90% use GenAI. 50% run agentic. 100% need better accountability. Martech Futurist | July 14, 2026
Nine in ten US marketing agencies now run generative AI, and half run agentic AI in live execution. Producing the work has stopped being the hard part. This week's research points at what replaced it. Forrester and the 4As found agencies booking AI as a cost of business and harvesting its productivity for margin, while their clients rarely ask for growth. Harvard Business Review showed that one digital and AI strategy fractures across a company's different selling models, because the decision rights differ in each. A Forrester field report described operators wiring their own AI workflows around processes that move too slowly for them. A MarketingProfs piece traced how brand strategy stalls the moment nobody owns the decisions it implies. The capability is now ambient. What decides whether it compounds into growth or settles into overhead is governance: who holds the decision rights, who sets the guardrails, and who answers for what the machine does.
That marks a shift from where this column landed last week. The measurable-return problem read as a routing problem, capability pooling where it was cheapest to count. The signal this week sits one layer deeper. Once the capability is everywhere, the scarce input becomes accountability for it.
Three themes that emerged
1. Capability went ambient. Accountability became the scarce input. When a capability saturates a market, it stops separating the leaders from everyone else. Generative tools sit inside creative, strategy, media, and reporting workflows across almost every agency, and agentic execution is now standard in half of them. The differentiation moves off the act of producing and onto the conditions around it: whether the output can be trusted, traced, and stood behind. Those conditions carry the reliability, legal, and privacy concerns that agencies themselves name as their top barriers, and they sit with a human who has to own the result. Jaclyn Wands of Phaedon put the accountability posture plainly when I interviewed her on The Agile Brand podcast: "As a leader in marketing, you need to have at least a conceptual understanding of just how frequently these models can be wrong. And I quote my favorite statistician, E.P. Box, 'All models are wrong, but some are useful.' And when you approach your AI strategy with this fundamental understanding, you are more likely going to implement fail-safes and the ability to do human in the loop checking that allows for a more powerful integration of AI." The capability is assumed. Owning its fallibility is the work.
2. The ledger line sets the ceiling. How a company files AI on its books governs how much value it can pull from it. Sixty-one percent of agencies treat AI as a cost of business. Six percent treat it as a line of business, a capability sold to clients. Booked as a cost, AI gets optimized downward toward margin and headcount. Booked as a capability, it gets governed, developed, and pointed at growth. The demand side reinforces the frame: only 27 percent of agencies say clients ask for revenue growth as the primary benefit of generative AI, and just 11 percent say the same about AI agents. Efficiency is an outcome. Filing a capability under one of its outcomes caps what the capability is allowed to become.
3. Governance is the operating discipline now. Three of this week's four sources converge on the same mechanics: decision rights, guardrails, and a named owner. HBR shows why one AI strategy cannot govern a company that sells through digital-first, hybrid, and relationship-led models at once, since each model assigns different decisions to people and to systems. MarketingProfs shows brand strategy failing to scale until someone defines which calls a team makes independently and which ones escalate. Forrester's field research shows what fills the gap when governance lags: operators route around slow processes and build their own AI workflows, and the function that owned the work risks being bypassed. Renu Upadhyay of Omnissa named the tension directly when I spoke with her on The Agile Brand podcast: "Marketing is moving at the speed of AI, sometimes faster than the governance, the processes, the operational models that an organization has been able to keep up with." She was careful to add that the workaround is not defiance: "Shadow IT isn't intentional. It's the line of business, marketing, trying to accomplish a business outcome, for which they're leveraging this technology." Governance closes that gap by deciding, in advance, who is permitted to act and where a human has to sign.
Featured Insights
Forrester and the 4As. Agencies Are Using AI to Boost Margins, Sacrificing Growth. July 6, 2026. A new Forrester study conducted with the 4As, based on an April poll of nearly 200 agency decision-makers at the VP level or higher, finds generative AI in use at 87 percent of US marketing agencies and agentic AI running marketing execution at 50 percent. Productivity and cost efficiency drive adoption, with 81 percent using generative AI to lift staff productivity. Sixty-one percent classify AI as a cost of business and only 6 percent as a line of business sold to clients. Jay Pattisall, who co-authored the report, said the industry "is at risk of mistaking efficiency for effectiveness" and that agencies are "mired in a legacy business model in which they are harvesting the productivity of AI for margin." Reliability and bias, legal risk, and privacy top the list of barriers to scaling. Audit how AI appears in your own plan and reporting. If it shows up only as a cost to reduce, your measurement is capping the return before the work starts.
Harvard Business Review. Tailor Your Digital Strategy to Reach Every Customer. June 29, 2026. Prabhakant Sinha, Arun Shastri, Sally Lorimer, and Saby Mitra of ZS argue that most companies run several go-to-market models at once and then try to govern all of them with a single digital strategy. That approach fails because the tools, the AI systems, and the decision rights have to be designed differently for digital-first, hybrid, and relationship-led selling. The authors press leaders to align design, decision rights, and governance to each model, and to keep adapting as customers and technologies move. Map your go-to-market models before you standardize a stack. Assign decision rights per model, so the same AI system does not carry authority it was never meant to hold in a relationship-led sale.
Forrester. The Rise of the "Claude Cowboy" in RevOps. Late June 2026. Clinton Fitch describes a new archetype: commercially minded operators using agentic AI and low-code automation to solve operational problems fast, often around formal processes that cannot keep up with the demand for answers. As AI lowers the cost of building reports, models, and workflows, insight commoditizes, and the value of the function moves from producing outputs to interpreting signals, managing risk, and guiding decisions. The counsel is to build guardrails, set standards, and redefine the function around judgment and governance, because the real risk is that the function fails to evolve and gets bypassed. Treat shadow AI as a demand signal, not a violation. Where your team routes around process, the process is too slow. Fix the guardrails so speed and governance stop competing.
MarketingProfs. Why Brand Positioning Must Drive Decisions, Not Just Messaging. July 7, 2026. Diana Rain shows positioning breaking down after approval, as sales reverts to old pitches, teams interpret the message differently, and partnership calls keep escalating to the founder. Her fix is governance built for decisions: define what must stay consistent, where teams can adapt, which calls they make independently, and when a decision needs review. She insists on a named owner for the governance system and warns that too little governance breeds inconsistency while too much creates bottlenecks. The test of positioning, she writes, is whether the organization can make aligned decisions when the strategist is no longer in the room. Name an owner for your brand and AI decision rules. Guidance that no one maintains decays into the same escalation loop it was meant to remove.
Key Takeaways
Adoption has stopped being a differentiator. With generative AI near-universal and agentic execution at half of agencies, the edge forms in the conditions around the output, not the output itself.
Watch the ledger line. AI filed as a cost of business gets optimized toward margin. AI treated as a capability gets governed and pointed at growth. The classification precedes the strategy.
Governance is concrete work: decision rights per selling model, guardrails that let teams move fast inside clear limits, and a named owner who keeps the rules current.
Shadow AI is diagnostic. When capable people build around your process, the process is the problem. Read the workaround before you police it.