How to Reject a Platform Pitch and Scope a Single Facility Fix Instead
Start with one broken process, a named owner, and a ninety-day test.

AI operations transformation in asset-heavy industries can start without a plan for a platform. The practical alternative is to reject enterprise-wide scope, audit one painful facility process, and define a deployment with a named owner, a measurable loss, and a ninety-day success condition. The decision is not whether to adopt AI, but which operational result earns the first commitment.
Why vendor platform pitches are growing more aggressive even as enterprise AI success rates fall
The pitch offers to connect procurement, scheduling, logistics, quality, and workforce management in one transformation program. Each department’s problems become evidence for the same purchase. The roadmap divides the work into phases and gives the ambition a timetable. It looks like an operating plan before anyone has examined an operating process.
Deloitte’s AI in Manufacturing 2026 survey found that most manufacturers report measurable value from AI, but only roughly one in five use cases has been scaled across sites or enterprise-wide. That gap gives vendors an opening to sell broader adoption as the missing step. Yet the HFS Research–Rackspace survey found that only about a third of large AI programs met or beat their business case, with success dropping further when AI spanned several functions. An operations leader therefore needs a way to separate a credible first engagement from a persuasive transformation promise.
What the platform approach actually does to a process that already has problems
A platform-first deployment scales infrastructure before a single workflow is trusted. If that workflow contains unresolved failures, automation can accelerate them. A system can connect departments without checking whether the output they pass between them can be trusted.
Construction scheduling makes the mechanism concrete. AI supply chain tools work from real-time data, but baseline schedules can contain broken logic ties and durations without grounding. An AI layer operating on such a schedule still depends on relationships and assumptions that have not been validated.
Making AI understand the work therefore takes priority over simply introducing it. That requires identifying the job, the owner, and the outcome before selecting the tool. Scope provides the boundary within which those relationships can be examined. Without that boundary, a program can spread the same unresolved workflow assumptions across several functions before anyone has established a dependable first result.
How to recognize a platform pitch before it has taken root
The first signal is a proposal organized around departments, with the failure to be addressed still unidentified. “Procurement, scheduling, and workforce management” describes coverage. The distinction is visible in the supporting evidence.
The second signal is a roadmap measured in quarters or phases without a named first deliverable. A live first process in weeks is absent. The third is a diagnostic built around a survey or workshop. No one has asked to see how the work runs today, where it stalls, or what people do when the formal workflow fails.
The fourth signal is ROI attached to the program rather than a workflow an operations leader can own and verify. A credible aggregate number still leaves that accountability unresolved. Deloitte’s survey identified high implementation costs as the most frequently cited implementation challenge, reported by 43% of respondents. Wide scope and diffuse ownership compound that burden because the commitment grows before a specific operating result has been established.
The language that redirects a platform pitch without killing the relationship
The useful response challenges the proposed scope while respecting the vendor’s legitimacy. Specific language keeps that distinction clear. “Before scoping the program, show the one process where the platform has the most evidence.” This invites the vendor to distinguish production references from pilots and gives a capable supplier room to propose a narrower engagement.
“Walk through what a live first process looks like at week six.” A concrete response establishes what would be running and who would be using it. A response that stays at the roadmap level leaves the first deployment undefined.
“The process needs to be audited before the solution is scoped.” “Let’s agree on the one KPI this first deployment moves, and what it needs to show in ninety days.”
Wharton’s research found that departments reporting strongly positive ROI were far more likely to have changed their management structure as a result of AI. That association supports a more demanding conversation about deployment without establishing that restructuring alone produces returns. The vendor needs to explain how the work will be organized.
How do you choose one broken plant process to fix with AI?
The strongest first candidate has a measurable loss, a single named owner, and a workflow failure that AI can address. Visibility is not the deciding criterion. Neither is strategic prominence. The loss must be named and quantified before the fix is built, because a failure that is felt but not counted cannot support a verifiable before-and-after comparison.
Ownership makes that comparison actionable. The first process needs one person who can say that it broke and that the intervention worked. If the proposed deployment must cross three departments before it produces an output, it already looks like the broader program being put off. The boundary is too wide for a clean first test.
The workflow criterion requires equal care because accurate predictions alone do not constitute an operating process. The intended use of the prediction, the responsible person, the response time, and the way the result will be recorded must be defined. Missing answers indicate an unresolved foundation. The first task is to establish the information and responsibility on which a usable workflow depends.
For construction and logistics operations, procurement lead time visibility provides a sharper diagnostic. If the team cannot state the actual lead time for the longest-lead item in a current project, that information gap is the first process to fix. Every downstream schedule depends on an unverified number. The operations leader should leave the meeting ready to answer one question. “Which process, right now, is producing outputs that someone is acting on without being able to verify they are correct?”
What a scoped process audit looks like in practice, and why it has to come before the solution
The operations leader runs or commissions the audit before accepting a deployment scope. The audit compares the documented workflow with the work people actually perform, then establishes which failure deserves intervention. A process document demonstrates only that a procedure exists.
An approval may be treated as a formality. People may carry out a scheduled reconciliation without looking into differences, or bypass a system control by working offline. These distinctions change the scope because they reveal whether the proposed AI would enter a functioning hand-off or depend on a control that exists only on paper.
The 2025 IIA Global Internal Audit Standards, effective January 9, 2025, require explicit linkage between the audit plan and enterprise risk. The same discipline applies to AI scoping. Findings should show how a departure from documented procedure puts the work at risk. Corrective action also needs evidence, because marking a plan complete does not establish that the underlying process has changed.
Planning sets the bounds of the process and its risk, while Fieldwork looks at how people do the work and handle hand-offs. Analysis links the failure found in the work to the proposed scope. Follow-up establishes whether the corrective change can be demonstrated.
The leader asks the vendor to show how an agent fits into the existing hand-off between roles. That request tests whether the design assumes a cleaner process than the facility has. If it does, the discrepancy belongs in the scope decision before a commitment.
What the platform vendors get right, and where their objection deserves a real answer
A fix for one facility can create integration debt. It can also reach a practical limit when the next improvement requires adoption across functions. The disagreement concerns the sequence in which shared infrastructure and wider coordination will become necessary.
Gartner’s report found that fewer than a third of AI use cases succeed in full and meet ROI expectations, while a fifth fail outright. The most common cause was expectations that exceeded what the deployment was designed to do. That is a warning against promising beyond the design. A narrow deployment still requires a defensible success condition, and broader infrastructure does not excuse a mismatch between capability and expectation.
Deloitte’s survey argues for a structured scaling pipeline from pilot to production to multi-site rollout. Replication, however, requires evidence from a validated first use case. The process needs to keep working across different shifts, different products, and changing constraints.
The practical answer to integration debt is a portable process design. A first deployment with the right scope sets out the data requirements, ownership model, hand-off logic, and performance baseline that the next deployment will need. This makes integration work more explicit than a commitment that depends on the vendor to maintain coherence across functions without those foundations.
The financial answer also needs to be explicit. The recommended 2027 budget should reserve room for data preparation, integration, security, training, and ongoing operating costs rather than treating the first deployment as the full expense. Setting aside that money helps with later adoption without requiring an enterprise-wide purchase at the start. The operations leader’s goal is a first deployment that works well enough to earn the right to scope the second.
How to bring a concrete first-process proposal back to the table after the redirect
The follow-up proposal opens with the process walkthrough, showing how the work runs today and where it hurts. It then names the current-state loss, proposes the single KPI the deployment will move, and sets the ninety-day check. This is the constructive answer to the roadmap.
The proposal must also define an AI agent’s authority task by task. Generating a recommendation, opening a work order, and changing a machine setting involve different responsibilities. They cannot be treated as equivalent actions within the same scope. The process owner needs an explicit boundary around what the deployment is being authorized to do.
Early pilots can face messy data, resistance, and failures, so a scoped proposal should not promise frictionless execution. Its value is that the result can be examined and used to inform the next commitment. The closing position is concrete. “The commitment is to this process, in this facility, with this owner, against this KPI, with a result to show in ninety days.”


