Every CMO budget review now includes a line for artificial intelligence, and almost every CMO is being asked to defend it. The question used to be “what’s the ROI of marketing.” Increasingly, it’s “what’s the ROI of the AI inside marketing.” That’s a harder question to answer than it sounds, and the reason isn’t a lack of AI marketing tools. It’s that most of what those tools measure was never the part of the business that moved the needle most.
The contradiction every CMO is living with
Adoption of artificial intelligence in marketing is close to universal. McKinsey’s 2025 State of AI survey puts organizational AI use at 88 percent, up from 78 percent the year before. What hasn’t followed is impact: only around 6 percent of organizations qualify as what McKinsey calls “high performers,” meaning AI contributes 5 percent or more to EBIT. Gartner’s 2026 CMO Spend Survey shows a similar picture from the marketing side. Seventy percent of CMOs want their function to lead on AI, and roughly the same share admit their processes aren’t ready for it. Marketing budgets, meanwhile, have stayed flat at under 8 percent of revenue. Everyone is using the tools. Almost nobody can point to what changed.
What marketing analytics actually measures, and the half it misses
Part of the answer sits in a well-established but underappreciated body of research on where advertising effectiveness actually comes from. NCSolutions and Nielsen’s meta-analysis of nearly 450 CPG campaigns, built on a machine learning model of eighteen features across creative, targeting, reach, and recency, found that creative quality drives close to half, 49 percent, of incremental sales lift, more than targeting, reach, and recency combined. Creative is, by a wide margin, the single largest lever in the entire marketing mix.
Here’s the part that explains the gap: a February 2024 Advertiser Perceptions study of marketers and agencies found they estimate creative’s contribution at roughly 19 to 20 percent, while ranking targeting as the more important driver. Marketers have the two numbers close to reversed. This isn’t a case of teams being careless or unscientific. Media could be tracked click by click for two decades, so measurement budgets, dashboards, and analytics headcount grew around media. Creative resisted that kind of instrumentation, so investment in measuring it stayed thin, sentiment analysis on social posts and campaign-level brand tracking aside. The result is a marketing analytics stack that’s excellent at the smaller lever and largely blind to the bigger one.
Why generative AI widened the gap before it could close it
The obvious fix would be for AI to finally bring measurement to creative. Instead, the first wave of generative AI in marketing went almost entirely toward making more of it. Gartner’s 2025 CMO Spend Survey found that 77 percent of marketers using generative AI apply it to creative development, by far the most common use case, ahead of anything analytics-related. Large language models, LLMs like ChatGPT, Google Gemini, and Anthropic’s Claude, now draft ad copy and product pages in seconds. Purpose-built ai writing assistants such as Jasper AI layer brand voice and templates on top of the same natural language processing engines so a hundred writers still sound like one brand. AI image generators, from Midjourney to Canva AI, do the same for visuals, and tools like Synthesia extend it to video. None of this requires much prompt engineering skill anymore to produce a usable first draft.
The reported return on all this is real, but it’s a productivity story, not an effectiveness one. Gartner’s marketers cite time savings (49 percent), cost savings (40 percent), and higher content output (27 percent) as the benefits of generative AI, three efficiency metrics and not one effectiveness metric among them. MIT’s Project NANDA found that 95 percent of generative AI pilots show no measurable P&L impact, even though more than half of enterprise generative AI budgets now sit inside sales and marketing. More assets are getting produced against the same evidence base as before. The ratio of output to insight has gotten worse, not better, which is the opposite of what these productivity tools were supposed to buy.
What AI in marketing analytics should actually deliver
Closing that gap means moving analytics from descriptive, telling you what happened, to decision-grade, telling you what to do next. That shift happens across four levels: benchmark (where does this asset stand), diagnosis (why does it work or not), recommendation (what should you select, improve, or allocate), and learning (how does each result compound into the next decision). Most marketing analytics today covers the first level reasonably well and thins out fast after that. A score without a why is a verdict, not analysis, which is exactly the gap the “They want the why, not just a score” mindset among evidence-driven marketing teams is pointing at.
Predictive analytics and ai-powered analytics are the terms usually used for this shift, but the mechanism that actually delivers it is workflow redesign, not a new dashboard. McKinsey’s data on AI high performers backs this up directly: fundamental workflow redesign is the single strongest correlate with EBIT impact, and it’s a step only a minority of companies have taken. Bolting an ai workflow onto an existing process rarely produces the same result as rebuilding the process around where the evidence actually needs to sit.
Why an AI content platform can’t solve a measurement problem
It’s worth being precise about what an AI content marketing platform is built to do, because it’s easy to assume it already covers this. These platforms are exceptionally good at the generation layer: producing copy variants, localizing them, scaling output across channels. What they aren’t built to tell you is which variant will actually work, or why. Generation and evaluation are two different layers of the marketing stack, and for most teams, only the first one has real tooling behind it.
This is the layer BCG’s 2026 CMO survey describes as a missing “brand intelligence layer” inside the emerging agentic marketing stack, the connective layer that ai agents and automation tools still route around rather than through. It shows up across the rest of the stack too. Salesforce Einstein scores leads and personalizes CRM outreach. HubSpot and marketing automation platforms manage the distribution. Albert.ai automates media allocation across channels. Semrush, Surfer SEO, and Perplexity handle research and seo optimization before a brief is even written. Hootsuite and other social media management tools schedule and track the social media output that comes out the other end, often with hashtags and sentiment analysis layered on for short-term read. Every one of these is a legitimate, well-built piece of ai marketing software. None of them is built to tell you whether a specific asset will move a customer, which is a different job entirely, and one that generative AI made more urgent by multiplying how many assets need that answer.
A short checklist for evaluating AI marketing analytics tools
For teams building out an ai marketing strategy and shopping the category, five criteria separate genuine analytics from a dashboard wrapped around a score:
- Asset- and channel-specific logic. A shelf pack and a TikTok clip are processed differently by attention and memory. A single model with the same KPIs across every channel is convenient, not accurate.
- Grounded in established science, not a black box. Teams that have already been burned by tools that promised transformation and delivered disappointment now rank methodological transparency second only to ROI as a buying criterion.
- Validated against in-market performance, not only survey results. The Marketing Accountability Standards Board’s validation framework is the reference point here, not self-reported scores.
- Embedded in the workflow, not sitting beside it. This is the same workflow-redesign principle McKinsey found correlates most strongly with EBIT impact, applied at the level of a single creative decision instead of the whole organization.
- Compounding. Every asset tested should make the next decision sharper. If a test doesn’t feed forward into the next brief, it’s reporting, not learning.
This matters as much for lead generation and customer engagement programs and influencer marketing partnerships as it does for a paid TV campaign; the same 49-versus-19-percent perception gap applies wherever a human ultimately looks at, reads, or scrolls past a piece of creative.
What this means for three roles
For a CMO, the case to the CFO shifts from “we adopted AI” to an effectiveness capability with evidence attached at the asset level, which is the level the budget conversation is actually about. For research and insights teams, it means scaling measurement without losing validity, moving out of the role of bottleneck and into the role of standard-setter. For brand management, it means a decision made with evidence in hand instead of one made by seniority or gut feel, which is the same shift performance marketers have already made on the media side over the last decade.
Know What Works. Understand Why. Increase Impact.
With budgets flat, asset quality is the one lever that doesn’t need a bigger line item to move, only better evidence about which assets are worth shipping. That’s the layer generative and evaluative AI working together is built to close, and it’s also the argument behind why content creation needs smarter evaluation as production scales.
If you want to see where your own assets stand against it, start a free trial of Brainsuite and test your first asset today.
