurances without documentation
❗Ambiguity about whether data is used to benefit other clients.
Ideal response: A KPI framework that separates business outcomes from proxy metrics and shows how trade-offs are managed. The framework should be defined jointly by the advertiser and agency, with campaign targets clearly aligned to the advertiser’s overall business objectives.
Why it matters: AI is very good at optimizing to the metric it is given, but that can backfire if the metric is a poor proxy for business value. KPIs should be defined clearly, separating brand outcomes from performance outcomes – and advertisers should be wary of metrics that aren’t clearly defined or independently measurable.
What good looks like:
✔️A one-page KPI hierarchy with primary outcome, supporting metrics and guardrails (e.g. reach/frequency, brand safety, cost controls).
✔️An explanation of optimization priorities; for example, 'we will accept higher CPM if it improves qualified reach' - but with evidence and limits
✔️A plan for cross-channel comparability where possible, so performance isn't defined differently in every platform.
Red flags:
❗KPIs that change by platform 'because that's what the platform gives us'
❗Engagement metrics with no clear definition, benchmarking or link to outcomes, or outcome frameworks that lack clear alignment to advertiser objectives.
Ideal response: Pacing and frequency guardrails, a clear optimization cadence, and reporting that explains why significant shifts have occurred, alongside transparency into the underlying data and clear ownership of that data by the advertiser.
Why it matters: AI-driven optimization introduces constant, incremental changes to campaigns. Without clear guardrails, oversight and accountability, it becomes difficult to maintain control or understand what is driving performance.
What good looks like:
✔️Agreed guardrails with pacing, frequency and brand safety thresholds and exclusions.
✔️A regular optimization rhythm (e.g. daily/weekly), plus thresholds for significant changes.
✔️Reporting that clearly explains what changed, why it changed, and what we expect next.
Red flags:
❗No guardrails, cadence or explanation beyond 'the model decided'.
❗Set-and-forget language for channels that clearly require monitoring.
❗An inability to explain clearly a significant change in campaign performance or optimization decisions.
Ideal response: A full list of platforms and technology used by the agency, what is agency-owned vs what’s third-party, and the level of transparency provided into decision logic and fees.
Why it matters: Agencies increasingly use proprietary platforms and AI-enabled tools, which can improve efficiency but may also reduce transparency when decision-making is driven within systems the advertiser cannot fully interrogate. Limited visibility into how decisions are made, particularly where fees or incentives are linked to specific tools, is a key consideration.
What good looks like:
✔️A transparent stack list: DSPs, buying tools, optimization layers, measurement and verification partners, and what is proprietary.
✔️Clear disclosure of fees linked to tech, data management and any bundles services.
✔️A straightforward explanation of how the agency ensures recommendations are aligned to the advertiser's outcomes, with transparency around any commercial arrangements.
Red flags:
❗A refusal to disclose decision logic, or an inability to separate strategic advice from commercial incentives.
❗'You don't need to worry about the tools - just trust us'
The most important step for advertisers is to ask their agency partners to show their AI workflow in its entirety, and to agree, in writing, on the level of transparency they will receive, including guardrails, reporting and access to data. This allows AI to be used as a performance advantage while maintaining appropriate governance and control.
The role of the advertiser is to maximize the value of the systems used to support their brand. This requires a clear understanding of how those systems operate, alongside appropriate levels of control and auditability across the tools and platforms used by the agency.