This article was based on the interview with Kaz Ohta and Karen Wood, CEO and CMO at Treasure Data by Greg Kihlström, AI and MarTech keynote speaker for The Agile Brand with Greg Kihlström podcast. Listen to the original episode here:
For years, enterprise marketing leaders have been engaged in a monumental construction project. With blueprints drawn up by the major marketing clouds and materials sourced from an ever-expanding MarTech landscape, you’ve meticulously built intricate technology stacks. The goal has always been the same: a unified, intelligent system capable of delivering personalized customer experiences at scale. Millions of dollars and countless hours have been invested in this pursuit. We’ve all seen the diagrams, marvelled at their complexity, and worked tirelessly to integrate the disparate pieces. But what if the very foundation of this structure, the one we’ve so carefully assembled, is fundamentally flawed for the era we are now entering?
The arrival of sophisticated AI is not merely adding another floor to our construction project; it’s conducting a seismic stress test on the entire foundation. The cracks are beginning to show. The fragmented data, disconnected systems, and walled-garden architectures that we’ve learned to work around are now becoming critical points of failure. AI, with its insatiable appetite for clean, comprehensive, and connected data, is finding itself with a form of digital amnesia, unable to recall a customer’s full history because it’s scattered across dozens of systems. This isn’t just an inconvenience; it’s a barrier to realizing the transformative potential of artificial intelligence. It forces a necessary, and perhaps overdue, conversation about reinventing our marketing architecture from the data up.
The Architectural Flaw: When More Tools Mean Less Intelligence
The paradox of the modern MarTech stack is that in our quest for best-of-breed capabilities, we often create a “worst-of-all-worlds” data environment. The digital transformation push, accelerated by the pandemic, led to a SaaS buying spree. While each tool solved a specific problem, the collective result was a spiderweb of data silos. Now, as the economic climate tightens and the C-suite demands consolidation and proven ROI, this complexity is being scrutinized. More importantly, as Kaz Ohta points out, this fragmentation is the single biggest impediment to effective AI.
“AI needs better data. If you’re garbage in, garbage out. And the more tool you have, you just have more fragmented data… The CIO and CFO is asking, okay, is this underutilized? You have to get rid of it, et cetera, et cetera. So at best the IT budget, marketing budget, it’s probably flat or even probably decline. So that’s one force for consolidation. The other one obviously is AI, which is if you are using one platform, you have clean data, it’s more consolidated, right? So that powers better AI.” – Kaz Ohta
Ohta’s point is one that resonates deeply with any leader who has tried to stitch together a coherent customer journey from a dozen different reports. The “garbage in, garbage out” principle isn’t new, but its consequences are magnified exponentially with AI. An AI model trained on incomplete or contradictory data won’t just produce a flawed report; it will generate misguided segments, write irrelevant copy, and orchestrate clumsy journeys. The push for consolidation, therefore, is being driven by two powerful forces: fiscal prudence from the CFO and a functional necessity dictated by AI itself. To build a truly intelligent marketing engine, the starting point can no longer be the channel or the campaign tool; it must be a clean, unified data core.
From Manual Tasks to Orchestrated Agents
The promise of AI in marketing isn’t just about doing the same tasks faster; it’s about fundamentally changing the operating model. The concept of AI agents—specialized, autonomous systems that can execute complex, multi-step tasks—represents this shift. However, early hype often glosses over a critical reality: general-purpose AI is still not reliable enough for high-stakes enterprise work. An AI that completes a complex task with a 50% success rate isn’t a helpful colleague; it’s a coin flip. The practical path forward, as Ohta describes, is through a collection of specialized agents, orchestrated by a “Super Agent,” that can reliably execute narrow tasks and improve over time.
“We created a bunch of small task agents. And then we also built this one orchestrator agent that actually splits the tasks into multiple small agents’ tasks… What’s amazing about this is, so we released one first AI agent called audience agent. It’s like creating segments, explore some data about it. And a lot of tasks took almost like 10 minutes to finish. And after 12 months, we can finish within a minute now. So there was a 10x improvement.” – Kaz Ohta
This is the pragmatic application of AI that enterprise leaders should be focused on. Instead of waiting for a mythical AGI, the strategy is to build an ecosystem of agents—one for research, one for data analysis, one for segmentation, one for journey creation—that can be called upon by an orchestrator to complete a larger goal, like launching a new nurture campaign. The 10x performance improvement Ohta mentions in just 12 months illustrates the compounding power of this approach. As these agents become faster and more accurate, the operational efficiency gains become exponential. This moves the marketer’s role from a hands-on-keyboard operator to a strategic director of an AI-powered team, defining the objectives and letting the agents handle the complex execution.
The Overlooked Imperative: AI, PII, and Permissions
As we grant AI agents more autonomy and deeper access to our systems, a critical and often-underestimated challenge emerges: security and governance. In a traditional SaaS world, permissions are managed at the application level. But when a single AI agent has the theoretical ability to operate across your entire stack—from your CDP to your email platform to your ad network—it creates what could be called a “super permission” risk. A breach of this single agent could expose every piece of customer data you hold. This is not a theoretical risk; it’s an existential one. Any viable enterprise AI strategy must be built on a robust, granular permissions framework.
“If AI can operate all the SaaS, that means this has a super permission. If this agent gets hacked, you basically leak every single record of every single system. That’s not going to be sustainable. So what we have built is this permission system that tightly integrated with CDP and AI on top of it… This agent recognizes you’re the CEO permission. We merge two permission together, agent and human, and we operate in this realm… The benefit of building AI on top of CDP is you can actually inherit all of the permission control you’ve built for your own PII data.” – Kaz Ohta
This is arguably one of the most important considerations for any marketing leader evaluating AI solutions. The idea of merging the permissions of the human user with the permissions of the AI agent is a sophisticated and necessary safeguard. An intern asking an AI agent about company revenue should receive a different answer than the CEO asking the same question. By building the AI layer directly on top of the CDP, the system can inherit the rigorous data governance, access controls, and consent management already established for the company’s most sensitive asset: its customer PII. This approach transforms the CDP from a simple data repository into the central nervous system for secure AI operations, ensuring that efficiency and innovation don’t come at the cost of trust and compliance.
A New Era of Marketing Intelligence
The journey we’ve been on as marketing leaders has brought us to a critical inflection point. The architectures we painstakingly built for the last decade are proving insufficient for the demands of the next. The rise of AI is forcing us to confront the data fragmentation and system silos that have been persistent, low-level frustrations and elevate them to the status of mission-critical problems to be solved. Simply layering AI on top of a broken foundation won’t work; it will only accelerate our ability to make data-poor decisions. The future requires a deliberate, data-first approach where a unified and secure customer data platform is not just a component of the stack, but its intelligent core.
This shift is more than just technological; it’s strategic. It redefines the role of marketers from campaign mechanics to architects of intelligent systems. It’s about leveraging orchestrated AI agents to handle the tactical execution, freeing up human talent to focus on strategy, creativity, and customer empathy. The evolution of an event like CDP World into “Agentic World,” as mentioned by Treasure Data’s team, is a clear signal of where the industry is heading. The conversation is no longer just about unifying data; it’s about securely activating that data with a new class of intelligence. For leaders who embrace this change, the opportunity is not just to build a better marketing machine, but to create more meaningful, respectful, and valuable connections with customers than ever before.
