Three-quarters of marketers say their measurement systems are broken, and they plan to rebuild them with AI. That figure — 75% — anchors eMarketer's latest analysis of marketing attribution and analytics, published this week.

What does "broken" actually mean here?

The headline number comes from a survey of marketing practitioners who described their measurement stacks as unreliable across attribution, ROI tracking, and campaign reporting. eMarketer framed the finding as a verdict on legacy martech, not a passing frustration. Marketers cannot tie spend to outcomes with confidence. Last-click attribution still dominates in many organizations, even though buyer paths now run through TikTok, Reddit, podcast ads, retail media networks, and AI-driven search.

Cookie deprecation has stripped another layer of signal. Internal teams disagree on definitions. CFOs demand proof; growth teams demand speed. The 75% figure summarizes that reality.

Why AI is the rebuild strategy

Per eMarketer's framing, machine-learning models are the practical answer to a measurement problem rules-based dashboards cannot solve. AI tools can stitch cross-channel data, model incrementality, and forecast outcomes where deterministic tracking fails. The pitch to a CFO is straightforward: stop arguing about which screen gets credit, and start asking which intervention lifted the baseline.

Who is actually using this?

Early adopters sit in performance marketing, retail media, and direct-to-consumer brands where every dollar is audited. Slower adopters — B2B, healthcare, financial services, regulated industries — cite compliance, data residency, and audit trails as blockers rather than capability gaps. In-house teams remain split: some centralize measurement under a single ops leader; others push it out to channel owners and accept the noise.

Should buyers be skeptical?

Yes. AI measurement still hallucinates when training data is thin. Models trained on one brand's data rarely transfer cleanly to another. Vendors rarely disclose confidence intervals on their lift claims. Marketers evaluating any AI measurement vendor should ask three things before signing:

• What training data was used?
• What is the holdout methodology?
• What is the false-positive rate on attributed conversions?

If the vendor cannot answer in writing, the deal is not ready.

What should marketers do now?

Treat the 75% figure as a procurement signal, not a survey artifact. Audit the current stack: what data sources feed it, what models score conversions, what dashboards reach the CFO. Identify the gap between marketing's view of performance and finance's view of incrementality — most organizations will find a meaningful delta on the same campaigns. Close that gap with AI tools that produce auditable lift estimates, not black-box attribution scores.

The downstream effects are concrete:

• Budget conversation moves from "what did we spend" to "what did we lift"
• Vendor RFPs increasingly ask for AI-native measurement, not bolt-on dashboards
• Headcount shifts toward data engineers and MLOps, away from analyst roles tied to legacy platforms

Bottom line

Marketers cannot keep reporting on attribution that CFOs no longer trust. AI is the rebuild strategy because the alternatives — better spreadsheets, more pixels, fewer cookies — have already failed. The 75% figure is not a complaint. It is a procurement signal.