You Can't Go to Mars on a Golf Helicopter
"The measure of a platform is not the features it lists — it is the architecture it was built on."
— The question every CIO should ask at renewal timeBuilt for a Different Planet
In 2003, NASA's Mars Exploration Rover program put two rovers — Spirit and Opportunity — on the surface of Mars. Neither was a modified car. Neither was a repurposed aircraft. They were designed from the ground up, for the specific physics of another planet: low gravity, abrasive regolith, extreme temperature swings, no GPS, no rescue. Every gram was engineered for the mission. That is what purpose-built looks like.
Now picture your ITSM vendor's AI story. The platform was designed in the mid-2000s or early 2010s. It was built to log tickets, route approvals, and close incidents. It worked. Then, over the last five years, the vendor attached a virtual agent here, a predictive intelligence module there, a "Now Assist" or "ITSM.ai" wrapper on top. The core architecture — the relational data model, the workflow engine, the process logic — remained structurally unchanged. Only the brochure changed.
Sending that platform to the AI era is like trying to reach Mars on a helicopter designed to ferry golfers across a country club. It is not a question of how much fuel you add. The vehicle was never designed for the mission.
The Feature-Layering Illusion
There is a pattern that repeats reliably across the history of enterprise software. An incumbent builds a dominant platform for the problems of its era. It accumulates customers. It accumulates revenue. Then a paradigm shift arrives — cloud, mobile, now AI — and instead of rebuilding, the incumbent acquires, wraps, and rebrands. The result looks like evolution. It is mostly cosmetics on a structure that was never designed for what it now claims to do.
ServiceNow's acquisition of Moveworks in March 2025 for $2.85 billion is a clear data point. Moveworks was built as a standalone conversational AI layer — an intelligent assistant that could understand intent and navigate enterprise systems. ServiceNow did not grow that capability organically. It bought a helicopter and is now trying to dock it to a space station. The integration story will take years. The seams will show.
BMC Helix followed a parallel path. Cognitive automation, predictive service management, AI-assisted incident correlation — all were acquired or assembled from external components and layered atop a process-management core that predates the iPhone. The underlying data structures were designed to track work items, not to reason about them.
"Bolting AI onto a legacy ITSM platform is not transformation. It is renovation theater. The wiring behind the walls is still the same."
This matters in ways that are not immediately visible in a product demo. AI does not just need a front-end. It needs a data substrate that captures relationships, context, and intent — not just event logs. It needs a workflow engine that can reason about processes dynamically, not just execute predefined approval chains. Legacy ITSM architectures were not built for any of this. They were built to move tickets through queues.
45% of enterprise leaders whose organizations are piloting or running vendor-supplied AI agents say those capabilities do not meet their company's expectations for promised business performance. Half of all surveyed leaders also reported that their organizations lack the technical and data stack readiness required for AI agent deployment.
Source: Gartner, "Gartner Survey Finds 45% of Martech Leaders Say Existing Vendor-Offered AI Agents Fail to Meet Their Expectations of Promised Business Performance," October 29, 2025
Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. Gartner also identified "agent washing" — the rebranding of existing chatbots and RPA tools as AI agents without genuine agentic capability — as a significant market problem, estimating only around 130 of thousands of claimed agentic AI vendors are credible.
Source: Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027," June 25, 2025
Four Questions Before Your Next Renewal
Before signing the next three-year agreement, every CIO and service operations leader should pressure-test their vendor's AI story with these four diagnostics. Not during the demo. During the contract negotiation.
Was AI architected in — or added on?There is a structural difference between a platform designed from day one to reason about data and a platform that bolted an AI services layer on top. Ask the vendor to show you the data model changes made to support AI in the last five years. If the answer references an acquired product or third-party integration, the answer is "added on."
Can the data model support contextual intelligence?AI reasoning requires relationships between entities — user, asset, incident, resolution, knowledge, access policy. If the underlying schema is still a flat relational model built around ticket states and workflow nodes, the AI being sold is operating on impoverished data regardless of how capable the model itself is.
Is the workflow engine adaptive or merely automated?Automation executes predefined rules faster. Intelligence modifies the rules based on context. If every process still requires a Change Advisory Board template and a nine-step workflow configuration, the platform is faster — not smarter. Real AI requires the workflow layer to be a runtime reasoning surface, not a static flowchart executor.
Are you buying transformation or feature accumulation?Count the new modules added in the last three releases. Count how many required core architectural changes versus how many were new licensing SKUs layered on top. If the ratio is three modules to zero re-architectures, the vendor is selling complexity packaged as sophistication.
The Risk Nobody Is Discussing
FOMO — the fear of missing out — is a powerful force in enterprise technology decisions. Every CIO receives decks showing competitors deploying AI chatbots, automated resolution pipelines, and predictive capacity management. The social pressure to be seen as modern is real. And the vendors know it. The annual conference keynote exists precisely to turn FOMO into renewal signatures.
But the risk that deserves attention is not moving too slowly. It is encoding your most valuable institutional knowledge — your processes, your resolution patterns, your access logic, your service relationships — inside a vendor's proprietary AI stack. When the AI is architecturally inseparable from the ITSM platform, your knowledge becomes theirs.
This is not hypothetical. When Moveworks gets fully absorbed into ServiceNow's platform, every organization running Moveworks-powered automations will face a binary choice: migrate into ServiceNow's unified model — on ServiceNow's timeline, pricing, and terms — or find an exit that requires rebuilding from scratch. The knowledge those automations embody stays with the platform. The organization walks away with logs.
Vendor-native AI creates a second layer of lock-in on top of the first. The first lock-in was your process data and workflow configurations. The second is your trained AI behaviors, resolution pathways, and automation logic. Two layers of captivity — but you only signed up for one. Most organizations discover the second layer at renewal time, when leverage is lowest.
The enterprises best positioned in three to five years are those treating AI not as a feature purchased from their ITSM vendor, but as an independent capability they own — one that sits above the ITSM platform, interfaces with it through standard APIs, and can be moved, retrained, or replaced without dismantling operations. This is vendor-agnostic AI architecture in practice. It requires more upfront intentionality. It provides substantially more long-term optionality.
"Real leaders do not chase the demo. They secure the exit before they enter the room."
What AI-Ready Actually Looks Like
The next era of ITSM will not be won by the platform with the most features. History is clear on this: the feature-maximalist rarely wins the technology generation shift. It will be won by architectures that are simpler at the core and smarter at the edge — where intelligence operates on clean, portable data and the AI reasoning layer is structurally decoupled from the process execution layer.
Your resolution patterns, escalation logic, and service relationships represent years of operational learning. That knowledge should exist in a format your organization controls — structured, versioned, and exportable — not embedded inside a vendor's proprietary fine-tuned model you cannot inspect, port, or own.
Your ITSM platform was designed to record, track, and route. Let it do that well. Bring AI in as a separate orchestration layer that reads from and writes to the ITSM platform through standard APIs, but is not architecturally dependent on it. This boundary pays dividends at every contract renewal.
A demo is optimized for the best-case scenario on a clean tenant. Ask vendors to show you data schema changes made to support AI in their last major release. Ask what happens to your automation logic if you terminate the AI SKU. The answers will reveal whether the AI is native or cosmetic.
The model performing best today may not lead in 18 months. Organizations building on a single vendor's embedded model inherit that vendor's model strategy. Multi-LLM orchestration — where the AI layer routes to different models by task type and performance — is the architecture that survives rapid model evolution. Gartner predicts that by 2028, 30% of total global enterprise AI spend will shift to open, domain-specific models precisely for this reason.
When a technology decision is primarily driven by what competitors appear to be doing, slow down. FOMO-driven purchases rarely survive the three-year review intact — and Gartner's prediction that 40%+ of agentic AI projects will be canceled by 2027 suggests the market is already beginning to feel this. Real leadership is not being first to deploy AI features. It is deploying AI in a way you will not have to unwind.
A Diagnostic Before Your Next Signature
Run this checklist before your next ITSM renewal or AI platform evaluation. None of these require a vendor proof of concept — only honest answers from the account team.
- Can the vendor demonstrate an AI use case — not a demo — running natively on your existing data model without custom development?
- Does AI licensing exist as a separate SKU, or is it included in core platform pricing at no additional cost?
- If you terminate AI features, what happens to the automation logic and resolution models built on top of them?
- Can your AI automations interface with non-ITSM systems — HR, ERP, identity management — without requiring the ITSM vendor as the integration hub?
- Has the vendor published an architectural roadmap — not a feature roadmap — showing how the core data model is evolving to support AI natively?
- Does the vendor support exporting AI training data, resolution history, and automation logic in an open, portable format?
- If this vendor is acquired in the next 18 months, what contractual protections ensure continuity of your AI capabilities and current pricing?
"The organizations that will lead the next decade of service operations are already making a different choice — not the loudest AI story, but the most defensible AI architecture."