How can AI help in marketing?

The Adoption-ROI Gap: Why 91% of Marketers Use AI but Only 41% Can Prove It Works

Nine out of ten marketing teams now have AI somewhere in their stack, yet fewer than half can prove it's paying off. That's the central paradox of AI marketing in 2025: adoption is nearly universal, but confidence is moving in the wrong direction.

According to MarTech's analysis of industry data, 91% of marketing teams use AI in some form — but only 41% of marketers can prove AI ROI, down from nearly half the prior year. Meanwhile, 70%+ of B2B organizations plan to increase AI investment anyway. Spending more while proving less is a pattern that eventually catches up with any budget line.

The root cause is what practitioners call the "tactical trap". Most teams deploy AI only for low-risk tasks — social posts, email subject lines, meeting summaries — producing what one analyst described as "random acts of AI." According to the same research, 71% of B2B firms use AI for content production, but 56% see its primary value in basic execution. That's a lot of effort spent on the cheapest part of the workflow.

The measurement problem is just as damaging. Teams count output volume — emails written, variants generated, posts scheduled — because those numbers are easy to capture. But volume isn't impact. As marketing strategist Brianna Miller put it: "If you're still tracking the amount of stuff created as your primary AI metric, you're probably not inspiring your CFO."

What separates teams that prove ROI from teams that can't isn't tool choice — it's maturity. High-maturity organizations are twice as likely to achieve solid ROI, and they share a few common traits:

• A written AI roadmap — yet only one in four teams has one, and up to 75% of companies lack a real plan for the next 1–2 years
• AI embedded in CRM, marketing automation, and attribution systems rather than used as standalone point tools
• AI pointed at revenue bottlenecks, with business-impact metrics rather than content-volume metrics
• Governance and review processes — governance concerns grew 3.4x over the past year, and disciplined governance enables speed rather than blocking it

The maturity gap also explains why trust remains low inside organizations. Only 6% of leaders trust AI with high-stakes tasks like market positioning, 57% believe strategic thinking is AI's biggest weakness, and 88% say they must refine AI's work before using it. AI earns strategic trust the same way it earns CFO confidence: through measured, governed, embedded use — not scattered experiments.

For teams building toward that higher-maturity agentic stage, agent skills can automate the connective tissue between tools — and SkillzBench is building a free directory and MCP connector to discover, verify, and benchmark those skills, since published research found roughly 26% carry at least one vulnerability. Proven skills, measured impact: that's how the ROI gap closes.

Where AI Actually Moves the Needle: Personalization, Analytics, and Targeting

If you want to know where AI genuinely earns its keep in marketing, follow the money: personalization, audience targeting, and analytics consistently deliver the measurable lift that tactical uses like subject-line generation don't.

The personalization numbers are hard to ignore. Deloitte Digital's biennial research found that 52% of consumers are more likely to purchase from brands that personalize their experience, and those consumers spend 30% more. High-maturity personalization programs see a 62% conversion lift versus 41% for low-maturity teams, along with stronger retention.

But there's a catch hiding in the same data: a 58-point credibility gap. While 87% of brand leaders rate their personalization as good or excellent, only 29% of consumers agree. Brands think they're nailing it; customers mostly don't notice or don't care.

What customers actually value is telling. It isn't more clever targeting — it's personalization that:

• Proactively resolves issues before the customer has to complain (67%)
• Saves them money through relevant offers (61%)
• Delivers tangible loyalty benefits (58%)

In other words, the winning personalization feels like good service, not surveillance. That distinction matters given that consumer comfort with brands using AI has declined, from 57% in 2023 to 46% in 2024, according to Statista's marketing research.

Targeting and analytics round out the high-impact trio. A peer-reviewed study of 893 consumers found AI significantly improves social media marketing through personalized content, better influencer selection (matching creator audiences to brand targets), and real-time interaction. Meanwhile, AI agents are compressing analytics cycles: multi-armed bandit testing turns weeks-long A/B timelines into real-time optimization, and agents can analyze visitor behavior, select assets, and orchestrate across channels without constant human input.

The throughline in all of this is maturity. Teams that embed AI into their CRM and attribution systems and measure business impact — not content volume — are twice as likely to achieve solid ROI than those dabbling in "random acts of AI." And since agentic workflows are the next frontier here, it's worth being deliberate about which agent skills you deploy: published research found roughly 26% of agent skills carry at least one vulnerability. SkillzBench is building a free directory that verifies, security-scans, and benchmarks agent skills, plus a free MCP connector, so teams can adopt agents for personalization and analytics with their eyes open.

The lesson from the data: point AI at revenue bottlenecks, keep humans on strategy, and measure lift in conversions and retention — not output volume.

The New Frontier: AI Agents, Real-Time Optimization, and Generative Engine Optimization

The old model of marketing automation followed rigid if-then rules: if a visitor clicks, show this; if a cart is abandoned, send that email. AI agents break that mold entirely. Unlike traditional automation, agents make independent decisions using machine learning — analyzing visitor behavior, selecting assets, and orchestrating campaigns across channels without waiting for a human to write the next rule.

That shift shows up in five core agent workflows identified in industry analysis of agentic content systems: real-time visitor behavior analysis matched to relevant content, automated testing via multi-armed bandits, dynamic asset transformation, cross-channel orchestration, and predictive identification of content gaps. The bandit approach is the standout — it compresses traditional A/B testing timelines that took weeks into real-time optimization, shifting traffic toward winning variants as data arrives. Agents can also trigger nurture flows, adjust messaging mid-campaign, and suggest budget reallocations, executing marketing operations autonomously under defined rules.

Content operations benefit too. AI repurposing engines now break a single white paper or webinar into blogs, social carousels, and email sequences — extending the lifespan of every asset a team produces. The practical implication for marketers: the tactical layer is increasingly automatable, which frees humans for strategy and oversight. If you want to experiment with agentic workflows, agent skills can automate much of this pipeline — and SkillzBench is building a free directory and free MCP connector so you can discover, verify, and benchmark skills before trusting them to your agents.

There's a second frontier, and it's less about what your AI does and more about who's reading your content: AI itself. Discovery is now AI-mediated. According to Deloitte Digital's personalization research, 63% of consumers have used AI-powered search tools to find products, and 1 in 5 begin their search with an AI assistant. Meanwhile, 32% have purchased AI-recommended products and 18% discovered a new brand through one.

This changes the optimization playbook. Traditional SEO aimed at ranking pages for human clicks; generative engine optimization (GEO) aims at being found, verified, and cited by AI systems answering questions directly. Deloitte's guidance is clear: product pages, social posts, and campaigns must now be "visible, understandable, consistent and credible — not just to customers, but to the AI systems now informing them." In practice that means:

• Structuring content in modular, self-contained blocks rather than long narrative arcs
• Leading with clear, direct answers AI systems can extract and cite
• Keeping product and brand facts consistent across every channel an AI might scan
• Earning credibility signals — citations, attributions, authoritative sourcing — that generative engines weigh

The stakes are real. Brands already take an average of 4.7 business days to activate personalization from new customer data, and 97% say AI innovation is outpacing their ability to optimize. Teams that master both autonomous agents and AI-readable content will close that gap; those that don't will find their carefully crafted campaigns invisible to a growing share of buyers.

From Random Acts to Real Returns: A Practical AI Marketing Roadmap

Most marketing teams don't have an AI problem — they have an aim problem. With 91% of teams already using AI somewhere in their stack but only 41% able to prove ROI, the difference between the two groups isn't budget or tools. It's whether AI is pointed at revenue bottlenecks or scattered across "random acts of AI."

Map AI to revenue bottlenecks, not busywork. High-maturity teams start by identifying where revenue actually stalls — lead conversion, retention, activation — and embed AI directly into CRM, marketing automation, and attribution systems. According to MarTech's analysis, these high-maturity organizations are twice as likely to achieve solid ROI. As strategist Brianna Miller puts it: "If you're still tracking the amount of stuff created as your primary AI metric, you're probably not inspiring your CFO."

Start with low-risk, high-signal use cases — email first. Before attempting omnichannel orchestration, prove value where failure is cheap. AI-personalized email is the classic starting point because results are measurable within days, not quarters. Deloitte's research shows the payoff is real: personalization drives a +62% conversion lift in high-maturity brands versus +41% in low-maturity ones, and 52% of consumers say they're more likely to buy from brands that personalize.

Close the trust and skills gap deliberately. Consumer comfort with brand AI use has dropped to 46%, down from 57% the prior year, per Statista data. Meanwhile, 34% of businesses cite limited AI skills as a barrier, and governance concerns have grown 3.4x. The fix is structural, not aspirational:

• Establish human review processes — 88% of leaders say they must refine AI's work before use, so build that step in rather than pretending it away.
• Invest in AI literacy: hallucination checking, output validation, and understanding where AI is a thinking partner, not an autonomous strategist.
• Set privacy protocols that make data consent and usage transparent to consumers.

Measure business impact, not output volume. Tie every AI initiative to a pipeline metric — conversion, retention, cost per acquisition — and review quarterly. Disciplined governance, as one analyst noted, "isn't a chokehold, it's a launchpad."

For teams ready to push further into agentic workflows, agent skills can automate much of the review-and-optimization loop — and SkillzBench is building a free directory and MCP connector to discover, verify, and benchmark those skills, since published research found roughly 26% carry at least one vulnerability. Browse the waitlist at skillzbench.com to see what's coming.

Equipping Your Agents Safely: Why Verified Skills Matter for Marketing Workflows

Every marketing workflow we've covered — content creation, personalization, analytics — eventually hits the same question: who actually executes it? Increasingly, the answer is an AI agent. According to industry analysis, the next frontier is agentic workflows that trigger nurture flows, adjust messaging mid-campaign, and suggest budget reallocations autonomously.

But handing execution to agents raises the stakes. The same research shows governance concerns grew 3.4x over the past year, and only 6% of leaders trust AI with high-stakes tasks like market positioning. That caution is well-placed, and it extends to the tools agents use. Agent skills — the SKILL.md instruction packages that power agents like Claude Code, Codex, Cursor, and Gemini CLI — are proliferating fast, but published research found roughly 26% of skills carry at least one vulnerability. An unverified skill that drafts your emails or touches your customer data is a governance problem waiting to surface.

The verification gap matters because the payoff is real. Deloitte's research shows high-maturity personalization programs achieve 62% conversion lift versus 41% for laggards — and disciplined execution is what separates them. As one analyst put it, "Disciplined governance isn't a chokehold, it's a launchpad."

Before you equip an agent with skills, a few checks are non-negotiable:

• Provenance: every skill version should be pinned to a specific commit, so you know exactly what code your agent is running.
• Security and licensing: automated scanning and license checks catch the vulnerabilities and usage conflicts manual review misses.
• Measured performance: skills should be benchmarked in a sandbox — the same model, tested with and without the skill — so you know whether they actually help.
• Scoped access: your agent should only see the skills and tools you deliberately chose for it, nothing more.

This is the problem SkillzBench is being built to solve. It's a free, independent directory of agent skills — currently in development — that continuously crawls and dedupes skills from GitHub, validates and security-scans each one, and benchmarks them in a sandbox to publish the measured Lift. A free MCP connector will let you toggle skills per agent, so each agent sees only its chosen loadout, while upstream changes are tracked and re-scanned automatically. Everything — the directory, the benchmarks, the connector — is free, and money never touches scores.

Agent skills can automate the rest of this workflow: the GEO-friendly content rewrites, the repurposing engines, the real-time optimization. The teams that win won't be the ones that adopt agents fastest — they'll be the ones that equip them with skills they can actually trust.