Engageably

05 — Journal

← Back to Journal

Most marketing AI stacks will not survive the next platform shift

May 19, 2026 · 8 min read · Capability · Benjamin Rodriguez

The Exposure

The marketing AI spend booked between 2023 and 2025 sits in the wrong layer of the stack. Most of it lives in workflow-coordination tools, single-purpose vendors, and thin tools built on top of a prompt. Buyers priced and depreciated these on a single assumption: the vendor would still be operating, and still differentiated, three years out. That assumption is now the exposure. Gartner expects the agentic AI category to consolidate as supply outstrips demand. Capital-rich incumbents will absorb or wind down the undifferentiated vendors (Gartner, 2025).

The market is already pricing this in. 45% of martech leaders say the AI agents their vendors sold them are failing to hit the promised business performance. 81% are piloting or implementing them anyway (Gartner, 2025). The gap between what they bought and what works is where impairment lives.

For the CFO, the risk is the depreciation schedule and the multi-year SaaS commitment, both written on the assumption that the vendor keeps operating. The market no longer underwrites that assumption. For the CMO, it is the operational dependency on a tool that may be acquired, sunset, or quietly de-prioritized inside a larger platform before the contract term runs out.

Why It Is Surfacing Now

The structural trigger is simple. The companies that build the base models and the big cloud providers are now moving directly into the exact capabilities mid-market companies paid third-party vendors to deliver: workflow coordination, personalization, agent governance, content generation. AWS, Microsoft, and Google now account for 66% of cloud infrastructure spending. Each has launched agent-focused tools inside its own platform, from Bedrock AgentCore to Copilot’s agentic operations to Vertex AI’s governance layer (Omdia / ERP Today, 2026). Gartner notes that GenAI advantages are eroding faster than in prior innovation cycles. Within 36 months, baseline GenAI capability will be table stakes inside larger platforms, not a reason to buy a separate product (Gartner, 2025).

The platform shift does not make AI investment wrong. It makes the current integration layer the single most impaired asset class in most marketing budgets.

This is why the exposure is surfacing in 2026, and not earlier. The cloud platforms’ own agent tools are now mature enough to absorb the work. The contracts signed in the first buying wave are entering their first renewal window at the same time.

How the Risk Plays Out

The scenarios below are not predictions. They are trajectories already visible in vendor filings, cloud-platform product releases, and regulatory calendars. Read each row against your current vendor contracts and how deeply each vendor is embedded in your live operations. The rows where a vendor is most deeply embedded are the rows where the write-down will be largest.

ScenarioLikelihoodBusiness ImpactLeading Indicator
Workflow-coordination vendor is acquired or discontinues the product lineHigh through 2027 as consolidation acceleratesForced migration mid-campaign cycle, write-down of remaining contract value, integration rework across the martech stackVendor funding round at flat or down valuation, key product leadership departures, slowed release cadence
Cloud-platform built-in feature absorbs a single-purpose tool already in useHigh where the company is already on AWS, Azure, or Google CloudDuplicate spend on a capability now included in the existing cloud commitment, lost leverage on the single-purpose-vendor renewalCloud-platform roadmap announcement naming the exact capability, built-in preview features in the cloud the company already uses
Data portability failure blocks migration to a replacement platformMedium to high for any agent platform that stores its memory and tool setup only in the vendor’s own servicesMigration cost exceeds residual contract value, the asset becomes a hostage rather than a choiceVendor stores the agent’s state, memory, or tool connections only in its own dashboard or proprietary services (xpander.ai, 2026)
Incumbent vendor falls behind EU AI Act high-risk obligationsMedium, rising sharply toward August 2026Vendor becomes unusable for in-scope work, campaign pause, fines up to €35 million or 7% of global turnover for the operator (Legiscope, 2026)Vendor silence on conformity assessment, missing technical documentation, no published timeline for high-risk compliance
Internal team cannot operate the replacement platformHigh, given half of martech organizations report they lack the technical and data readiness and the talent to run AI agents (Gartner, 2025)Migration slips, replacement platform underused, second write-down on the replacement itselfHiring requisitions open for months, pilot projects stalling at the same integration step, reliance on a single vendor’s professional services

The Controls That Hold

The primary control is a living vendor dependency map, owned jointly by the CMO and the CFO’s technology finance function, reviewed on a fixed quarterly schedule. The map records, for every AI vendor in live use, the workflows that depend on it, the data that lives inside it, the remaining contract value, and whether the capability is now built into a platform the company already licenses. The team reviews it off-cycle the week a vendor announces a funding event, an acquisition, a leadership change, or a product discontinuation. Gartner expects fewer than one in five GenAI projects to deliver the business value originally promised through 2026 (Gartner, 2025). That is why this review cannot be annual. A year is longer than the half-life of the assumptions behind the contract.

The second control is the structure of every contract signed or renewed from this point forward. Three provisions are non-negotiable. A sunset clause that forces vendor notice and a defined transition window if the product line is discontinued or acquired. The right to export your data, covering the agent’s state and memory, your prompts, any fine-tuned model, and the tool setups, in a portable format, not just a download from the vendor’s dashboard. Termination-for-convenience with a pro-rated refund tied to a milestone the CFO can verify. Without these, the asset only lives in the vendor’s dashboard. And an asset that only lives in a vendor’s dashboard is not an asset. It is a hostage.

The third control is a capability audit run in parallel. Can the internal team operate the replacement platform before a migration is forced? That is the question most finance-led reviews miss. It is the one that turns a write-down into two.

Escalation and Ownership

The CMO owns the signal. That means monitoring vendor health indicators, product roadmap changes, cloud-platform feature releases that overlap a current vendor, and any compliance gap against the August 2, 2026 high-risk obligations under the EU AI Act (Legiscope, 2026). The CFO owns the threshold. That means defining, in advance and in writing, the impairment trigger at which the company renegotiates a contract, lets it expire, or writes it down instead of renewing it.

The decision path runs in that order. A leading indicator surfaces in the CMO’s quarterly review. The team matches it against the dependency map to size the operational exposure. If the exposure crosses the materiality threshold the CFO has set, the contract moves into renegotiation while the vendor still needs the renewal. That is the last window in which the company has pricing leverage. If the indicator is a confirmed sunset or a confirmed compliance failure, the contract moves to impairment recognition and the migration plan executes against the capability audit already on file. Waiting for a vendor to announce end-of-life is waiting past the last moment of leverage.

Executive Next Step

Within 60 days, the CMO and CFO should jointly audit every AI vendor contract signed since 2023 against three criteria: data portability in a usable format, contractual exit rights including sunset and termination-for-convenience, and whether the core capability is now built into a platform the company already licenses. Any contract failing two of the three goes into renegotiation this cycle.

Sources

← Back to Journal