We've recently had conversations with marketing leaders at two very large brands who were expressing skepticism about artificial intelligence and its ability to move humans out of the loop for complex marketing use cases like orchestrating multi-channel campaigns, personalized journeys, and loyalty incentives.
Within their organizations, they've successfully proven AI's value in generative content and data analysis, embedded within very specific point solution platforms, like Jasper for content writing or Personetics for analytics.
But the bigger promise that many industry-leading marketing technology vendors like Salesforce and Adobe are making, the one that paints a picture of marketers executing highly complex campaigns with the simple push of a button or a conversational chat, hasn't shown to be much beyond hype.
From a software sales perspective, the promise is on point. The need to get more out the door, at less cost, with fewer people and without introducing risk has been a significant pain point of campaign operations and marketing teams for nearly 20 years.
A significant swath of the martech ecosystem was built to solve this problem. First with simple triggers, then with rules-based automation, then robotic process automation, and now with AI.
Nearly 20 years later, however, the problem is still there. So when it comes to agentic AI as the latest solution flavor, leaders are naturally skeptical. And they're right to be.
Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, undone by escalating costs, unclear business value, or inadequate risk controls. Gartner also estimates that of the thousands of vendors marketing agentic AI, only around 130 are building solutions on top of frontier models with specifically-developed agentic capability. The rest is what they call "agent washing," established automation and chatbots repackaged with an agentic price tag.
Numbers like these make the point plainly.
It's hard to cut through the hype to know what's real.
And cutting through it isn't one leader's job. Bringing agentic orchestration online touches data, technology, operations, and the lines of business, and all of them need the same read of what's real.
Five pre-flight checks
- Check 1: No solution orchestrates across a mixed stack out of the box. Across the swath of enterprise-grade orchestration tools in the martech ecosystem, there are no platforms in-market today that run true cross-stack, cross-platform agentic orchestration on its own out of the box. True end-to-end orchestration across the mixed, multi-vendor environment a large organization runs has not been demonstrated at scale. For agentic orchestration, even leading vendors like Adobe, Salesforce, or Pega work primarily across their own platform applications. Orchestration with non-native applications is based on emerging protocols where agentic capability is still at an early stage.
The data supports this.
Boston Consulting Group's 2026 survey of 300 CMOs found only about 8% are running campaigns where multiple agents operate autonomously, while 42% still use AI to assist people with discrete tasks.
On Salesforce's own CRMArena-Pro benchmark, leading agents completed roughly 58% of single-step enterprise tasks and only about 35% of multi-step ones, a rate one analyst called a non-starter for enterprise use.
- Check 2: Orchestration cannot outrun the data underneath it. An agent making decisions at speed on fragmented, poorly governed data does not produce personalization at scale. It produces chaos moving faster. Salesforce puts it plainly in its own analysis: A brilliant model with bad data access makes confident mistakes.
Setting AI aside, marketing teams often struggle with pulling enough coherent context out of fragmented data to make the right segmentation, targeting, and personalization decisions. A system that can reach the data and still answer the wrong question, confidently and at volume, is worse than a slower operation working from sound, complete data.
In a regulated context, moving faster than the data can support can be catastrophic in the size and scale of fines and reputational damage to the brand.
In 2025, all 50 states introduced AI-related legislation, more than 1,200 bills in total, and by early 2026 states had introduced over 1,500 more. A consistent target across them is automated decision-making that materially affects consumers, with transparency, risk-assessment, and anti-discrimination requirements attached.In financial services the bar is higher still, with a February 2026 U.S. Treasury framework setting more than 200 operational control objectives for AI governance. - Check 3: No tool on the market cleans the data autonomously. The tempting shortcut for a fragmented data estate is to find a product that ingests the mess and returns a governed, orchestration-ready foundation, and then run the orchestration on top. That product does not exist. Scott Brinker and Frans Riemersma's Martech for 2026 research found data quality to be the single largest obstacle to AI implementation, with 56.3% of respondents reporting that missing, stale, or inconsistent data was holding their AI back. The tooling has not resolved this. It has surfaced it.
The reason is structural. Most enterprise martech platforms currently implemented in enterprise marketing contexts, including CDPs, warehouses, and attribution tools, were not built with agentic AI in mind, and out of the box they rarely meet its requirements. Identity resolution, real-time data refresh, and schema alignment across systems take deliberate engineering and ongoing governance. AI can assist inside that work, matching records and flagging anomalies, but the methodology choices still require human judgment, and agentic systems assume clean inputs and clear decision authority that most organizations do not yet have. The foundation gets built the hard way or it does not get built.
- Check 4: The connective tissue for cross-system orchestration is still early. Reaching across a mixed stack depends on protocols that let agents talk to tools and data across vendors, and the leading one, the Model Context Protocol, is barely more than a year old. Anthropic introduced it in late 2024 and only handed it to a Linux Foundation body at the end of 2025, with its first enterprise-ready specification due in mid-2026. In a May 2026 advisory, the NSA and CISA noted that the protocol's spread has outpaced the development of its security model, and that its early design left real ambiguity around safe use. Production deployments keep hitting the same gaps. Things like no standardized audit trails, authentication tied to static secrets, and undefined behavior once a request passes through a gateway.
Those gaps are not academic in a regulated setting. Audit trails and authorization controls are compliance requirements, and where they are missing, autonomous decisioning at scale is not something a compliance team or a regulator will sign off on. The deployments that do work at scale show the pattern. When Pinterest put MCP into production in 2026, it did so with human approval built in for sensitive operations, alongside governance and gateway controls. The protocol is promising, and it is moving quickly. It has not yet reached the point where a regulated organization can safely run agentic orchestration across its stack without a human in the loop.
- Check 5: The benefit isn't worth the effort — yet. The business case for agentic orchestration remains highly volatile and, for most organizations, is reason enough to pause.
On the cost side of the equation, AI is shifting the software pricing paradigm from subscription-based to consumption-based. Unlike traditional SaaS per-seat license costs that can be planned in advance, agentic orchestration prices against usage. And the cost of token consumption is still a moving target.
In its June 2026 analysis of agentic AI costs, EY documents that a single interaction rises from four cents to $1.20 once it becomes an orchestrated process that retrieves, plans, and calls tools. Beyond the visible cost, the EY analysis frames the full cost of an agentic system as seven categories and finds most organizations budget only the first three. There is broad consensus among practitioners, analysts, and vendor-driven research that 30-40% of the total cost for agentic solutions are "below the line" of the AI tool, in the data, operations, and governance needed to support them. Gartner found in a 2025 survey that 63% of organizations either lack or are unsure they have the right data management practices for AI, and initiatives that start without addressing that foundation consistently run over budget in later phases.
The return side of the equation is no more certain. In S&P Global's 2025 survey of more than a thousand enterprises, 46% of the organizations that had invested in generative AI reported that not one enterprise objective had seen a strong positive impact from it. Abandonment is climbing at the same time, with the share of companies scrapping most of their AI initiatives before production rising from 17% to 42% in a single year.
For a contained pilot, that uncertainty is acceptable. That is what a pilot absorbs. Cross-stack orchestration is the opposite of contained. It is among the most time-consuming and resource-intensive undertakings a marketing organization can commit to, and standing it up means putting serious budget and years of effort against a cost that keeps climbing and a return most organizations cannot yet demonstrate.
Few enterprises want to be first through that door without a defensible case, and today that case cannot be built.
Get ready for what's next
Leading martech vendors are making strong progress toward the vision of agentic orchestration of multi-channel campaigns across the tech stack with minimal human intervention. But it's still a vision that's several years away from a proven solution.
The ambition driving the vision isn't wrong. The marketing technology ecosystem has always run on casting and chasing its own vision. What's different about this particular innovation cycle is the hype surrounding it. While not necessarily a "bubble" in the truest sense, AI is pushing organizations to commit budget and make mandates before viability has been proven for their business. Gartner's 2026 Hype Cycle places agentic AI at the peak of inflated expectations, with only 17% of organizations having deployed AI agents so far while more than 60% expect to within two years, the most aggressive adoption curve of any emerging technology Gartner has measured.
The vendor conference circuit compounds it. Adobe, Salesforce, Oracle, and Databricks all fill keynote stages and panels with customers describing the success they have had with AI. When a CMO watches peers report those wins, the urgency to follow becomes real, whether or not the foundation is there to support it.
Recently, a marketing leader described this exact bind. Her CMO, new to the organization and arriving from a leading digital brand, moved quickly to issue substantial AI mandates. She was left doing two things at once. Hunting for solutions that do not yet exist. And drawing on her own tenured experience, which told her that AI is not a magic answer to a problem that's 20 years in the making.
The instinct for any marketing capabilities leader in that position is to tell the CMO that AI isn't up to the challenge. But the better approach is leading the CMO to redirect their urgency toward fixing the foundation.
As the in-market solutions for cross-capability orchestration mature, the time is best spent maturing the data, business architecture, and operational governance that will let an agentic solution come online later at lower total cost of ownership and a faster path to value.
Establishing readiness for what's next, whatever the transformation demands, is the core discipline of marketing capabilities maturity. When agentic orchestration matures, the organizations that did the work to get ready will benefit, while the ones that rushed into early AI adoption are still walking back the collateral impact of moving too fast.
