Every week, AI systems make thousands of decisions for advertisers—shifting budgets, rotating creative, selecting audiences, and setting bids. Yet verification still runs on a human calendar.
As frontier AI capabilities become universally available, the models themselves no longer differentiate advertisers. Meta’s investments in AI systems and AI-enabled solutions such as Advantage+ benefit every advertiser equally. When the model becomes universal, the edge is no longer the model.
For agencies, durable advantage moves to three capabilities: proprietary data and business context structured for AI; deeper integration with platforms around hard business outcomes; and the speed at which an organization closes the loop from measurement to action.
Compounding works both ways
At machine velocity, compounding is symmetric. Validated learning compounds into advantage. Unexamined error compounds just as quickly—into trajectories that look like performance while eroding it.
When an agentic orchestration layer makes thousands of decisions in real time and if it optimizes towards an outcome not tied to the bottom line, risk grows exponentially.
Is your data ready?
Only 7% of enterprises say their data is completely ready for AI, while a further 27% describe it as not very or not at all ready.
Four foundations matter:
- Establishing a semantic layer ensuring stable business definitions such as “cost per conversion,” “ROAS,” or “active customer” resolve consistently across queries
- Making sure unstructured data such as briefs or transcripts (which brings unique context but can produce wrong artifacts when snippets lose context) can be reliably leveraged by AI
- Solving governance and lineage, because AI generates data at unprecedented volume
- Rebuilding how you manage access and governance as AI democratizes access to analysis and insights across the organization
Incrementality keeps the system honest
There is no single source of truth. Attribution is timely but can overvalue the last click. Marketing mix models capture the whole funnel but move slowly. Experiments establish causality but cannot run everywhere.
The solution is to triangulate, calibrating the rest against the causal ground truth experiments provide. AI raises the stakes: as signals erode and decision velocity rises, grounding the suite of truth in incremental business outcomes matters more, not less.
According to eMarketer, 64% of worldwide AI users have used an AI tool to compare products and 53% to discover a new brand. As consumers begin journeys inside AI assistants, click- and view-based signals degrade further.
On Meta, the response is to keep running incrementality experiments. Meta has opened its ads Model Context Protocol server to any developer with a Meta app, allowing AI applications to connect directly to an ad account, create and manage A/B tests and conversion lift studies, and read results.
Compound assurance
AI systems are non-deterministic: The same prompt can produce different responses and reasoning paths. Traditional quality assurance therefore needs evaluation—structured tests that grade outputs against defined criteria and run continuously as context shifts.
When you put incremental experiments and evaluations together, you achieve compound assurance.
Outcome assurance uses causal measurement to verify that learning fed back into the loop is true. System assurance uses evaluations and operational monitoring to verify that the machinery compounding that learning is behaving as designed.
No AI-driven marketing decision should scale without both assurance layers.
What it looks like in practice
Based on Monks’ global experiments conducted from June 2025 to June 2026 on the Meta platform, the company’s portfolio analysis puts experimentation adoption at roughly twice industry benchmarks: 77% of total investment on Meta globally was measured, 2.1X the benchmark vs. global holdcos. More than 63% of Monks advertisers ran experiments globally, 1.7X the benchmark.
In a Monks analysis of more than 3,000 conversion-optimized Meta ad sets in the U.S. from June 2025 through May 2026, accounts that completed at least five brand lift, conversion lift, or A/B tests had a median cost per conversion 49% lower than accounts that ran fewer experiments. This is an observed association, not a controlled causal estimate; accounts that test more may differ in other ways.
Monks’ Experimentation Engine standardizes experiment design across lift studies, A/B tests, multi-armed bandits, and geo-holdouts, applies consistent statistics, and retains learnings in structured, queryable form—so each new experiment starts from accumulated evidence rather than zero.
Monks.Flow, Monks’ agentic AI orchestration platform, provides the final pillar: acting at AI speed across insights and strategy, creation and adaptation, and delivery and performance, closing the loop between governed data, retained learning and action.
The agency opportunity
If platform AI benefits everyone equally, advantage increasingly lies in orchestration—and who governs it.
“AI enables in-housing” is credible: conversational campaign management can bring more execution inside brands. But validation, evaluation engineering, cross-platform governance, and data-foundation design become more important—requiring platform-neutral expertise and cross-client pattern recognition.
“Platform AI will absorb the agency’s job” is also partly true. The automatable parts will be automated. Agency value migrates to deciding which outcomes are worth optimizing, verifying automation, and assembling the stack across platforms.
The next advantage
The differentiator is the ability to give AI trustworthy context, establish causal truth, retain what has been learned, and act on it quickly—while checking both outcomes and the systems producing them.
For agencies, the opportunity is to build the system that determines which decisions can be trusted.
The question is no longer whether AI will make marketing decisions. It is whether your organization knows which decisions deserve to compound.
—
Shekhar Deshpande leads strategy, insights, and thought leadership for Meta’s global clients and agencies team. He works with Meta’s top global clients and agencies, identifying and unlocking opportunities for growth.