Choosing an industrial AI consultant means separating two very different profiles: the consultant who advises and produces strategy documents, and the operator who has personally deployed AI in a production environment and takes accountability for the outcome. For industrial AI, the difference is decisive — plant-floor integration with MES, SCADA, and PLC systems cannot be navigated theoretically. This guide provides selection criteria, questions to ask, red flags to watch for, and a decision checklist that helps you identify the person whose experience matches your specific production challenge rather than the person whose slide deck looks the most polished.
This guide is for executives, operations leaders, and procurement teams who need to select an external AI advisor for a manufacturing operation. It covers selection criteria, evaluation questions, red flags, and the operator-vs-consultant distinction. For direct access to vetted practitioners, see our industrial AI consulting service and our work in the chemical industry.
The criteria below are designed to evaluate whether a candidate has the production experience your engagement requires. Each criterion should be assessed with evidence — a specific deployment, a measured outcome, a named plant — not with assertions. If a candidate cannot provide evidence for a criterion, treat that criterion as unmet rather than assuming competence from seniority or firm reputation.
| Criterion | What to look for | What it tells you |
|---|---|---|
| Production deployment experience | Has personally deployed AI in at least one production environment comparable to yours — same industry, similar process type. | Whether they can anticipate integration challenges or will learn on your budget. |
| Measured outcomes | Can cite specific, quantified results from past deployments (scrap reduction %, downtime hours saved, yield improvement). | Whether they track impact or just deliverables — and whether their work actually moved KPIs. |
| Integration literacy | Demonstrates working knowledge of MES, SCADA, PLC, ERP integration — not just cloud ML platforms. | Whether they understand the plant-floor reality or only the data-science layer. |
| Independence | No undisclosed vendor partnerships or reseller agreements that could bias technology recommendations. | Whether their advice is objective — critical for vendor selection and due diligence. |
| Accountability structure | Willing to define success criteria, milestones, and exit conditions in the engagement contract. | Whether they commit to outcomes or only to effort and deliverables. |
Weight the criteria according to your engagement type. For board advisory, prioritize integration literacy and measured outcomes. For project leadership, prioritize production deployment experience and accountability structure. For due diligence, independence is non-negotiable — the person assessing vendors must have no commercial ties to any candidate being evaluated. For a broader planning perspective, see our guide on building an industrial AI roadmap.
The questions below are designed to be asked directly in an evaluation conversation. The right answers are specific and verifiable. The wrong answers are vague, deflective, or generic. Use the red flags column to identify patterns that indicate the candidate is a presenter rather than an operator.
| Question to ask | Good answer | Red flag |
|---|---|---|
| Which production environments have you personally deployed AI in? | Names a specific plant, industry, and use case with a measurable outcome. | Refers to "projects" or "engagements" without naming environments or outcomes. |
| What integration challenges did you encounter with MES, SCADA, or PLC? | Describes a specific technical obstacle and how it was resolved. | Gives a generic answer about "data silos" or "legacy systems" without specifics. |
| How do you handle model drift in production? | Describes a monitoring and retraining process with thresholds and ownership. | Treats drift as a theoretical concept or says "it depends" without a process. |
| Can you provide a reference from an operations leader? | Provides a reference from a plant manager or operations director. | Only offers references from IT or data-science colleagues, not from operations. |
| What deployment failed, and what did you learn? | Describes a specific failure with a candid root-cause analysis. | Claims no failures, or blames the client without self-reflection. |
| Do you have vendor partnerships that could influence your recommendations? | Discloses all partnerships clearly and explains how conflicts are managed. | Evasive or defensive about vendor relationships, or refuses to disclose. |
The question about failure is particularly revealing. Operators who have deployed AI in production have failures — that is the nature of the work. A candidate who claims no failures has either not deployed enough or is not being honest. The quality of the root-cause analysis matters more than the failure itself: an operator who can explain why a deployment failed and what they changed as a result is more valuable than one who has never faced the situation. For evaluating the financial side of an engagement, see our guide on evaluating ROI of industrial AI projects.
The candidate who gives the most polished pitch is not necessarily the one who can navigate a plant-floor integration. Evaluate evidence of production outcomes, not the quality of the slide deck. A practitioner who is less polished but can name specific deployments is usually the safer choice.
A prestigious firm name does not guarantee that the specific person assigned to your engagement has production experience. Always evaluate the individual who will do the work, not the firm. Ask who will be on-site and verify their credentials independently.
If you are hiring a consultant to assess AI vendors, undisclosed partnerships create a fundamental conflict of interest. The consultant may steer you toward a partner's platform regardless of fit. Require written confirmation of independence and check for reseller agreements.
An engagement without defined success criteria and exit conditions drifts. If the consultant resists committing to measurable outcomes or milestones, that resistance itself is a signal. The right operator defines what success looks like and agrees on when the engagement ends.
A consultant advises, analyzes, and produces recommendations or strategy documents. An operator has personally deployed AI in a production environment and takes direct accountability for the outcome. For industrial AI, the distinction matters because plant-floor integration challenges require someone who has faced them before, not just someone who can describe them theoretically.
It depends on the engagement. Large consultories bring breadth, methodology, and capacity for multi-site programs. Specialists bring depth, specific production experience, and faster time-to-insight for a defined use case. For a first AI deployment, a specialist with relevant plant experience is usually the better choice; for a multi-site rollout program, a larger firm may be more appropriate.
Ask: Which production environments have you personally deployed AI in? What was the measured outcome? What integration challenges did you encounter with MES, SCADA, or ERP? How do you handle model drift? Can you provide a reference from a plant manager? The answers reveal whether the person has operated in production or only advised from a distance.
Red flags include: inability to name specific production environments they have worked in, emphasis on technology names rather than business outcomes, no references from operations leaders, reluctance to discuss failures, and vendor partnerships that create conflicts of interest in vendor selection engagements.
Costs vary widely based on scope, duration, and the seniority of the expert. Board advisory engagements are typically retainer-based; project leadership is usually day-rate or fixed-fee; due diligence is often scoped as a fixed deliverable. The key is not the absolute rate but whether the engagement structure aligns payment with outcomes and includes clear exit criteria.
Your sector, your challenge and the stage of the project. That is enough for us to prepare the proposal.